CANDE-2007

Culvert Analysis and Design

Solution Methods and Formulations

Developed under National Cooperative Highway Research Project NCHRP 15-28

i
ii
 CANDE-2007
Culvert Analysis and Design
Solution Methods and Formulations

Developed under National Cooperative Highway Research Project NCHRP 15-28

Michael G. Katona – Consultant

Gig Harbor, WA

Mark Mlynarski – Michael Baker Jr. Inc.

Moon Township, PA

Timothy J. McGrath – Simpson, Gumpertz & Heger Inc.

Arlington, MA

iii
(This page intentionally left blank)

iv
Table of Contents:

1 SOLUTION LEVELS AND ASSUMPTIONS .....................................................................................................1-1

1.1 Elasticity Solution........................................................................................................................................1-1
1.1.1 Conceptual model...............................................................................................................................1-1
1.1.2 Nonlinear aspects ...............................................................................................................................1-3
1.1.3 Utility of Level 1 ................................................................................................................................1-3
1.2 Finite Element Methodology .......................................................................................................................1-4
1.2.1 Element types .....................................................................................................................................1-5
1.2.2 Global assembly and incremental construction. .................................................................................1-6
1.2.3 Nonlinear solution strategy.................................................................................................................1-7
2 BEAM-COLUMN ELEMENTS – PIPE TYPE MODELS...................................................................................2-1
2.1 General Form ...............................................................................................................................................2-1
2.1.1 Beam kinematics. ...............................................................................................................................2-1
2.1.2 Incremental stress-strain model ..........................................................................................................2-2
2.1.3 Internal thrust and moment increments ..............................................................................................2-3
2.1.4 Beam-column virtual work.................................................................................................................2-5
2.1.5 Finite element interpolation functions. ...............................................................................................2-5
2.1.6 Element stiffness matrices. .................................................................................................................2-6
2.1.7 Transformation to global coordinates.................................................................................................2-7
2.1.8 Equation solving and recovery of structural responses. .....................................................................2-8
2.1.9 Nonlinear solution strategy.................................................................................................................2-9
2.2 Corrugated Metal .......................................................................................................................................2-11
2.2.1 Overview of corrugated metal pipe type ..........................................................................................2-11
2.2.2 Design criteria for corrugated metal .................................................................................................2-11
2.2.3 Nonlinear model for corrugated metal..............................................................................................2-12
2.3 Reinforced Concrete ..................................................................................................................................2-17
2.3.1 Overview of reinforced concrete pipe type ......................................................................................2-17
2.3.2 Design criteria for reinforced concrete .............................................................................................2-17
2.3.3 Nonlinear model for reinforced concrete..........................................................................................2-19
2.4 Thermoplastic Pipe ....................................................................................................................................2-27
2.4.1 Overview of thermoplastic pipe type................................................................................................2-27
2.4.2 Design criteria for thermoplastic pipe ..............................................................................................2-27
2.4.3 Nonlinear model for local buckling in profile plastic pipe ...............................................................2-28
2.5 Basic Pipe Type .........................................................................................................................................2-31
3 SOIL MODELS.....................................................................................................................................................3-1
3.1 Continuum Elements....................................................................................................................................3-1
3.1.1 Triangle elements ...............................................................................................................................3-1
3.1.2 Quadrilateral elements........................................................................................................................3-2
3.1.3 Finite element development ...............................................................................................................3-2
3.2 Isotropic Linear Elastic ................................................................................................................................3-5
3.3 Orthotropic Linear Elastic............................................................................................................................3-6
3.3.1 Orthotropic properties from testing specimens...................................................................................3-6
3.3.2 Orthotropic properties for reinforced soil...........................................................................................3-6
3.4 Overburden Dependent Soil Model .............................................................................................................3-8
3.4.1 Input data for overburden dependent model.......................................................................................3-9
3.4.2 Overburden model development ......................................................................................................3-10
3.4.3 Overburden dependent secant moduli data tables in CANDE..........................................................3-11
3.5 Duncan and Duncan/Selig Soil Models .....................................................................................................3-13
3.5.1 Plane-strain constitutive matrix ........................................................................................................3-13
3.5.2 Duncan Young’s modulus development...........................................................................................3-14
3.5.3 Bulk modulus formulations ..............................................................................................................3-17
3.5.4 Summary of Duncan and Duncan/Selig soil models. .......................................................................3-19
3.5.5 Behavioral characteristics and special considerations ......................................................................3-20
3.5.6 Implementation of soil models and nonlinear solution strategy .......................................................3-20

v
 3.5.7 Recommended Duncan and Duncan/Selig parameters for standard soils.........................................3-24
3.6 Extended Hardin Soil Model......................................................................................................................3-26
3.6.1 Plane-strain constitutive matrix ........................................................................................................3-26
3.6.2 Hardin shear modulus development .................................................................................................3-27
3.6.3 Poisson ratio development................................................................................................................3-30
3.6.4 Summary of extended Hardin soil model functions .........................................................................3-32
3.6.5 Implementation of soil models and nonlinear solution strategy .......................................................3-33
4 INTERFACE ELEMENT .....................................................................................................................................4-1
4.1 Introduction..................................................................................................................................................4-1
4.2 Virtual Work and Constraint Equations.......................................................................................................4-2
4.2.1 Constraint equations general form......................................................................................................4-3
4.2.2 Global virtual work statement with constraints ..................................................................................4-4
4.2.3 General element constraint matrix and load vector ............................................................................4-4
4.3 Interface Element Matrix and Load Vector..................................................................................................4-5
4.3.1 Definition of interface element...........................................................................................................4-5
4.3.2 Incremental and total responses (notation) .........................................................................................4-6
4.3.3 Element constraint matrices and load vectors for three interface states. ............................................4-7
4.4 Nonlinear Solution Strategy for Interface ..................................................................................................4-11
4.4.1 Selecting a new trial interface state ..................................................................................................4-11
4.4.2 Computing load vector parameters for next iteration .......................................................................4-12
4.4.3 Algorithm summary and convergence..............................................................................................4-13
5 LARGE DEFORMATIONS AND BUCKLING ..................................................................................................5-1
5.1 Updated Lagrange Formulation ...................................................................................................................5-1
5.1.1 Coordinates and incremental relationships.........................................................................................5-1
5.1.2 Bernoulli-Euler beam kinematics. ......................................................................................................5-1
5.1.3 Total Lagrangian strain for large rotations. ........................................................................................5-1
5.1.4 Updated Lagrangian strain increments. ..............................................................................................5-2
5.1.5 Incremental stress-strain model. .........................................................................................................5-2
5.1.6 Internal thrust and moment increments. .............................................................................................5-3
5.1.7 Virtual work for beam-column element. ............................................................................................5-4
5.2 Finite Element Development .......................................................................................................................5-6
5.2.1 Finite element interpolation functions. ...............................................................................................5-6
5.2.2 Element matrices and vectors .............................................................................................................5-6
5.2.3 Transformation and global assembly..................................................................................................5-8
5.3 Solution Strategy..........................................................................................................................................5-9
5.3.1 Iterative methodology.........................................................................................................................5-9
5.3.2 Recovery of element forces ................................................................................................................5-9
5.3.3 Update coordinates ...........................................................................................................................5-10
5.4 Buckling Capacity......................................................................................................................................5-11
5.5 Illustration – Simply Supported Beam.......................................................................................................5-12
5.5.1 Development of closed-form solution ..............................................................................................5-12
5.5.2 Example test problem for closed-form solution ...............................................................................5-13
5.5.3 CANDE model of test problem ........................................................................................................5-14
5.5.4 CANDE simulating total Lagrange approach...................................................................................5-15
5.5.5 CANDE Solution for updated Lagrange approach. ..........................................................................5-16
5.6 Illustration -- Soil-Structure Interaction.....................................................................................................5-18
5.6.1 CANDE soil-structure model and parameters ..................................................................................5-18
5.6.2 Study # 1 -- Linear soil and bonded interface..................................................................................5-19
5.6.3 Study # 2 -- Nonlinear soil and bonded interface ............................................................................5-23
5.6.4 Study # 3 -- Linear soil and frictionless interface............................................................................5-27
5.6.5 Comparing Studies #1, #2 and #3 with design criteria .....................................................................5-29
6 DESIGN CRITERIA AND LRFD DESIGN METHODOLGY............................................................................6-1
6.1 Introduction..................................................................................................................................................6-1
6.2 Objective and Scope ....................................................................................................................................6-1
6.3 Service Loads...............................................................................................................................................6-2
6.4 LRFD Loads for Strength-limit States .........................................................................................................6-3

vi
 6.4.1 Load Factors. ......................................................................................................................................6-3
6.4.2 Load Modifiers. ..................................................................................................................................6-4
6.4.3 LRFD load factors and nonlinear soil models. ...................................................................................6-4
6.5 Design Criteria .............................................................................................................................................6-5
6.5.1 Corrugated metal ................................................................................................................................6-5
6.5.2 Reinforced concrete............................................................................................................................6-6
6.5.3 Plastic pipe .........................................................................................................................................6-7
6.6 Illustration of LRFD Factors and Evaluation with CANDE ........................................................................6-9
6.6.1 Construction increment number 1. .....................................................................................................6-9
6.6.2 Construction increment numbers 2 through 8 ....................................................................................6-9
6.6.3 Construction increment number 9 (live load). ..................................................................................6-10
6.6.4 Final evaluation of LRFD design criteria .........................................................................................6-11
7 BANDWIDTH MINIMIZATION.........................................................................................................................7-1
7.1 Background..................................................................................................................................................7-1
7.2 Objective and Scope ....................................................................................................................................7-2
7.3 Bandwidth Minimization Methodology in CANDE ....................................................................................7-3
7.3.1 Element connectivity matrix...............................................................................................................7-3
7.3.2 Algorithm Cycle. ................................................................................................................................7-3
7.3.3 Post algorithm-cycle updates..............................................................................................................7-5
7.4 Transparency of Node Numbers in Solution Output....................................................................................7-7
7.5 Illustration of Bandwidth Minimization.......................................................................................................7-8
8 MODELING TECHNIQUES................................................................................................................................8-1
8.1 Live Loads ...................................................................................................................................................8-1
8.1.1 Live load modeling problem ..............................................................................................................8-2
8.1.2 Elasticity-based correction for live-load strip pressure ......................................................................8-3
8.1.1 AASHTO-based method to compute live-load strip pressure ............................................................8-5
8.2 Pipe Group Connections and Combinations ................................................................................................8-9
8.2.1 Connections among element types .....................................................................................................8-9
8.2.2 Stiffeners and culvert rehabilitation with parallel pipe groups.........................................................8-10
8.2.3 Illustrations of pipe group combinations ..........................................................................................8-12
8.3 Construction Increments ............................................................................................................................8-14
8.3.1 Rules and insights for construction increments ................................................................................8-14
8.3.2 Techniques for initial construction increment ..................................................................................8-15
8.3.3 Soil compaction and construction increments ..................................................................................8-16
9 REFERENCES......................................................................................................................................................9-1
10 CANDE ANALYSIS SOURCE CODE ..............................................................................................................10-1
10.1 Overview of CANDE Analysis Engine Architecture.................................................................................10-1
10.2 Executive Routine (Cande_dll) ..................................................................................................................10-1
10.3 Pipe-type Subroutines ................................................................................................................................10-2
10.4 Solution-level Subroutines.........................................................................................................................10-4
10.4.1 Elasticity solution Level 1 ................................................................................................................10-4
10.4.2 Finite element solutions Level 2 and Level 3...................................................................................10-4
10.5 Extensions to CANDE-2007......................................................................................................................10-7
10.5.1 New pipe-type model .......................................................................................................................10-7
10.5.2 New soil models ...............................................................................................................................10-7
10.5.3 New canned mesh for Level 2 ..........................................................................................................10-8
11 CANDE GUI Source Code..................................................................................................................................11-1
11.1 Overview of CANDE GUI Architecture....................................................................................................11-1
11.2 CANDE Files .............................................................................................................................................11-3
11.2.1 CANDE Input Definition .................................................................................................................11-3
11.2.2 CANDE analysis XML output files..................................................................................................11-6
11.2.3 CANDE table of contents files .........................................................................................................11-7
11.2.4 CANDE help files ............................................................................................................................11-9

vii
(This page intentionally left blank)

viii
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

1 SOLUTION LEVELS AND ASSUMPTIONS

This document focuses on the engineering mechanics and methods that lie behind the capabilities in CANDE-2007.
Each chapter is essentially a stand-alone reference that describes the theory and engineering approximations used for
the solution methods and nonlinear models employed in the program. The reader is referred to CANDE-2007 User
Manual and Guideline for an overall understanding of CANDE capabilities and architecture from a user’s perspective.

Two distinct solution methods are contained in CANDE. The first is called Level 1and is an extension of a closed-form
elasticity solution by Burns and Richard (Reference 7). The second is a finite element methodology modified and
extended from Herrmann (Reference 8). Input for the finite element method has two input options called Level 2 and
Level 3. Level 2 offers a completely automated mesh generation scheme but it is restricted to basic culvert shapes and
symmetric installations, whereas Level 3 is virtually unrestricted in modeling capability, but requires the user to define
the mesh topology.

Fundamental to both the closed form solution and the finite element methodology is the assumption of plane strain
geometry, two-dimensional loading, and real-time independence. Naturally, the elasticity solution is more restrictive
than the finite element solutions for Levels 2 and 3. Detailed capabilities and restrictions of the solution methods are
discussed in the following sections.

1.1 Elasticity Solution

The elasticity formulation provides an exact solution for an elastic cylindrical conduit encased in an isotropic,
homogeneous, infinite, elastic medium (soil) with a uniformly distributed pressure acting on horizontal planes at an
infinite distance. Thin-shell theory is assumed for the conduit, and continuum elastic theory is employed for the
surrounding infinite medium. The conduit-medium interface is modeled with a choice of two boundary conditions:
bonded interface, where both normal and tangential forces are transmitted across the interface, and frictionless
interface, where only normal forces are transmitted across the interface. Table 1.1-1 identifies the parameters that
describe the idealized boundary value problem and summarizes the elasticity solutions of key structural responses for
the two interface assumptions. Key structural responses, including radial and tangential soil pressure on conduit, radial
and tangential displacements of conduit wall, along with moment, thrust and shear resultants are given as a function of
the angle theta measured counterclockwise from the springline. The solutions in Table 1.1-1 are expressed in terms of
the dimensionless parameters alpha and beta. Alpha is a measure of the conduit’s hoop stiffness relative to the soil’s
ability to resist uniform compression, and beta is a measure of the conduits bending stiffness relative to the soil’s ability
to resist ovaling deformation. The expressions in Table 1.1-1 are developed in References 9 and 10.

1.1.1 Conceptual model

At first encounter, the applicability of the infinite regions described above to model culvert systems with finite burial
depths may seem questionable. However, it has been shown that the interaction between conduit and medium (or pipe
and soil) occurs primarily within a three-radius area of the pipe center. Beyond this area the soil response is practically
unaffected by the pipe inclusion for overburden loading. Therefore, the pipe-soil system can be visualized with the
finite boundaries and overburden loading as shown in Figure 7.1.1. In this representation, Po is the equivalent
overburden pressure of the fill soil above the pipe given by Po = γH, where the parameters are identified in the figure.
The elasticity solution becomes progressively less valid when H, the depth of cover, is less than 3R and should not be
used for cover depths less than 2R.

1-1
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

Table 1.1.1-1 Elasticity parameters and solutions for bonded and frictionless interfaces.
Elasticity Solution Parameters
Soil Properties Pipe Properties Dimensionless Parameters
• P0 = Overburden pressure • R = Average radius • θ = Angle in polar coord.
• A = Wall thrust area
• G = Shear modulus per unit length • α = EA/(2GR)
• I = Wall moment of inertia (Relative hoop stiffness)
• K = Lateral pressure coeff. per unit length
K = μ/(1- μ), where μ • E = Young’s modulus in • β = EI/(2GR3)
is Poisson ratio of soil plane strain form (Relative bend stiffness)
2
E = Epipe/(1- μpipe )

Bonded Interface: Structural Response Solutions

Structural Response
Bonded Interface – Expression to be multiplied by common factor
R
θ
Common Denominator term: D = (1+K) + 3(5-K)β + (3+K)α + 12(3-K)αβ
Factor

Radial pressure on pipe Po α/(1+ α) – {(1-K)(-2α + 18β + 24αβ)/D}cos2θ

Tangential pressure on Po 0 + {(1-K)(4α + 24αβ)/D}sin2θ

pipe
Radial displacement of Po R/(2G) 1/(1+ α) – {(1-K)(2 + 4α)/D}cos2θ
pipe
Tangential displacement Po R/(2G) 0 + {(1-K)(2 + 2α + 6β)/D}sin2θ
of pipe
Moment in pipe wall per Po R2 β /(1+ α) + {(1-K)(6β + 12αβ)/D}cos2θ
unit length
Thrust force in pipe wall Po R α/(1+ α) + {(1-K)(2α + 6β + 24αβ)/D}cos2θ
Per unit length
Shear force in pipe wall Po R 0 – {(1-K)(12β + 24αβ)/D}sin2θ
per unit length
Frictionless Interface: Structural Response Solutions
Structural Response
Frictionless Interface -- Expression to be multiplied by common factor
R
θ
Common Denominator term: D = (1+K) + 3(5-K)β
Factor

Radial pressure on pipe Po α/(1+ α) – {(1-K)(18β )/D}cos2θ

Tangential pressure on Po 0.0

pipe
Radial displacement of Po R/(2G) 1/(1+ α) – {(1-K)2/D}cos2θ
pipe
Tangential displacement Po R/(2G) 0 + {(1-K)/D}sin2θ
of pipe
Moment in pipe wall per Po R2 β /(1+ α) + {(1-K)(6β)/D}cos2θ
unit length
Thrust force in pipe wall Po R α/(1+ α) + {(1-K)(6β)/D}cos2θ
Per unit length
Shear force in pipe wall Po R 0 – {(1-K)(12β)/D}sin2θ
per unit length

1-2
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

Figure 1.1.1-1 Conceptual view of elasticity solution with finite boundaries.

1.1.2 Nonlinear aspects

Although Level 1 is based on a linear elasticity solution, a fair degree on nonlinear modeling is achieved for both soil
and pipe in the following manner. First the overburden pressure, Po, may be divided into “n” load increments, ΔPi, for i
= 1 to n, and each load increment is applied in a series of load steps. During each load step the material properties of the
soil may be redefined in accordance with current overburden pressure. The structural responses as presented in Table
1.1-1, with Po replaced by ΔPi , are summed in a running total thereby providing a load-deformation history record. The
concept of overburden-dependent soil properties is elaborated in a later chapter on soil models.

With regard to nonlinear behavior of the pipe, each pipe type model (discussed later) can create changes in the effective
bending stiffness EI and the effective hoop stiffness EA at each point around the pipe periphery. These modified
properties are used directly to predict stress and strain at each pipe point based on the moment, thrust and shear at that
point. However, an average value of the modified properties is used in the closed-form equations to predict the primary
unknowns (displacements, moment, thrust, and shear) at each point on the pipe periphery.

1.1.3 Utility of Level 1

The Level 1 approach does not have the versati1ity and generality of the Level 2 and 3 counterparts. Nonethe1ess its
efficiency and app1icabi1ity particularly for design is quite remarkable. From a design viewpoint, the exact nature of

1-3
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

the soil system, loading and boundary conditions may not be known with enough certainty to warrant a finite element
solution. Thus the simplifying assumptions of Level 1 are often commensurate with knowledge of the design problem.
It follows that the simple data preparation and quick computer time make Level 1 an attractive and powerful design
tool for routine applications for deep burial loading.

1.2 Finite Element Methodology

As previously mentioned Level 2 and Level 3 share a common finite element solution program and differ only with
respect to the mode of input: automatic or user-defined. Although the development and formulation of the finite
element method is well established in the literature, an overview will be given herein as it applies to culvert
installations.

A static, displacement-based finite element formulation is developed based on incremental virtual work. Incremental
virtual work is ideally suited for characterizing buried structures problems because the load may be applied in a series
of steps representing increments of overburden pressure, temporary construction loading and live loads from vehicles.
In the case of culvert-soil systems, the incremental approach takes on a larger meaning than just incremental load steps.
To wit, not only the load, but also the structural system may be assembled in increments. This process is termed the
"incremental construction" technique and is the mathematical analogue of the physical process of constructing the soil
system in a series of compacted layers or lifts. The structural predictions from an incremented system are more realistic
than an equivalent monolith system. Another advantage of the incremental virtual work formulation is the relative ease
of incorporating and solving various nonlinear models, which are discussed throughout this document.

For a general structural-continuum system, incremental virtual work may be expressed in matrix notation as:

∫ δε Δσ dV = ∫ δu T Δτ dS + ∫ δu T Δf dV
T
Equation 1-1
V S V

Where, σ = stress vector

ε = strain vector
u = displacement vector
τ = surface-traction vector
f = body-force vector
δx = small virtual variation of any vector x, δ is an operator
Δx = x i+1 - x i , incremental change in any vector x from load step i to i+1, Δ is an operator
S = surface area of traction loads at load step i + 1
V = volume of structural system at load step i + 1

The physics behind the above incremental virtual work statement is as follows. We assume that the structural system is
in equilibrium at load step i, and we are in the process of applying incremental loads to reach load step i + 1. With this
understanding, Equation 1-1 states, "The increment of internal virtual-strain-energy is equal to the increment of
external virtual work of body and traction loads as the system undergoes a virtual movement compatible with the
kinematical constraints of the system."

To express the virtual work statement in a displacement formulation, the stress vector is expressed in terms of strains
by using an incremental constitutive relationship, symbolized as ;

Δσ = CΔε Equation 1-2

Where, C is the constitutive function (matrix) that relates stress and strain increments from load step i to load step i +
1. In general, the coefficients may be nonlinear and dependent on the total stress and strain history.

1-4
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

Next, the strain increments are expressed in terms of displacement increments as;

Δε = ΔQ(u) = Q(Δu) Equation 1-3

where, Q is an operator matrix composed of partial derivatives. For small-deformation theory, the operator matrix
contains linear operations so that Q is commutative with Δ and δ operators. In a later section of this chapter, the
strain-displacement operator is extended to include large deformations, however for present purposes small-
deformation theory is assumed.

Using the above relationships, the incremental virtual work may be expressed in terms of displacement fields over the
entire domain as;

∫ Q(δu) C Q(Δu) dV = ∫ δu Δτ dS + ∫ δu T Δf dV
T T
Equation 1-4
V S V

The components of C and Q are dependent on element type, material behavior, and kinematical assumptions. Later, a
nonlinear form of the Q operator is used to formulate large deformation analysis. The specific forms will be developed
in subsequent chapters, for now, the concern is with the general formulation.

At this juncture, the finite element approximation is introduced by subdividing the domain V into a discrete set of
elements interconnected at common nodal points. The unknown displacement fields within each element are
approximated with prescribed functions such that continuity is maintained at the nodes and along the boundaries
between elements. The assemblage of elements and nodes is termed the finite element mesh or topology.

The unknown displacement fields within each element are approximated by specified interpolation functions with
unknown nodal-point displacements values, symbolically expressed as;

u e = Nuˆ e Equation 1-5

where , u e = displacement vector-field within element

N = matrix of prescribed interpolation functions (spatial variables)
û e = nodal-point displacement vector (unknown degrees of freedom)

The subscript “e” implies the above relationship holds within a given element. The form of the interpolation matrix and
nodal displacement vector is dependent on element type.

1.2.1 Element types

The heart of any finite element formulation is the description of the elements themselves. There are three basic element
types employed in the CANDE program.
• Quadrilateral and triangular elements: for in-situ soil, bedding material, fill soil, pavement, etc.
• Interface element: for interfaces such as between pipe and soil.
• Beam-column element: for culvert structure like pipes, boxes, arches, etc.

The quadrilateral element is a nonconforming element developed by Herrmann (Reference 11) that has superior
qualities in all basic deformation modes. The quadrilateral is composed of two triangles with complete quadratic
interpolation functions initially specified within each triangle. Upon applying appropriate constraints and static
condensation procedures a four-node quadrilateral with an 8 x 8 stiffness matrix is formed, wherein each node has two
degrees of freedom representing horizontal and vertical displacements. Associated with the quadrilateral/triangle
element are four choices for material characterization: (1) linear elastic, isotropic or anisotropic; (2) incremental elastic,
dependent on overburden pressure; (3) hyperbolic soil models by Duncan and Duncan & Selig, and (4) variable

1-5
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

modulus soil model by Hardin. Specific derivations of the quadrilateral/triangular element are given in Section 7.3 in
conjunction with the constitutive models for soil.

The interface element allows consideration of two subassemblies meeting at a common interface such that under
loading the subassemblies may slip relative to each other with Coulomb friction, or separate, or re-bond. A natural
application of this element is simulating the pipe-soil interface, other applications include trench soil-to-in-situ soil
interface. The interface element is composed of two nodes, each associated with one subassembly and initially meeting
at a common contact point. Each contact node has two degrees of freedom, horizontal and vertical displacement. In
addition, a third node is assigned to the "interior" of the contact point to represent normal and tangential interface
forces. The three nodes produce a 6 x 6 element "stiffness" matrix in a mixed formulation. Actually, the element
stiffness is a set of constraint equations with Lagrange multipliers. Constraint equations impose conditions on normal
and tangential displacements, and Lagrange multipliers are interface forces. The interface element derivation is
presented in Chapter 4

Lastly, the beam-column element is the familiar structural-matrix element for two-dimensional bending and axial
deformation. It is defined by two nodes with three degrees of freedom per node, horizontal and vertical displacement
and a rotation. Bending deformation is approximated by a cubic interpolation function and axial deformation is
represented by a linear interpolation function. The beam-column element derivation employs a general nonlinear stress-
strain model that is specialized to the material behavior of different pipe types. Specifically, elastic-plastic behavior for
corrugated metal, tensile cracking and compressing yielding for reinforced concrete, and local buckling of wall profile
elements for thermoplastic pipes. The beam-column element derivation is presented in Chapter 2 for each pipe type.

With the above background on specific element types, we continue with the general finite element formulation, wherein
the global integrations expressed in Equation 1-4 may now be obtained as a summation of all element integrations.
Specifically, within each element we replace the global displacement vector Δu by NΔuˆ e and δu by N δ uˆe . Since
Δuˆ e and δuˆ e are nodal-point parameters, they may be factored out of the element integrations. Thus, the element
integrations are performed over known interpolation functions resulting in the so-called element stiffness matrix and
load vector, expressed as;

∫ Q(N) C Q(N) dV
T
ke = e Equation 1-6
Ve

∫ N Δτ dS ∫N
T T
Δpe = e + Δf dVe Equation 1-7
Se Ve

Where, k e = element stiffness matrix

Ve = volume of element
Δpe = element load vector
Se = surface of element where traction is applied

1.2.2 Global assembly and incremental construction.

Each element is assigned a construction increment number, which corresponds to the load step number that the element
stiffness matrix and load vector is assembled into the global system. Once an element stiffness enters the system it
remains active for all subsequent load steps (there is no element death option). Of course, the element body-load vector
is only applied during the load step corresponding to the element construction increment number; it is not reapplied on
subsequent load steps. Similarly, surface pressure and/or point loads (i.e., force boundary conditions) are also assigned
a particular construction increment number and are only applied during the corresponding load step. Likewise,
displacement boundary conditions are applied during the load step they are specified and remain fixed for all
subsequent load steps.

1-6
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

To facilitate the global assemblage of elements, the current list of all active nodal degrees of freedom are aligned in a
particular sequence order corresponding to the numerical sequence of the number-tag assigned to each node. Each
active node is assigned two sequential positions in the global list for the horizontal and vertical nodal degrees of
freedom. For those nodes that have a beam-column element attached, a third sequential position is assigned to the
global list for the rotational degree of freedom. The global displacement vector, û G , represents the ordered list of all
active degrees of freedom.

With the above understanding, the finite-element equivalent of Equation 1-4 is obtained by summing all active element
contributions and assembling them into the global incremental virtual work statement. Since δuˆ G is an arbitrary virtual
movement of all active degrees of freedom, the virtual work statement requires that the following set equilibrium
equations be satisfied for each load step.

K G Δuˆ G = ΔPG Equation 1-8

where, KG = ∑
elements
k e = incremental global stiffness matrix

ΔPG = ∑
elements
Δpe = incremental global load vector

Δuˆ G = increment of global displacement vector (unknown degrees of freedom)

If the global system is linear, (that is, linear models are selected for the pipe materials, soil zones, interface conditions,
and deformation theory), then the global stiffness matrix is directly calculable and constant for each load step. With this
simplifying assumption, Equation 1-8 represents a set of linear algebraic equations that may be solved by standard
methods such as Gauss Elimination to obtain Δuˆ G . The running summation of each Δuˆ G over all load steps provides
the total solution for the global displacement vector. Unfortunately, most culvert problems exhibit some type of
nonlinear behavior so that the global stiffness matrix is not constant during the load step, requiring a nonlinear solution
strategy discussed next.

1.2.3 Nonlinear solution strategy

CANDE employs a solution strategy known as the direct iterative method, or more simply called trial and error. This
method has proven to be robust and readily accommodates the wide variety of nonlinear models such as tensile
cracking and elastic-plastic behavior of pipe models, hyperbolic constitutive laws for soil models, frictional sliding and
separation for interface models, and geometric nonlinearity for large deformation analysis. Most importantly, the
solution accuracy is not dependent on the magnitude of load increments; that is, a large load increment will produce
essentially the same results as the case where the load increment is divided into two or more sub-increments.

To illustrate the strategy, we start with the assumption that a valid (converged) solution is in hand for load step i, and
we seek to increment the solution from load step i to i+1. That is, we know the mechanical responses (displacements,
stresses, strains, etc.) at load step i and we seek to update the mechanical responses at load step i+1. The global set of
equations for load step i+1 is expressed with an iteration counter k as;

K Gk Δuˆ Gk = ΔPG k Equation 1-9

where, K G k = global stiffness matrix for iteration k, (assembly of active elements)

Δuˆ Gk = increment of global displacement vector for iteration k (active dof)

ΔPG k = incremental global load vector for iteration k

1-7
Chapter 1 – Solution Levels and Assumptions CANDE-2007 Solution Methods and Formulations

k = iteration counter; 1, 2, 3, …

For the first iteration k =1, the global stiffness matrix is assumed to remain the same as computed at the end of the
previous iteration cycle for the step i, except for the addition of new elements entering the system for the first time at
load step i+1. Note, if load step i+1 happens to be the first step, meaning the initial configuration, then the global
stiffness matrix is constructed based on unloaded, linear-elastic elements belonging to the initial configuration prior to
loading.

With the above understanding, the set of linear-like equations (Equation 1-9 for k = 1) is solved by a standard Gauss
elimination scheme for the first trial solution Δuˆ G1 . The trial solution is temporarily added to the known mechanical
responses at load step i to form a new and better estimate of the displacements, stresses and strains at load step i+1.
Next, all the nonlinear models for each element are re-evaluated based on current estimate of mechanical response for
load step i+1, typically requiring small modifications to the elements stiffness matrices.

The process is repeated for k = 2, 3, 4 … until convergence is witnessed. That is, iteration k produces a trial
solution Δuˆ Gk that is based on the revised stiffness matrix from the previous iteration Δuˆ G k-1 . When two consecutive
iterations produce the same stiffness matrices for all elements within small error limits, then the solution has converged
and we proceed to the next load step. The mechanics behind updating each element stiffness during the iteration cycle
depends on the particular nonlinear model, discussed in subsequent chapters.

Once a converged solution increment has been found, all the mechanical responses are updated based on the last
iteration solution, all intermediate iteration solutions are discarded. Thus prior to starting the next load step, the
mechanical responses are permanently updated as symbolically shown below

q i+1 = q i + Δq Equation 1-10

where q stands for all structural responses such as displacements, stresses, strains, moments, thrusts, etc. The data is
saved in the program and the output file thereby providing a load-response history record.

As a final comment, it must be recognized that convergence is never guaranteed to occur. Sometimes convergence does
not and should not occur because the system or portion of the system is physically incapable of carrying additional
loading (singular system). In the normal default mode, CANDE will terminate on non-convergence with a message
describing what nonlinear models are not converging. However, CANDE also offers the user control over the number
of iterations with a special command to continue processing load steps even after a non-convergence load step has been
encountered. In this way the user may inspect the post-convergent results to ascertain the cause of the problem.

1-8
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

2 BEAM-COLUMN ELEMENTS – PIPE TYPE MODELS

Presented in this section is a complete development of the beam-column elements used for modeling corrugated metal,
reinforced concrete and thermoplastic pipe materials (or any two-dimensional structure). The initial development is
focused on the general finite-element formulation that is applicable to all pipe-type material models. Subsequent
developments describe specific stress-strain models that distinguish one pipe-type material from another.

2.1 General Form

The major assumptions and limitations used for the beam-column element are listed below:
1. Two-dimensional framework in a plane strain formulation.
2. Bernoulli-Euler beam kinematics without shear deformation.
3. Small deformation theory (this restriction removed in a subsequent chapter).
4. Material nonlinearity is a function of normal stress and strain and their history.
5. Incremental virtual work formulation with incremental stress-strain relationships.

2.1.1 Beam kinematics.

Based on the assumption that cross-sectional planes remain plane in bending and axial deformation, the Bernoulli-Euler
assumption for displacement increments Δu at any station x along the beam length and at any point y in the beam’s
cross section may be expressed as an increment from load step i to i+1 as,

Δu ( x, y ) = Δa ( x ) + ( y*-y ) Δv′ Equation 2-1

where, Δu ( x,y ) = displacement in x direction from column and bending deformation.

Δa ( x ) = uniform displacement in x direction from column action, independent of y.
Δv ( x ) = displacement in y direction as function of x, independent of y.
Δv′ = d ( Δv ) /dx , derivative of Δv with respect to x (local deformation slope).
y* = reference plane in the beam cross-section at y = y*, yet to be specified.
Δq = q i+1 - q i = incremental change in any function q from load step i to i+1.

The Bernoulli-Euler assumption states that Δu at any point in the cross-section may be described by a uniform
displacement Δa plus a rotational-like motion due to the slope of the transverse displacement increment. Note that the
axis, y*, is not specifically fixed in cross-section under these assumptions. Figure 2.1.1-1 illustrates the beam-column
element in local beam coordinates, wherein x is aligned with the longitudinal axis and y is in the transverse direction
locating positions in the beam’s cross section.

Figure 2.1.1-1 Local coordinates for beam element and cross section.

2-1
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Applying small strain theory to the above displacement functions produces one non-zero strain component, which is the
strain normal to the beam cross section, Δε=d(Δu)/dx . In terms of the Bernoulli-Euler functions, the incremental
normal strain function is given by:

Δε ( x,y ) = Δa ′ + ( y*-y ) Δv′′ Equation 2-2

where ( )′ = d ( ) / dx , prime symbol denotes derivative of any quantity with respect to x.

2.1.2 Incremental stress-strain model

The nonlinear stress-strain laws used in CANDE are specific to each pipe material such as corrugated metal, reinforced
concrete, and profile plastic pipe, which are presented in subsequent sections. However, all pipe materials stress-strain
models conform to the same generic form expressed as;

Δσ = E c Δε Equation 2-3

where, Δσ = increment of axial stress from load step i to i + 1, (Δσ = σi+1 - σi )

Δε = increment of axial strain from load step i to i + 1, (Δε = εi+1 - εi )
E c = chord modulus of total stress-strain curve from load step i to i + 1

The chord modulus is dependent on the type of material and is generally dependent on the history of stress and strain
throughout all loading steps. It is determined iteratively during each load step by repeating the solution process until the
value of Ec converges for each point within the element to a small tolerance of error. The chord modulus is illustrated in
Figure 2.1.2-1 for a generic, nonlinear stress-strain relationship.

Replacing Δε in Equation 2-3 with Equation 2-2, the fundamental stress-displacement relationship may be expressed
as,

Δσ = E c (Δa ′ + ( y*-y ) Δv′′) Equation 2-4

Next, we will use the above equation to define the internal thrust and internal moment acting on the cross section at any
station x.

2-2
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Figure 2.1.2-1. Generic incremental stress-strain relationship.

2.1.3 Internal thrust and moment increments

As is customary, we define the thrust increment as the integral of Δσ over the cross section area A, and the moment
increment as the integral of Δσ (y*-y) over the cross section area A. Specifically,

ΔN = ∫ Δσ dA
A
= ∫ E (Δa′ + ( y*-y ) Δv′′ ) dA
A
c Equation 2-5

ΔM = ∫ Δσ ( y*-y ) dA = ∫ E (Δa′ + ( y*-y ) Δv′′ ) ( y*-y ) dA

c Equation 2-6
A A

where, ΔN = thrust increment from load step i to i + 1, (ΔN = ΔN (x))

ΔM = moment increment from load step i to i + 1, (ΔM = ΔM (x))
dA = b(y)dy, a differential area of cross section where b(y) is the width
Figure 2.1.3-1 shows the incremental thrust and moment resultants for an arbitrary cross-section with the coordinate y
measured from bottom.

2-3
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Figure 2.1.3-1 Thrust and moment increments from integration of stress distribution

The location of the arbitrary reference position y* is now conveniently selected so that the first moment of integration
is zero, i.e.,
∫A Ec (y*-y) dA = 0. Therefore, the location of y* measured from the bottom of the cross section is,

y* = ( ∫ E c y dA ) / ( ∫ E c dA) Equation 2-7

A A

With the above definitions for y*, the thrust and moment increments simplify to the following equations,

ΔN = EA*Δa ′ Equation 2-8

ΔM = EI*Δv′′ Equation 2-9

where, EA* = ∫E
A
c dA = effective axial stiffness of beam element

∫ E ( y*-y )
2
EI* = c dA = effective bending stiffness of beam element .
A

In the above equations, the thrust expression is written in the familiar linear form as the product of axial stiffness and
column strain; likewise, the moment expression is written in the familiar form as the product of bending stiffness and
beam curvature. Although the equations for thrust and moment increments may appear linear-like, the integrals
defining y*, EA*, and EI* depend on the nonlinear chord modulus, which in turn depends on the stress-strain state at
each point in the cross-section in advancing from load step i to i+1.

From an analytical viewpoint, the only difference between one nonlinear pipe material and another is calculation of y*,
EA*, and EI* along with the associated convergence criteria. Except for these calculations, the fundamental beam-
column formulation is identical for all pipes as presented in the remainder of this section.

2-4
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

2.1.4 Beam-column virtual work.

The general framework of incremental virtual work as previously presented is now specialized for the beam-column
element. Specifically, the increment of internal virtual-strain-energy is written as,

δΔU e = ∫ (δa′ + ( y*-y ) δv′′)Δσ dV

V
Equation 2-10

where the virtual strain is expressed in terms of the Bernoulli-Euler functions prescribed in Equation 2-2, and V
represents the volume of the beam-column element.

Separating the volume integral into area and length integrals, dV = dAdx, we arrive at,

δΔU e = ∫ {δa′∫ ΔσdA + δv′′∫ Δσ ( y*-y ) dA} dx

x A A
Equation 2-11

Since the two area integrals in the above expression have been identified as ΔN and ΔM, the above expression may be
equivalently written as,

δΔU e = ∫ {δa′ΔN
x
+ δv′′ΔM } dx Equation 2-12

Lastly, replacing ΔN and ΔM by Equations 2-8 and 2-9, we arrive at the desired displacement form for internal virtual
strain energy of the beam-column element,

δΔU e = ∫ {δa′ EA* Δa′ +

x
δv′′ EI*Δv′′}dx Equation 2-13

In a similar manner, the incremental external virtual work for body loads is specialized for the beam-column element
as,
δWe = ∫ (δaΔFx + δvΔFy )dx Equation 2-14
x

where, ΔFx = ∫ f x dA = axial body force per unit beam length in local x-direction,
A

ΔFy = ∫ f y dA = transverse body force per unit beam length in local y-direction.
A
The body force per unit volume is assumed to be generated by gravity acting in the global vertical direction so that fx
and fy are the component body forces per unit volume in local beam coordinates. Recall that all surface tractions and
pressures are incorporated in the system at the global level, not at the element level.

2.1.5 Finite element interpolation functions.

To utilize the internal and external virtual work expressions in a finite element formulation for the beam-column,
elements, we introduce specific interpolation functions for the displacement functions that become exact as the element
lengths become small. The interpolation functions are expressed in terms of unknown nodal variables shown in the
sketch below.

y θa ua ub θb
va vb

x ( )

2-5
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

where, Δu a , Δu b = incremental nodal displacements in x direction at nodes a and b

Δva , Δv b = incremental nodal displacements in y direction at nodes a and b
Δθ a , Δθ b = incremental nodal rotations in counter-clockwise direction at nodes

The incremental axial displacement function is approximated with a linear interpolation function given by,

Δa(x) = φ1 (x) Δu a + φ 2 (x) Δu b Equation 2-15a

Or in vector notation,

Δa(x) = <φ1 φ 2 > <Δu a Δu b > T Equation 2-15b

where, φ1 (x) = 1 - x/L = first interpolation function defined over beam length L.
φ 2 (x) = x/L = second interpolation function defined over beam length L.

The incremental vertical displacement function is approximated by a cubic polynomial, known as a Hermetian
interpolation function, and is expressed in vector notation by,

Δv(x) = < γ1 γ 2 γ 3 γ 4 > < Δva Δθ a Δv b Δθ b > T Equation 2-16

where, γ1 (x) = 1 - 3 (x/L)2 + 2 (x/L)3 = Hermetian interpolation function one,

2
γ 2 (x) = L (1 - x/L) x/L = Hermetian interpolation function two,
γ 3 (x) = 3 (x/L)2 - 2 (x/L)3 = Hermetian interpolation function three,
γ 4 (x) = L (1 - x/L) (x/L)2 = Hermetian interpolation function four.

The above interpolation functions for axial deformation and bending lead to the element stiffness matrices developed
next.

2.1.6 Element stiffness matrices.

Taking the necessary derivatives of the interpolation functions for Δa(x) and Δv(x) and inserting them into the two
internal virtual work terms of Equation 2-13, we arrive at the following expressions for internal virtual work in terms
of axial virtual work and bending virtual work.

δΔU e = δ<Δu a Δu b > K a < Δu a Δu b > T + δ< Δva Δθ a Δv b Δθ b >K b < Δva Δθ a Δv b Δθ b > T

L
where, K a = (EA*) ∫ [<φ1′ φ′2 > T < φ1′ φ′2 >]dx (axial stiffness) Equation 2-17
0

L
and, K b = (EI*) ∫ [<γ1′′ γ′′2 γ′′3 γ′′4 > T <γ1′′ γ′′2 γ′′3 γ′′>]dx (bending stiffness) Equation 2-18
0

Performing the integrations with respect to x over the element length L, we arrive at the final evaluation for the axial
stiffness and bending stiffness as recorded below.

2-6
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

⎛ 1 -1 ⎞
K a = (EA*/L) ⎜ ⎟ Equation 2-19
⎝ -1 1⎠

⎛ 12/L2 6/L -12/L2 6/L ⎞

⎜ ⎟
6/L 4 -6/L 2
K b = (EI*/L) ⎜ ⎟ Equation 2-20
⎜ -12/L2 -6/L 12/L2 -6/L ⎟
⎜⎜ ⎟⎟
⎝ 6/L 2 -6/L 4 ⎠

For linear materials EA* and EI* are constant and remain the same through out the analysis. For nonlinear materials the
stiffness factors EA* and EI* are determined iteratively for each load step wherein the factors are computed from the
average moment and thrust at the center of the element. This implies that the elements will be sufficiently small so that
thrust and moment do not vary substantially over any one element.

The above two matrices are combined into a single 6 x 6 matrix by re-grouping the nodal unknowns into a single vector
as expressed below

Δuˆ e = < Δu a Δva Δθ a Δu b Δv b Δθ b > T Equation 2-21

where, Δuˆ e is a vector of the six nodal unknowns in element coordinates. Accordingly, the 6x 6 element stiffness
matrix is formed from the adding Ka and Kb to get the complete element stiffness matrix, Ke.

⎛1 0 0 -1 0 0⎞ ⎛ 0 0 0 0 0 0 ⎞
⎜0 0 0 0 0 0
⎟ ⎜ 2 2 ⎟
⎜ ⎟ ⎜ 0 12/L 6/L 0 -12/L 6/L ⎟
⎜0 0 0 0 0 0⎟ ⎜ 0 6/L 4 0 -6/L 2 ⎟
K e =EA* ⎜ ⎟ + EI* ⎜ ⎟
⎜ -1 0 0 1 0 0⎟ ⎜ 0 0 0 0 0 0 ⎟
⎜0 0 0 0 0 0⎟ ⎜ 0 -12/L
2
-6/L 0 12/L
2
-6/L ⎟
⎜ ⎟ ⎜⎜ ⎟⎟
⎝0 0 0 0 0 0⎠ ⎝ 0 6/L 2 0 -6/L 4 ⎠

2.1.7 Transformation to global coordinates.

The last step before adding the element’s contribution into the entire system is to transform the nodal variables from
local coordinates to global coordinates.

Let β = angle from global X-axis to local x-axis, then the local nodal variables may be expressed as global nodal
variables by,

Δuˆ e = T Δuˆ E Equation 2-22

where, Δuˆ e = < Δu a Δva Δθ a Δu b Δv b Δθ b > T = local nodal variables for element

Δuˆ E = < Δu A Δv A Δθ A Δu B Δv B Δθ B > T = global nodal variables for element

2-7
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

⎛ cosβ sinβ 0 0 0 0 ⎞
⎜ -sinβ cosβ 0 0 0 0
⎟
⎜ ⎟
⎜ 0 0 1 0 0 0 ⎟
T= ⎜ ⎟ = transformation matrix
⎜ 0 0 0 cosβ sinβ 0 ⎟
⎜ 0 0 0 -sinβ cosβ 0 ⎟
⎜ ⎟
⎝ 0 0 0 0 0 1 ⎠

With the above transformation matrix, the global element stiffness matrix may be expressed in global coordinates as,

T
KE = T Ke T = global element stiffness matrix Equation 2-23

It should be clear that lower-case subscripts are used for element quantities that are in expressed in local coordinates,
whereas upper-case subscripts are used for element quantities that are in expressed in global coordinates.

The element load vector due to body weight may be directly expressed in global coordinates because gravity operates
in the global Y direction (negative). Thus the element’s weight is divided equally to both nodes and acting in the -Y
direction associated with Δv A and Δv B as expressed below.

⎛ 0 ⎞
⎜ 1/2 ⎟
⎜ ⎟
⎜ 0 ⎟
Δp E = -ρAL ⎜ ⎟ = global element load vector Equation 2-24
⎜ 0 ⎟
⎜ 1/2 ⎟
⎜ ⎟
⎝ 0 ⎠

where, ρ is the density of the beam-column material, A is the cross-sectional area and L is the length. The element
stiffness matrix KE and load vector ΔpE are in the proper form for global assembly.

2.1.8 Equation solving and recovery of structural responses.

After the entire set of elements is assembled and solved by Gauss elimination as described in Sections 1.2.2 and 1.2.3,
the solution (or trial solution) gives numerical results for the incremental displacements at all nodes. Using these
results, other key structural responses of the beam-column element are calculated (recovered) based the previously
developed relationships. Key responses include internal thrust, moment and shear resultants and stress and strain
distributions. .

First, the internal force and moment increments at the ends of the beam-column element are recovered by multiplying
the element stiffness matrix by the calculated incremental displacements (transformed to local coordinates) as shown
below.

< ΔN a ΔQa ΔM a ΔN b ΔQ b ΔM b > T = K e Δu$ e Equation 2-25

.
where, ΔN a , ΔN b = thrust force increments (local xi direction) at nodes a and b

2-8
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

ΔQa , ΔQ b = shear force increments (local yi direction) at nodes a and b

ΔM a , ΔM b = moment increments at nodes a and b (invariant to x-y coordinates)

The fact that the incremental end forces can be recovered by multiplying the current element stiffness matrix times the
incremental displacements is a direct result of applying incremental virtual work to a free-body beam-column element
cut just short of the end nodes. Said another way, the incremental forces and moments at the cut ends are the additional
end forces that are required to hold the free-body element in equilibrium as a result of the incremental load. As a result
of the assumed interpolation functions, thrust and shear forces are constant within the element and the moment varies
linearly from end to end. Thus, ΔN and ΔM are known at every point x along the element length.

Next, the incremental strain, defined by Equation 2-2, may be evaluated at any point (x,y) by using Equation 2-8 to
evaluate the constant axial strain increment (Δa’= ΔN/EA*) and Equation 2.9 to evaluate the bending curvature (Δv”
= ΔM/EI*) as shown below. The strain distribution over any cross-section is always linear irrespective of the material
stress-strain model.

Δε ( x,y ) = ΔN/EA* + ( y*-y ) ΔM/EI* Equation 2-26

Knowing the strain distribution, the stress increment at any point (x,y) may be calculated using the incremental stress-
strain relationship (chord modulus) defined by Equation 2-3, which is dependent on the particular type of material.

Δσ ( x,y ) = E c Δε ( x,y ) Equation 2-27

where, E c = E c (σ(x,y)) = chord modulus as a function of total stress and stress history

Unlike the linear strain distribution, the incremental stress distribution is, in general, nonlinear over the cross-section
because the chord modulus is dependent on the total stress magnitude and history.

After the incremental structural responses have been computed, the total structural responses at load step i + 1 are
computed by adding the incremental values to the values of previous load step as indicated below.

q i+1 = q i + Δq Equation 2-28

where q stands for any structural responses such as displacements, stresses, strains, moments, thrusts, and so on. If the
solution represents a converged load step (or a linear system), then the q i+1 values are printed to the output file and
permanently stored in the computer in preparation for the next load step. On the other hand, if the solution has not
converged, the q i+1 values are used temporarily to compute improved estimates of the element stiffness matrices and
then discarded.

2.1.9 Nonlinear solution strategy

The key section properties, EA*, y* and EI*, are the heart of the nonlinear algorithm, and they are computed at every
node for every iteration. Shown below is a summary of the key section property definitions.

EA* = ∫E
A
c dA = effective axial stiffness of beam element Equation 2-29a

y* = ( ∫ E c y dA ) / ( ∫ E c dA) = reference axis Equation 2-29b

A A

2-9
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

EI* = ∫E
A
c (y*-y) 2 dA = effective bending stiffness of beam element Equation 2-29c

To perform the above area integrations, the chord modulus, Ec, must be estimated at each point in the cross-section. As
previously illustrated in Figure 2.1.2-1, the chord modulus for nonlinear materials is dependent on the known stress-
strain state at load step i and the unknown stress-strain at load step i + 1, thereby requiring an iterative solution strategy
as outlined below.

1. For the first iteration, the chord moduli are assumed to remain unchanged from the converged results of load
step i. Thus, the key section properties and element stiffness also remain the same as they were. Based on
these known element stiffness matrices, the first trial solution is obtained for the new incremental load vector
from load step i to load step i+1.

2. Using the first trial solution, the recovery process discussed above provides estimates for σi+1 and εi+1, which,
in turn, provides new estimates for the chord moduli depending on the particular stress-strain model assigned
to the element.

3. Knowing the new estimates for the chord moduli, the area integrations in Equations 2-29a,b,c are performed to
provide new estimates for the key section properties, EA*, y* and EI*, and hence, revised element stiffness
matrices.

4. After assembling the revised element stiffness matrices, the next trial solution is obtained, which leads to new
estimates for σi+1 , εi+1, and Ec, for each point in the cross-section. Once again, the area integrals are evaluated
to provide better estimates of the key section properties at each nodal location, and so on.

5. The above process is repeated for each iteration until two successive solutions produce essentially the same
section properties for every element within 1% relative difference. When this occurs, the solution is said to
have converged and all the updated structural responses are printed and saved for the next load step.

The area integrations for section properties, EA*, y* and EI*, are accomplished in each pipe-type subroutine. In some
cases, the integrations are done numerically and in other cases the integrations are accomplished in a semi-analytical
manner depending on the stress-strain model of the pipe material. In subsequent sections of this chapter the calculation
procedure to obtain EA*, y* and EI* for each pipe material are presented for each pipe-type material.

2-10
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

2.2 Corrugated Metal

2.2.1 Overview of corrugated metal pipe type
Corrugated metal applies to both corrugated steel and corrugated aluminum. Both metals are represented by the same
fundamental models and differ only with respect to the numerical values of key parameters such as modulus and yield
stress. Wall properties of corrugated metal are characterized by cross-sectional area, moment of inertia and section
modulus, which represent the geometry of the corrugation’s waveform per unit length. The aluminum pipe-type
subroutine and the steel pipe-type subroutine have built-in tables for commercially available corrugation sizes as well
as realistic default values for all linear and nonlinear material properties. Steel and aluminum material behavior is
simulated with a bilinear stress-strain model with an initial elastic response up to yield stress followed a hardening
plastic response, identical in tension and compression. All unloading is assumed linear elastic.

2.2.2 Design criteria for corrugated metal

Design criteria for corrugated metal includes strength limits for thrust stress against material yielding in hoop
compression, global buckling and seam strength rupture. A new strength criterion is a limit on the amount of plastic
penetration through the cross section. Finally, a performance limit on the allowable defection completes the set of
design criteria. The design criteria are summarized in the following table.

TABLE 2.2.2-1 Corrugated Metal Design Criteria.

Design Criterion Demand Capacity
(Strength limits)
(1) Thrust stress (psi) σ = N /A f y = yield strength
max max
(2) Global Buckling (psi) σ max = N max /A f b = buckling capacity
(3) Seam strength (psi) σ max = N max /A f s = seam strength
(4) Plastic Penetration *(%) pp = computed % plastic failure = 100%
(Performance Limits)
(at Service Load)
(5) Allowable deflection *(%) Δ max =computed deflect % Allowable = 5%
(Long Spans = 2%)

The above design criteria are equally applicable to working stress or LRFD design methodologies. For the working
stress approach, the demand and the capacity quantities are unfactored, and the design evaluation is given by safety
factors defined as capacity divided by demand. Typically safety factors on the order of 2.0 are desirable for strength-
related criteria. For the LRFD approach the demand and the capacity quantities are factored (see Chapter 6), and the
design evaluation is given by ratios of demand-to-capacity. Demand-to-capacity ratios less than or equal to 1.0 are
acceptably safe. Further discussion on the design criteria is provided below.

1. Thrust stress. Thrust-stress demand is determined by finding the element with the largest thrust force, Nmax,
and dividing by the cross-sectional area. The corresponding yield-strength capacity is typically 33,000 psi for
steel and 24,000 psi for aluminum.

2. Global buckling. For the buckling-strength capacity, CANDE-2007 offers the user two choices; (1) an
approximate and generally conservative estimate based on the simplified AASHTO LRFD equations 12.7.2.4-
1-2, and (2) a much more accurate solution based on CANDE’s large deformation formulation with linearized
buckling prediction.

3. Seam strength. If longitudinal seams are present in the corrugated metal culvert, the seam-strength capacity is
typically less than the material yield strength. In the absence of experimental test data, seam-strength capacity
is often specified as 67% of material yield-strength capacity.

2-11
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

4. Plastic penetration. On the demand side, the percentage of the cross-section that is strained into the plastic
range from thrust and bending is calculated directly from the nonlinear corrugated metal model. The limit of
plastic penetration is100% of the cross section, meaning cross section is incapable of carrying any additional
load. Note that some amount of plastic yielding is expected to occur in the outer fibers of most well designed
corrugated metal culverts under service loading. The plastic penetration design criterion is a precaution against
full 100% yielding of the entire cross section, not against moderate outer fiber yielding.

5. Allowable deflection. Computed deflection is the relative vertical movement between the top and bottom of
the culvert structure, and the percent deflection is relative the vertical distance. The service load limit for
allowable deflection is generally taken as 5% for all corrugated metal structures except long spans, which are
usually limited to 2% of the vertical rise.

2.2.3 Nonlinear model for corrugated metal

The one-dimensional stress-strain relationship for metal is approximated by a bilinear curve as shown in the figure
below.

Figure 2.2.3-1 Stress-strain relationship for metal culverts

The model is identical in tension and compression and applies to both steel and aluminum, which differ only in
parametric values as shown in the table below.

Table 2.2.3-1 Typical material parametric values for steel and aluminum
Metal type Yield strength Elastic Young’s modulus, E1 Upper curve modulus, E2
psi psi psi
Aluminum 24,000 10,000,000 500,000
Steel 33,000 29,000,000 0.0

The following material constants are used for plane strain conditions:

2-12
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

E e = E1 /(1-ν 2 ) = Effective plane strain modulus in elastic zone

ν = Poisson ratio of metal
r = E2/E1 = Ratio of upper-curve modulus to initial modulus
r E e = Effective plane strain modulus for upper-curve zone.

Initial section properties are defined (input) as follows:

A0 = Cross-sectional area of corrugation per unit length
I0 = Moment of inertia of corrugation about mid-height
S0 = I0 /(h/2) = Section modulus of corrugation
h = Total height of corrugation

Within the framework of the general nonlinear model, the objective is to determine the integral quantities EA*, y*, and
EI* defined by Equations 2-29a,b, & c. To accomplish this, the chord modulus Ec must be determined consistently with
the bi-linear model, and, secondly, the area of integration must be defined. With regard to the latter, a manageable
integration area can be obtained by approximating the actual corrugation geometry by a saw-tooth pattern, such that the
same area A0 and depth of section h is preserved, as shown in Figure 2.2.3-2

Figure 2.2.3-2 Saw-tooth approximation of corrugation for integrating in nonlinear range

As shown in the above figure, integration of the saw-tooth approximation yields the correct area and centroid location,
however, the moment of inertia = A0h2/12 is an approximation of actual corrugations, which are more rounded at the
top and bottom than the saw-tooth approximation. By inspecting sectional properties of standard corrugation tables the
moment of inertias from the saw-tooth approximation are 5 to 10% lower than the reported moment of inertia values.
To correct for this discrepancy, integrations for the effective bending stiffness will be multiplied by the correction
factor C0 defined in the above figure in order to reproduce the correct moment of inertia. Of course if the metal is not
yielding anywhere in the cross section, the chord modulus has the constant value Ee and may removed from inside the
integral and integrations need not be performed since the section properties are already known.

Recalling the chord modulus relates increments of stress to increments of strain, Δσ = EC Δε, the bilinear model results
in three distinct zones as illustrated in Figure 2.2.3-3. Case 1occurs when the strain increment lies entirely in the elastic
zone, bounded by the yield strain, as illustrated by Δε1 . Case 2 occurs when the strain increment starts in the elastic

2-13
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

region and exceeds the yield strain, as illustrated by Δε2. Finally Case 3 occurs when the strain increment starts at a
strain greater than yield strain and ends at a greater value, as illustrated by Δε3. All other cases wherein the strain
increment start from a large strain value and ends with a smaller strain value is elastic unloading and treated the same
as Case 1.

Figure 2.2.3-3 Illustration of strain increments in three zones; elastic, transition and plastic

With the above understanding the general chord modulus can be defined in term of the bilinear model for the elastic,
transition and plastic zones as follows;

EC = Ee (Elastic zone) Equation 2.2-1a

EC = Ee[(ε y – ε i) /(Δεi+1) + r (εi+1 – ε y) /(Δεi+1)] (Transition zone) Equation 2.2-1b

EC = r Ee (Plastic zone) Equation 2.2-1c

In the above, the transition zone chord modulus is defined generally from load step i to i+1 where ε i < ε y < εi +1 .

The foregoing considered the stress-strain relationship at a point. To obtain EA*, y*, and EI*, the stress-strain
relationship must be defined over the cross-sectional area. To this end we take great advantage of the fact the strain
profile always remains linear due to Bernoulli-Euler kinematics. Figure 2.2.3-4 illustrates two typical strain
distributions at load step i and at load step i+l. The cross-section depth is divided into the regions elastic, transition, and
plastic. The elastic region is that portion which remains totally elastic during the load step. The transition region is the
zone that begins elastic and becomes plastic during the load step. And finally, the yield region signifies the zone where
the material remains plastic.

2-14
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Figure 2.2.3-4 Typical strain profiles and resulting zone descriptions

Knowing the strain profile at step i and having obtained a trial strain profile at step i+1, it is a simple matter to locate
the elastic-transition boundaries and the transition-plastic boundaries using straight-line equations as indicated in the
above figure. The elastic-transition boundaries are located where the strain profile at step i+1intersects the yield strain
limits, and the transition-plastic boundaries are located where the strain profile at step i intersects the yield strain limits.

The chord modulus in the elastic and plastic zones is a constant value, however in the transition zones, the chord
modulus varies with y because the strain increments vary with y. To simplify the integrations of the chord modulus in
the transition zones, an average value of EC is computed from Equation 2.2-1b based on the strain increment in the
center of the transition zones. Letting z t1 and z t2 be the y-distance to the centroid of the lower and upper transition
zones and computing the strain increments at these levels, we can express the chord moduli as,

EC1 = α1 Ee Average chord modulus in lower transition zone Equation 2.2-2a

EC2 = α2 Ee Average chord modulus in upper transition zone Equation 2.2-2b

where, α1 = [(ε y – ε i) /(Δεi+1) + r (εi+1 – ε y) /(Δεi+1)], with Δεi+1 evaluated at z t1

α2 = [(ε y – ε i) /(Δεi+1) + r (εi+1 – ε y) /(Δεi+1)], with Δεi+1 evaluated at z t2

The alpha terms are constants and are within the range, r < alpha < 1.0, depending on the transition zone boundaries,
which may be truncated at the top and bottom of the cross-section. This transition zone integration technique has
proven to be very accurate. The following table summarizes the data that is required to compute the key section
properties.

Table 2.2.3-2 Zone integration parameters to compute key section properties

Zone Zone Description Zone Thickness Distance to zone Zone weight due to
Number centroid nonlinear modulus
1 Elastic Δh1 z1 w1 = 1.0
2 Lower transition Δh2 z2 w2 = α1
3 Upper transition Δh3 z3 w3 = α2
4 Lower plastic Δh4 z4 w4 = r
5 Upper plastic Δh5 z5 w5 = r

2-15
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

It should be understood that not all five zones might be active at any given time. Un-active zones are characterized
5

with Δh j = 0, so that h = ∑ Δh j always hold true.

j=1

With the above understanding, the effective axial stiffness for trial load step i+1 is given by,

EA*= ∫ E C dA = E e A* Equation 2.2-3a

A
5
where, A*= (A0 /h)∑ w jΔh j Equation 2.2-3b
j=1

The current reference axis is given by,

5
y*= ( ∫ E C y dA)/( ∫ E C dA) = (∑ w jz jΔh j )/A* Equation 2.2-4
A A j=1

And finally, the effective bending stiffness is given by,

EI* = ∫ E C (y*-y) 2 dA = E e I* Equation 2.2-5a

A
5
where, I*= C0 (A0 /h)∑ w j (Δh j3 /12 + Δh j (z j -y*)2 ) Equation 2.2-5b
j=1

In CANDE-2007 the above computations are carried out in Subroutine EMOD wherein convergence is achieved when
two successive trial solutions produce A*, y* and I* within 1% of the previous trial solution for all elements.

2-16
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

2.3 Reinforced Concrete

2.3.1 Overview of reinforced concrete pipe type
Wall sections for reinforced concrete culverts are defined by the concrete wall thickness with up to two rows of
reinforcing steel, typically placed near the inner and outer surface with specified cover depths. In tension, concrete
behavior is characterized by cracking when tension stress levels exceed the tensile strain limit. When this occurs the
pre-existing tensile stresses are redistributed to the uncracked section, and the cracked location is assumed not to heal
for any subsequent tensile loading. In compression, concrete is simulated with a tri-linear stress-strain curve. Initially,
the concrete response is linear up to a specified strain level after which the concrete exhibits plastic-hardening
behavior. When the compressive stress reaches the ultimate strength limit (fc’), the stress-strain response becomes
perfectly plastic with no increase in stress as compressive strain increases. Reinforcing steel behavior is characterized
by an elastic-plastic stress-strain model, which becomes perfectly plastic when the steel yield stress is reached in
tension or compression.

2.3.2 Design criteria for reinforced concrete

Design criteria for reinforced concrete culverts include strength limits for yielding of steel reinforcement, crushing of
concrete in compression, diagonal cracking due to shear failure, and radial cracking due to curved tension steel (also
called bowstringing). Finally, a performance limit on the allowable flexure crack width, typically taken as 0.01 inches,
completes the set of design criteria. It is believed that the proposed design criteria faithfully represent the intent and in
some cases improve the clarity of the criteria as presented in the AASHTO LRFD specifications.

Table 2.3.2-1 Reinforced Concrete Design Criteria.

Design Criterion Demand Capacity
(Strength limits)
(1) Steel yielding (psi) f = max steel stress f y = yield strength
max

(2) Concrete crushing (psi) σ max = max compression f c′ = compressive strength

(3) Shear failure (lb/in) Vmax = max shear force Vult = concrete shear capacity
(4) Radial tension failure (psi) t max = max radial stress t ult = ultimate radial strength
(Performance Limits)
(at Service Load)
(5) Allowable crack width* (in) CWmax = max crack width CWAllow = allowable CW ( 0.01inch )

The above design criteria are equally applicable to working stress or LRFD design methodologies. For the working
stress approach, the demand and the capacity quantities are unfactored, and the design evaluation is given by safety
factors defined as capacity divided by demand. Typically safety factors on the order of 2.0 are desirable for strength-
related criteria. For the LRFD approach the demand and the capacity quantities are factored (see Chapter 6), and the
design evaluation is given by ratios of demand-to-capacity. Demand-to-capacity ratios less than or equal to 1.0 are
acceptably safe. Further discussion on the design criteria is provided below.

1. Steel yielding. The maximum steel-stress demand is computed directly from the nonlinear reinforced concrete
model. On the capacity side, the steel yield strength is an input or default value, nominally 60,000 psi for
deformed bars and 65,000 psi for smooth wire fabric.

2. Concrete crushing. On the demand side, the maximum outer-fiber concrete compressive stress is determined
directly from the reinforced concrete model as a result of thrust and compression bending. The ultimate
compressive strength or capacity is an input or default value typically in the range of 4000 to 6000 psi.

3. Shear failure. The maximum shear-force demand is computed directly from the beam-column internal forces.
Shear-force capacity is the shear force causing diagonal tension failure at a given cross-section. AASHTO

2-17
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

LRFD specifications prescribe three different shear-force capacities depending on structural shape and burial
depth.

• For concrete pipes and arches, the shear capacity is specified by Vmax = Equations 12.10.4.2.5,
which yield variable values for shear capacity dependent on the values for moment, thrust and
shear around the culvert wall.

• For boxes and 3-sided structures with 2 or more feet of soil cover, the shear capacity is specified
by Vmax = Equations 5.14.5.3-1 wherein the value is dependent on moment and shear (not thrust).

• For boxes and 3-sided structures with less than 2 feet of soil cover, the shear capacity is specified
by equations in Section 5.8.3.3 that, in some cases, depend on the nature of the traverse
reinforcement (stirrups).

Clearly there is a need to unify the AASHTO shear-capacity equations because rational mechanics indicates
that shear capacity should be a function of the cross-section properties and state of loading, not on the culvert
shape or depth of burial. CANDE allows the user to select among the three choices for shear-force capacity,
however, Equation 12.10.4.2.5 is considered the best predictor of shear-force capacity because of the large
experimental data base (Reference 14). CANDE also provides the option to choose the classical shear-strength
method where shear-strength = β f c′ , where β is a specified factor, typically = 2.0.

4. Radial tension failure. On the demand side, the concrete radial tensile stress is caused by tensile forces in
curved inner cage reinforcement steel as it tends to straighten out and exerts radial tensile stresses on the
interior concrete cover thickness. The phenomenon is sometimes called bow stringing. CANDE predicts the
radial tensile stress by dividing the interior cage steel force (maximum tensile force per unit length) by the
radius of curvature of the steel cage, or letting As = steel area per unit length., we have;

t max = A s f max /Radius Equation 2.3-1

The ultimate radial stress is related to the tension strength of concrete and the structure span. The radial tensile
strength capacity is adapted from AASHTO LRFD Equation 12.10.4.2.4c-1. and restated in psi units as:

t ult = 37.92 f c′ /1000 Frt Equation 2.3-2

where Frt is a scale factor dependent on structure span, specified in the contents of Equation 12.10.4.2.4c-1.

5. Allowable crack width. The allowable crack width at service loading is generally taken as 0.01 inches in
accordance with AASHTO LRFD specification 12.10.3. CANDE offers three empirical formulas to predict
crack width; the traditional Gergely-Lutz formula (Reference 16), the recently developed Heger-McGrath
formula (Reference 15), and a simple concrete strain-based formula proposed by the author.

The Gergely-Lutz and Heger-McGrath equations are similar in form and are driven by the computed tension
steel stress when it exceeds f0 , the threshold stress for initial cracking. Their crack width predictions are
concisely stated in the equation below,

CW = S(f s - f 0 ) > 0 Equation 2.3-3

where, CW = the crack width in inches,

fs = computed tension steel stress in psi,
S and f0 = model parameters specified in the table below.

2-18
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Table 2.3.2-1 Crack width model parameters S and f0

Gergely-Lutz Heger-McGrath
(Reference 16) (Reference 15)

S = (0.122x10-6) (2tb2sl )1/3 S = (0.333x10-6) (tbsl / 2n)1/3

where, tb = concrete cover thickness (inches) where, tb = concrete cover thickness (inches)
sl = spacing between rows of steel sl = spacing between rows of steel
reinforcement (inches) reinforcement (inches)
n = number of steel reinforcement layers in
the tension zone (1 or 2)
f0 = 31.6C1(h/d)2 fc ' /ρ
f0 = 5,000.
where, C1 = 1.0 for smooth wire, 1.5 for welded wire
fabric, and 1.9 for deformed wire/bars.
h/d = total concrete thickness –to- effective
height (tension steel to other face)
ρ = reinforcement area ratio (As/h).

The Heger-McGrath prediction is well calibrated for predicting crack widths near 0.01 inches, however it
tends to underestimate the prediction of smaller crack widths. The Gergely-Lutz formula is more accurate in
predicting smaller crack widths.

The concrete strain-based crack width prediction is given by,

CW = LS (ε tension - ε crack ) Equation 2.3-4

where, ε tension = computed tensile strain in outer fiber of concrete.

ε crack = tensile strain at initial concrete cracking, a concrete model property.
LS = characteristic length for crack spacing, nominally about 10 inches.

This crack width prediction is useful for fiber-reinforced and plain concrete or when there is no tension steel to
drive the Heger-McGrath or Gergely-Lutz crack width predictions.

2.3.3 Nonlinear model for reinforced concrete

Concrete model. The concrete constitutive model was originally developed in References 3 and 17. As shown in
Figure 2.3.3-1, the model for plain concrete is a tri-linear curve in compression and an abrupt tension rupture at initial
tension cracking.

2-19
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Figure 2.3.3-1 Concrete stress-strain model and parameters.

Primary model input parameters are defined below along with typical values shown in parenthesis

ε t = concrete strain at initial tensile cracking (0.0001 in/in)

ε y = concrete strain at initial elastic limit in compression (0.0006 in/in)

ε c = concrete strain at onset of unconfined compressive strength (0.002 in/in)

fc’ = unconfined compressive strength of concrete (4,000 psi)

E1 = Young's modulus of concrete in linear zone (3,800,000 psi)

Using the above input variables, three additional parameters are derived as follows:

E2 = (fc' - E1 ε y ) / ( ε c’ - ε y ) = Young's modulus in compression yielding zone

f y c = E1 ε y = compressive stress at initial yielding (2,000 psi)

f t = E1 ε t = tensile strength at initial cracking and rupture (380 psi)

The following material constants are used for plane strain conditions:

E e = E1 /(1-ν c 2 ) = effective plane strain modulus of concrete in elastic zone

νc = Poisson ratio for concrete

2-20
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

r = E2/E1 = ratio of compression yielding modulus to initial modulus

r E e = effective plane strain modulus for compression yielding zone.

The above model has the following behavior characteristics. In tension, the concrete is linear until the initial tensile
strain exceeds the cracking strain limit ε t. When cracking occurs, the tensile stress becomes abruptly zero (redistributed
to non-cracked portions of concrete). Once a point in the cross section is cracked, the crack is assumed not to heal so
that there is no future tensile strength. Thus after initial cracking, the tensile strength parameter f t is set to zero for all
subsequent reloading in tension.

For initial compression loading, the concrete behaves linearly until the stress level reaches the initial yield strength f y c
after which plastic hardening begins to occur in the yield zone. Perfect plasticity occurs when the stress level reaches
compressive strength fc’. Unloading is elastic and with permanent plastic strain, and reloading is elastic until the stress
reaches its previous maximum value after which it follows the original stress-strain curve.

Steel model. The assumed stress-strain behavior for reinforcing steel is shown in Figure 2.3.3-2.

Figure 2.3.3-2 Stress-strain model for reinforcing steel.

Stress

fy

E0

Strain
Elastic unload path

- fy

Steel behavior is assumed to be elastic-perfectly plastic defined by the input variables:

E0 = Young's modulus for steel

fy = steel yield strength

Behavior in compression and tension is identical so that material is elastic whenever the stress magnitude is less than
the yield strength. Non-hardening plastic flow occurs when the stress attempts to exceed the yield strength. Unloading
from the plastic range is elastic and results in permanent plastic strains.

For the purposes of a plane-strain formulation, the steel modulus is denoted as;

E s = E 0 /(1-υs 2 ) = effective plane-strain modulus of steel

υs = Poisson ratio for steel

2-21
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Section geometry. Figure 2.3.3-3 shows a unit width of concrete wall with steel reinforcement located near the inner
face (bottom) and the outer face (top).

Figure 2.3.3-3 Section geometry of reinforced concrete wall

Aso
yo

Asi

yi

Unit length

Geometric measures of the reinforced concrete section are defined below:

h = concrete wall thickness

Asi = area of steel for inner cage per unit length of wall

Aso = area of steel for outer cage per unit length of wall

yi = distance to centroid of Asi from bottom face

yo = distance to centroid of Aso from bottom face

The uncracked, transformed, elastic section properties are computed as,

EA* = E e (h + (n-1)(A si +A so )) Equation 2.3.3-1

y* = ( h 2 /2 + (n-1)(y i A si + y o A so ))/A* Equation 2.3.3-2
3 2 2 2
EI* = E e [ h /12 + h(h/2-y*) + (n-1)(A si (y i -y*) +A so (y o -y*) )] Equation 2.3.3-3

2-22
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

where, n = Es/Ee = ratio of steel modulus to concrete modulus (transform method).

The uncracked, transformed elastic section properties are used for the first iteration of the first load step to obtain a trial
solution. If the loading does not cause cracking, concrete yielding or steel yielding, then the section properties as
currently computed are correct, and the next load step is considered. More generally however, nonlinear responses are
observed and iteration within the load step is required to obtain the solution as described next.

Nonlinear Strategy. We assume we have a converged solution at load step i and we seek an incremental solution for
load step i +1. Using the values of EA*, y* and EI* from the previous load step, a trial solution is obtained for the first
iteration thereby providing new estimates of the strain distribution at each reinforced concrete cross-section. To
compute the next estimate for EA*, y* and EI*, numerical integration is used over the concrete wall section to cope
with nonlinear chord modulus as illustrated in the figure below.

Figure 2.3.3-4 Strain profile from step i to i+1 and 11-point Simpson integration

Concrete chord modulus. The effective chord modulus, E’, from the known stress-strain state (σi, εi) to the estimated
stress-strain state (σi+1 , εi+1) is computed at each integration point using the concrete stress-strain curve as indicated in
Figure 2.3.3-5 and quantified in Table 2.3.3-1.

2-23
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Figure 2.3.3-5 Illustration of concrete chord moduli from σi, εi to σi+1 , εi+1

The above figure implies the chord modulus, E’, connects the known starting point (σi εi,) to the estimated end point
(σi+1, εi+1 ) where εi+1 is obtained from the trial solution and σi+1 is the corresponding stress determined from the stress-
strain relationship as quantified in the table below.

Table 2.3.3-1. Concrete chord modulus and modulus ratio for specified starting and ending zones.
Starting Zone Ending Zone Corresponding Stress Chord modulus Modulus ratio
εi εi+1 σi+1 E’ F’(y) = E’/ Ee

elastic elastic σi + EeΔε Ee 1.0

elastic yield σi + rEe(εi+1 - εy) rEe(εi+1 - εy)/ Δε < 1.0
elastic plastic f c’ (fc’ - σi)/ Δε < 1.0
yield yield σi + rEeΔε rEe r
yield plastic f c’ (fc’ - σi)/ Δε < 1.0
plastic plastic f c’ 0.0 0.0
any compress zone unloading σi + EeΔε Ee 1.0
elastic rupture 0.0 - σi/ Δε < 1.0
rupture rupture 0.0 0.0 0.0

In the above table, the concrete-modulus ratio is defined as the chord modulus divided by the initial elastic modulus,
F’(y) = E’/ Ee. Thus, F’ varies through the cross-section in the range, 0 ≤ F’(y) ≤ 1.

2-24
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Elastic unloading occurs from any compression zone whenever │εi+1│< │εi│ with the provision that neither strain is in
the rupture zone.

For the case when the starting strain is tensile but not yet at initial crack strength, 0 < εi < εt, and the ending strain is in
the rupture zone causing initial concrete cracking, we set E’ = 0 (instead of a negative softening modulus) and the pre-
existing tensile stress is distributed to the uncracked portion of the cross section by adjusting the thrust and moment
increment to compensate for the loss of stress at the cracked location. Once a point in the cross section is cracked the
tensile strength is set to zero for all future loading conditions implying the crack does not heal. Crack depth is directly
predicted from the concrete model.

The above algorithm requires maintaining data records at each integration point to keep track of the cracking history as
well as the maximum compressive stress encountered during the loading schedule to properly simulate unloading and
reloading.

Steel chord modulus. Reinforcing steel is lumped at the inner and outer cage locations, measured by yi and yo. The
chord modulus for steel at either location is easily deduced from the elastic-plastic relationship as shown in the
following table.

Table 2.3.3-2. Steel chord modulus and modulus ratio for specified starting and ending zones.
Starting Zone Ending Zone Corresponding Stress Chord modulus Modulus ratio
εi εi+1 σi+1 Es’ W’ = Es’/ Es

elastic elastic σi + EsΔε Es 1.0

elastic plastic fy (fy - σi)/ Δε < 1.0
plastic plastic fy 0.0 0.0
unloading elastic σi + EsΔε Ee 1.0

In the above table, the steel-modulus ratio is defined as the chord modulus divided by the initial elastic steel modulus,
W’ = Es’/ Es, where the ratio is in the range, 0 ≤ W’ ≤ 1.

To summarize, the relationships for chord moduli of concrete and steel are listed below:

E′(y) = E e F′(y) = concrete chord modulus at location y in cross section Equation 2.3.3-4
E′si = E s Wi′ = steel chord modulus at inner cage location, yi. Equation 2.3.3-5
E′so = E s Wo′ = steel chord modulus at outer cage location, yo. Equation 2.3.3-6

Re-expressing the steel chord moduli with concrete-transform parameters, we have

E′si = n i E e Wi′ Equation 2.3.3-7

E′so = n o E e Wo′ Equation 2.3.3-8
where, n i , n o = E s /E e -1, if concrete is not cracked around the steel
n i , n o = Es /E e , if concrete is cracked around the steel

The transform method expresses the elastic steel modulus as a multiple “n” of the concrete elastic modulus with the
understanding that the uncracked concrete area needs to be reduced by the steel area. On the other hand, if the concrete
is already cracked then further reduction of concrete area by the steel area is not required.

Key section properties. The equations for the key section properties are expressed in transformed parameters as:

2-25
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

h
EA* = E e ( ∫ F′(y)dy + n i Wi Asi + n o Wo Aso ) Equation 2.3.3-9
0
h
y* = ( ∫ F′(y)ydy + n i Wi Asi yi + n o Wo A so yo )/A* Equation 2.3.3-10
0
h
EI* = E e ( ∫ F′(y)(y-y*)2 dy + n i Wi A si (yi -y*)2 + n o Wo Aso (y0 − y*)2 ) Equation 2.3.3-11
0

The integrals over the concrete wall thickness is achieved with 11-point Simpson integration to evaluate the following
three sums where F’(yi) is the concrete modulus ratio at integration point i.

h
S1 = ∫ F′(y)dy = h/30[F′(y1 )+4F′(y 2 )+2F′(y3 )+...+F′(y11 )] Equation 2.3.3-12
0
h
S2 = ∫ F′(y)ydy = h/30[F′(y1 )y1 +4F′(y2 )y2 +2F′(y3 )y3 +...+F′(y11 )y11 ] Equation 2.3.3-13
0
h
S3 = ∫ F′(y)y 2dy = h/30[F′(y1 )y12 +4F′(y 2 )y 2 2 +2F′(y3 )y32 +...+F′(y11 )y112 ] Equation 2.3.3-14
0

Using the above integration results, the final results for the key section properties are expressed as,

EA* = E e (S1 + n i Wi A si + n o Wo A so ) Equation 2.3.3-15

y* = (S2 + n i Wi A si yi + n o Wo A so yo )/A* Equation 2.3.3-16

EI* = E e (y*2S1 -2y*S2 +S3 + n i Wi A si (yi -y*) 2 + n o Wo A so (y 0 -y*) 2 ) Equation 2.3.3-17

Iterations with in the load step continue until successive calculations for EA*, y* and EI* are within 1% relative error
for all cross section..

2-26
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

2.4 Thermoplastic Pipe

2.4.1 Overview of thermoplastic pipe type
Thermoplastic materials include high-density polyethylene (HDPE), polyvinyl chloride (PVC) and polypropylene (PP).
CANDE-2007 provides three options to characterize the wall sections for thermoplastic pipe: smooth, general or
profile. Smooth refers to a uniform wall (gun barrel) whose cross-section properties are completely defined by the wall
thickness. General refers to an arbitrary properties described generically by the wall’s area and moment of inertia per
unit length. Profile refers to the majority of manufactured plastic pipe whose wall section properties are characterized
by the geometry of profile elements including web, valley, crest, liner, and link elements.

Material properties are assumed linear elastic with default values provided for high-density polyethylene, polyvinyl
chloride and polypropylene for both short-term and long-term loading conditions. Short-term properties are generally
used when vehicular loading is dominant and long-term properties are used when earth loading is dominant. A
nonlinear local buckling algorithm is provided for the profile option wherein the profile’s section properties are reduced
in proportion to the amount of compressive strain computed in the sub elements due to local buckling.

2.4.2 Design criteria for thermoplastic pipe

Design criteria for thermoplastic pipes include strength limits for thrust stress against material failure in hoop
compression and global buckling. Another strength state is a limit on the maximum outer fiber combined strain (hoop
plus bending strain). Performance limit states include allowable vertical deflection and maximum allowable tensile
strain, dependent on type of plastic. The design criteria are summarized in the following table.

Table 2.4.2-1 Thermoplastic Design Criteria.

Design Criterion Demand Capacity
(Strength limits)
(1) Thrust stress (psi) σ = N /A f u = ultimate strength
max max
(2) Global Buckling (psi) σ max = N max /A f b = buckling capacity
(3) Combined strain (in/in) ε max = bending + thrust ε ult = ultimate strain
(Performance Limits)
(at Service Load)
(4) Allowable tensile strain ε t-max =max tensile strain ε t-allow = allowable tensile strain
(5) Allowable deflection *(%) Δ max = computed deflect % Allowable = 5%
(recommended)

The above design criteria are equally applicable to working stress or LRFD design methodologies. For the working
stress approach, the demand and the capacity quantities are unfactored, and the design evaluation is given by safety
factors defined as capacity divided by demand. Typically safety factors on the order of 2.0 are desirable for strength-
related criteria. For the LRFD approach the demand and the capacity quantities are factored (see Chapter 6), and the
design evaluation is given by ratios of factored demand-to-factored capacity. Ratios less than or equal to 1.0 are
acceptably safe. Further discussion on the design criteria is provided below.

1. Thrust stress. Thrust-stress demand is computed by the dividing the maximum thrust force in the culvert by
the cross-sectional area. The ultimate strength for thrust stress is dependent on the type plastic and the load
duration. Nominal values, taken from the ASSHTO LRFD specifications and elsewhere, are shown in the table
below.

2-27
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Table 2.4.2-2 Recommended plastic properties for short and long-term loading
Effective Young’s Modulus Ultimate strength
(PE) (PU)
Type of Short-term Long-term Short-term Long-term
plastic (ksi) (ksi) (ksi) (ksi)

HDPE – 110.0 22.0 3.00 0.90

PVC – 400.0 140.0 6.00 2.60

PP – 135.0 27.0 3.10 1.00

2. Global buckling. The thrust stress level that causes global buckling may be conservatively
approximated from the simplified AASHTO LRFD equations 12.7.2.4-1-2. A more accurate alternative
is to utilize the new large deformation formulation with buckling capacity prediction available in
CANDE-2007 (see Chapter 5).

3. Combined strain. Combined strain means the maximum outer-fiber strain from thrust and bending and the
demand is the largest combined strain anywhere in the culvert. The combined strain limit value (or capacity)
as recommended by AASHTO LRFD specifications is equal to 1.5 times the long-term strength divided by the
long-term modulus. Accordingly, HDPE = 0.06 in/in, PVC = 0.028 in/in. and PP = 0.045 in/in.

4. Allowable tensile strain. This criterion is intended to preclude cracking and crazing under service load due the
maximum outer-fiber tensile strain. The allowable tensile strain as recommended by AASHTO LRFD
specifications is 0.05 in/in for HDPE and 0.035 or 0.05 in/in for PVC depending on cell class.

5. Allowable deflection. Computed deflection is the relative vertical movement between the top and bottom of
the culvert structure, and the percent deflection is relative the vertical distance. The service load value for
allowable deflection is generally taken as 5% of the diameter for all plastic pipes; however, the deflection limit
is not directly specified in the AASHTO LRFD design specifications.

Local Buckling. Local buckling is not a direct design criterion for plastic pipe, but it does influence the demands and
capacities of the design criteria listed above. Similar to corrugated metal pipe wherein some amount of outer fiber
yielding is permitted, some amount of local buckling is permitted in plastic profile pipe. The nonlinear model to
account for local buckling in profile plastic pipe is presented next.

2.4.3 Nonlinear model for local buckling in profile plastic pipe

Although the stress-strain models for all thermoplastic pipe materials are assumed linear and characterized by the
elastic moduli in the above table, local buckling is a nonlinear phenomenon. Local buckling is caused by compressive
strains that induce the elements of profile wall pipe to deform out-of-plane in a wrinkled pattern. Section 12.12.3.5.3 of
AASHTO LRFD specifications provides a methodology to simulate the effect of local buckling by reducing the
effective area of the profile elements dependent upon the average level of compressive strain in each element. This
AASHTO methodology is incorporated into CANDE-2007 as described below.

The geometry of the profile wall is defined with two web elements and up to four horizontal elements (valley, liner,
crest and link) as shown in the figure below.

2-28
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

Figure 2.4.3-1 Profile wall geometry and sub elements

Period

Link (4) Crest (3) Link (4)

Web Web
Height

Valley (1) Liner (2) Valley (1)

The web element lengths, which form a mirror symmetric pair within each period, are defined to span the distance
between the valley and crest element so that the web elements offer support to the ends of the crest and valley elements.
Hence, the defined length of the crest element is the unsupported span between the web supports plus the length of the
web supports, taken as twice the web thickness. Similarly, the defined length of the valley element is twice the web
thickness plus the unsupported valley span between the web supports. The defined length of the liner element is the
horizontal distance between two valley elements so that the liner’s defined length and unsupported length are identical.
Similarly, the defined length of the link element is the horizontal distance between two crest elements so that the crest’s
defined length and unsupported length are identical.

Given the defined lengths and thicknesses of all elements, the angle of the web elements, and the overall height of the
profile wall, it is a straightforward matter to calculate the initial section properties of the profile wall. The area per unit
length, A*, is equal to the sum of all element areas (lengths multiplied by thickness) divided by the period length. The
wall centroid, y*, is the first area moment taken about the bottom fiber divided by the area. And finally, the moment
inertia per unit length , I*, is the second area moment taken about y* divided by the period length. Thus in the absence
of local buckling, the beam-column section properties are initially known for the first iteration of the first load step to
provide the first trial solution.

The AASHTO LRFD methodology, which accounts for local buckling by reducing the effective length of the elements
due to compressive strain, is achieved by using the following two equations.

λ = (w/t) (ε/k) > 0.673 Equation 2.4-1

where, λ = measure of propensity to buckle, called slenderness ratio

w = unsupported length of element
t = thickness of element
ε = average compressive strain in element
k = edge support coefficient (typically taken as 4.0 for fix-ended elements)

ρ = (1/λ) (1 - 0.22/λ) < 1.0 Equation 2.4-2

where, ρ = reduction factor applied to w

ρw = remaining unsupported length of element

Each profile element (web, valley, liner, crest and link ) is evaluated with above equations. Equation 2.4-1 is evaluated
to compute λ using the predicted compressive strain at each element’s centroid as determined from the trial solution. If

2-29
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

λ < 0.673, then the compressive strain is not large enough to trigger local buckling for the element under consideration
so that ρ remains = 1. Otherwise if λ > 0.673, we evaluate Equation 2.4.2 producing a value for ρ less than 1.0 wherein
ρ is the fraction of the element’s unsupported length that remains effective in the cross section. Note that the total
effective length for the crest and valley elements also includes the web support thicknesses. The element length
measures are clarified in the table below:

Table 2.4.3-1 Local buckling profile dimensions

Element length Symbol Derivation Description
measures

Defined length L Defined by user (input) The actual length of the element

Unsupported length w w = L, (for web, liner and link) The free distance between supports
w = L – 2tweb, (for crest & valley) where buckling can occur
Effective length Leff Leff = ρL, (for web, liner and link) The effective remaining length of
Leff = ρw + 2tweb, (for crest & valley) element after local buckling

Since the central portion of each element is the most prone to local buckling damage, we assume the original element
length has effective gap length equal to L - Leff located in the central portion of the element. This assumption is
needed to compute the effective moment of inertia.

Once the effective lengths of the elements have been determined and the location of the gaps understood, it is, once
again, a straightforward matter to calculate the new section properties for the next trial solution as;
A* = Sum of all effective element areas (effective lengths times thicknesses)
y* = Centroid of A* measured from bottom fiber
I* = Sum of all effective element moment of inertias about y*

The above process dovetails with the general nonlinear solution strategy described in Section 2.19 by forming EA*
and EI* where E is the appropriate linear elastic modulus (Table 2.4.2-2). When two successive trial solutions produce
the same compressive stain values within a 0.1% tolerance level for all elements at all nodes, the load step is said to
converge and we proceed to the next load step.

2-30
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

2.5 Basic Pipe Type

The so-called basic pipe-type is not associated with any particular pipe material. Therefore, it is restricted to linear
material behavior, and it is not associated with any design criteria for evaluating its safety or performance. The material
model is defined by an elastic Young’s modulus and Poisson’s ratio, and the cross-sectional properties are defined an
area and moment of inertia.

If desired, each element in the basic pipe-type group may be assigned different elastic and cross-sectional properties,
and the element group has the option to include large deformation theory and buckling capacity predictions. In addition
to academic studies, the basic pipe-type groups are useful for modeling struts and braces and discrete reinforcement in
the soil mass.

Lastly, if it is desired to use only continuum elements without beam-column elements in particular problem in the
CANDE-2007 program, the basic pipe type is declared in the input with zero elements assigned to the group.

2-31
Chapter 2 – Beam-Column Elements-Pipe Type Models CANDE-2007 Solution Methods and Formulations

(This page intentionally left blank)

2-32
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

3 SOIL MODELS
Soil models, or constitutive forms, define a relationship between stress and strain based on phenomenological
observations of material behavior at the macroscopic level. The term soil model and constitutive form are used
interchangeably throughout this chapter. CANDE-2007 offers the following suite of soil models; isotropic elastic,
orthotropic elastic, overburden dependent, Duncan and Duncan/Selig, and extended Hardin. The common feature of
these models, which have proven effective in simulating the soil layers in culvert installations, is that the models are
based on elastic-like or variable modulus constitutive forms as opposed to plasticity based constitutive forms.

Each of the soil models is discussed in subsequent sections. However in order to properly set the stage, we first discuss
the basic formulation of the continuum elements in which the soil models are contained.

3.1 Continuum Elements

Plane-strain continuum elements are used to represent the soil zones in the soil-structure system. Two continuum
element shapes are available in CANDE, triangular and quadrilateral. Both element shapes utilize identical
interpolation functions and are classified as non-conforming elements. These elements, developed by Herrmann in
Reference 11, have superior qualities in all basic deformation modes and out perform the well-known linear strain
triangle and eight-node isoperimetric quadrilateral, respectively. For example, one row of CANDE’s quadrilateral
elements is capable of properly replicating beam-bending behavior.

3.1.1 Triangle elements

The triangular element, shown below, employs area coordinates (the natural coordinate system for triangles) to define
interpolation functions for the x-displacement function u(x,y) based on 3-external nodal degrees of freedom at the
triangle vertices and 3-internal element degrees of freedom. Similarly, the y-displacement function v(x,y) is based on 3-
external and 3-internal degrees of freedom.

Figure 3.1.1-1. Characteristics of triangular continuum element

Linear interpolation functions are used for the external nodal degrees of freedom, which by themselves are equivalent
to a constant strain triangle. Superimposed on the linear displacement functions are complete quadratic interpolation

3-1
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

functions associated with internal degrees of freedom within the element. Constraint equations are applied to the
internal degrees of freedom so that the element is convergent when the element size approaches zero and passes the
patch test. The internal degrees of freedom are statically condensed at the element level resulting in a 6x6 element
stiffness matrix and a 6x1 body-load vector associated with the external degrees of freedom, and ready for global
assembly.

3.1.2 Quadrilateral elements

As shown in the figure below, the quadrilateral element is composed of two triangles wherein the x-displacement
function u(x,y) is based on 4-external nodal degrees of freedom at the vertices and 6-internal degrees of freedom within
the element. Similarly, the y-displacement function v(x,y) is based on 4-external and 6-internal degrees of freedom.

Figure 3.1.2-1. Characteristics of quadrilateral continuum element

To insure that the quadrilateral element is convergent and passes the patch test, the influence of the internal degrees of
freedom must approach zero as the element becomes vanishingly small. This requirement places three constraints on
the twelve internal degrees of freedom, resulting in nine internal degrees of freedom. The nine internal degrees of
freedom are statically condensed at the element level resulting in an 8x8 element stiffness matrix and an 8x1 body-load
vector associated with the external degrees of freedom, and ready for global assembly.

3.1.3 Finite element development

Interpolation functional forms. The interpolation functions used for triangle and quadrilateral elements may be
generically expressed using the notation presented in Chapter 1 as,

3-2
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Δu e = NΔuˆ e Equation 3.1-1

Δv e = NΔvˆ e Equation 3.1-2

where, Δu e , Δve = displacement functions for x- and y-directions in triangle or quadrilateral element.
Δuˆ e , Δvˆ e = column-vector of degrees-of-freedom for x and y displacement functions in element.
N = row-vector of interpolation functions in area coordinates, also expressible in x and y variables.
Of course, the length of the interpolation row-vector is larger for the quadrilateral element than for the triangle element,
as are also the lengths of the degree-of-freedom column-vector.

Strain-displacement relationship. Small strain and small deformation theory are used for all continuum elements.
Accordingly, the strain components for plane strain are given by the partial derivatives of the displacement functions as
expressed in the matrix relationship below,

⎛ Δε x ⎞ ⎛ N,x 0 ⎞
⎜ ⎟ ⎜ ⎟ ⎛ Δuˆ ⎞
⎜ Δε y ⎟ = ⎜ 0 N,y ⎟ ⎜ e ⎟ Equation 3.1-3
Δvˆ
⎜ Δγ ⎟ ⎜ N,y
⎝ ⎠ ⎝ N,x ⎟⎠ ⎝ e ⎠

where, Δε x , Δε y , Δγ = strain increments for x-direction, y-direction and shear components, respectively.
N,x & N,y = partial derivatives of interpolation functions in the row vector with respect to x and y.

Stress-strain relationship. All soil models conform to a general constitutive form relating increments of stress to
increments of strain via a matrix of variable coefficients dependent on the stress-stain state at step i and step i+1 as
indicated below,

⎛ Δσ x ⎞ ⎛ C11 C12 C13 ⎞ ⎛ Δε x ⎞

⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Δσ y ⎟ = ⎜ C21 C22 C23 ⎟ ⎜ Δε y ⎟ Equation 3.1-4
⎜ Δτ ⎟ ⎜ C31 C32 C33 ⎟⎠ ⎜⎝ Δγ ⎟⎠
⎝ ⎠ ⎝

where, Δσ x , Δσ y , Δτ = stress increments for x-direction, y-direction and shear components, respectively.
Cij = variable coefficients generally dependent on current stress and strain state.

Components of the constitutive matrix are symmetric (Cij = Cji) and are explicitly defined for each soil model in
subsequent sections.

Element stiffness matrix. Using the above relationships, the element stiffness for the triangle or quadrilateral is
obtained by integrating the following matrix products,

T
⎛ N,x 0 ⎞ ⎛ C11 C12 C13 ⎞ ⎛ N,x 0 ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
∫
k e = ⎜ 0 N,y ⎟ ⎜ C21 C22
⎜C
C23 ⎟ ⎜ 0 N,y ⎟ dA Equation 3.1-5
A ⎜ N,y N,x ⎟
⎝ ⎠ ⎝ 31 C32 C33 ⎟⎠ ⎜⎝ N,y N,x ⎟⎠

3-3
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Integration of the interpolation functions is performed exactly using area coordinates whereas the components of the
constitutive matrix are taken as constant within the element based on the stress-strain state at the center of the element.
After the internal degrees of freedom are statically condensated from the element, the element stiffness for the triangle
element is a 6x6 matrix associated with the 6-external degrees of freedom, two per node. The quadrilateral element
stiffness is an 8x8 matrix, associated with the 8-external degrees of freedom, two per node.

3-4
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

3.2 Isotropic Linear Elastic

The linear elastic soil model in isotropic form is the simplest soil model. Linear elastic implies the soil stiffness remains
constant irrespective of the stress state, and isotropic implies the soil stiffness is uniform in all spatial directions. The
model is useful for characterizing stiff in-situ soils and pre-consolidated soils such as the soil remaining after
excavation. Moreover when the exact character of the soil is not well known, the linear-elastic isotropic soil model is
useful for parametric studies and/or to conservatively represent soil stiffness with modest moduli values.

The isotropic form of the elastic constitutive matrix is expressed by the following incremental stress-strain relationship,

⎛ Δσ x ⎞ ⎛ C11 C12 0 ⎞ ⎛ Δε x ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Δσ y ⎟ = ⎜ C12 C11 0 ⎟ ⎜ Δε y ⎟ Equation 3.2-1
⎜ Δτ ⎟ ⎜ 0 0 C33 ⎟⎠ ⎜⎝ Δγ ⎟⎠
⎝ ⎠ ⎝
where, C11, C12 and C33 are material constants defined by two elastic parameters. In the above matrix, the C22
coefficient is identified as C11 to emphasize they are identical in value for the isotropic case.

In CANDE, the elastic parameters used to characterize C11, C12 and C33 are Young’s modulus and Poisson ratio as
shown in the second column of Table 3.2-1. The exact equivalent using other pairs of elastic parameters are shown in
the third column for the confined modulus and lateral coefficient, and in the fourth columns for the bulk modulus and
shear modulus.

Table 3.2-1 Components for constitutive matrix defined by elastic parameters

Elastic parameter equivalent pairs

Components of
Constitutive matrix E = Young’s modulus Ms = confined modulus B = bulk modulus
ν = Poisson ratio K0 = lateral coefficient G = shear modulus
(E, ν) (Ms, K0) (B, G)
C11 = E(1- ν) / (1+ ν)(1-2 ν) Ms B + (4/3)G
C12 = Eν / (1+ ν)(1-2 ν) Ms K0 B - (2/3)G
C33 = E / 2(1+ ν) Ms (1-K0)/2 G
The inference in the above table is that if any two elastic pairs are known, they may be equivalently expressed with
any other elastic pairs using the above relationships.

Depending on the soil quality and compaction, representative ranges of Young’s modulus and Poisson ratio are shown
in Table 3.2-2 for three broad classes of soil.

Table 3.2-3 Representative ranges of Young’s modulus and Poisson ratio

Elastic parameters, nominal range
Soil type
Young’s modulus, E Poisson ratio, ν
psi (-)
Granular 600 to 2000 0.30 to 0.35
Mixed 400 to 1400 0.30 to 0.40
Cohesive 200 to 400 0.33 to 0.40

Well-compacted soils are characterized by the high-range values of Young’s modulus, whereas poorly compacted soils
are characterized by the low-range values.

3-5
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

3.3 Orthotropic Linear Elastic

Like the isotropic model above, the linear-elastic designation means the stiffness of the orthotropic model remains
constant irrespective of the stress state. However unlike the uniform isotropic model, the stiffness of the orthotropic
model may be different in one direction, say x’, than it is an orthogonal direction say, y’. Some stratified sedimentary
soil deposits exhibit orthotropic properties, but this is rather rare in culvert installations. A more common occurrence of
orthotropic stiffness is due to man-made inclusions in soils such as reinforced earth and geo-textile fabrics.

The orthotropic form of the elastic constitutive matrix is expressed by the following incremental stress-strain
relationship wherein the x-y coordinate system is assumed to be aligned with the principal material axis.

⎛ Δσ x ⎞ ⎛ C11 C12 0 ⎞ ⎛ Δε x ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Δσ y ⎟ = ⎜ C12 C22 0 ⎟ ⎜ Δε y ⎟ Equation 3.3-1
⎜ Δτ ⎟ ⎜ 0 0 C33 ⎟⎠ ⎜⎝ Δγ ⎟⎠
⎝ ⎠ ⎝
C11, C22, C12 and C33 are four independent material constants, however their values are constrained by the requirement
the constitutive matrix must be positive definite. If the x-y coordinate system is not coincident with the principal
material directions, then the angle between the two coordinate systems is used to transform the constitutive matrix to
the x-y system.

3.3.1 Orthotropic properties from testing specimens

Direct measurements to characterize C11, C22, C12 and C33 in naturally occuring orthotropic soils can be achieved with a
tri-axial testing machine by performing K0-tests wherein the axial load is applied and the confining pressure is adjusted
so that there is no net lateral strain, also called a confined compression test, or uni-axial strain test. For the case when
the material x-axis is aligned with specimen’s vertical axis, the results of the K0-test may be used to determined the
following three coefficients,

C11 = σaxial /ε axial Equation 3.3-2

C12 = σ 0 /ε axial Equation 3.3-3
C33 = (σ axial - σ 0 ) / 2ε axial Equation 3.3-4

where, σ axial = net axial stress including confining pressure,

ε axial = net axial strain including confining strain,
σ 0 = confining pressure.
In a similar manner, the K0-test may be repeated with the test specimen rotated 90 degrees such that the material y-axis
is aligned with the vertical direction in order to determine the last coefficient C22 as,

C22 = σ axial /ε axial Equation 3.3-5

This second set of test results may also be used to confirm the previously computed results for C12 and C33.

3.3.2 Orthotropic properties for reinforced soil

Placing strips of metal or geo-textile fabrics into soil has proven to be very effective in improving the directional
stiffness of soils in embankments and pavement systems. Directional soil reinforcement has also been studied for use in
long-span culvert installations in Reference 18. As developed Reference 18 and outlined below, the orthotropic model
is useful for simulating the directional stiffness obtained from reinforced soil. This is based on the unit cell concept.

The upper portion of Figure 3.3.2-1 illustrates a unit cell of soil with a single reinforcing strip at the center of the cell
extending in the x-direction. The y-z face of the cell measures d inches by b inches with the inference that the

3-6
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

reinforcement strips repeat themselves every d inches in the y-direction and every b inches in the z-direction to form a
uniform grid in some large zone of soil.

Figure 3.3.2-1 Unit cell concept for modeling reinforced earth

Assuming the reinforcing strips act as one-dimensional elements and remain bonded to the soil, the composite stiffness
in the x-direction is determined by applying a uniform strain εx while holding all other strains in the soil cell to zero. As
illustrated in the bottom of Figure 3.3.2-1, the uniaxial straining produces stresses in the soil and the reinforcing strip as
expressed below.

σsoil = M s ε x Equation 3.3-6

σ rs = E rs ε x Equation 3.3-7

where, σ soil , σ rs = stress in x-direction for soil and reinforcing strip, respectively.
Ms = confined modulus of isotropic soil
Ers = Young’s modulus of reinforcing strip

To determine the composite stress-strain modulus in the x direction, we evoke force equilibrium such that the net force
from the composite stress over the cell face is equal to force contributions from the soil and reinforcement,

3-7
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

A cell σ x = A soil σ soil + A rs σ rs Equation 3.3-8

where, A cell = bd = area of unit cell face

A soil = bd -wt = soil area per unit cell
A rs = wt = reinforcing strip area per unit cell

Replacing σ soil , σ rs in the above equilibrium equation with their strain equivalents, we have the key stress-strain
expression for the x-direction as,

σ x = (M s + α(E rs -M s ))ε x Equation 3.3-9

where, α = wt/bd = ratio of reinforcement area to unit cell area.

To simplify and summarize, let the second term in the above expression be written as, R = α (Ers – Ms), which is the
contribution of reinforcing strip to the C11component of the constitutive matrix for a unit cell area of soil. Thus,
the final form of the orthotropic constitutive matrix may be written as,

⎛ Δσ x ⎞ ⎛ M s +R Ms K 0 0 ⎞ ⎛ Δε x ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Δσ y ⎟ = ⎜ M s K 0 Ms 0 ⎟ ⎜ Δε y ⎟ Equation 3.3-10
⎜ Δτ ⎟ ⎜ 0 0 G ⎟⎠ ⎜⎝ Δγ ⎟⎠
⎝ ⎠ ⎝
where, R = α(Ers – Ms) = additional soil modulus stiffness in x-direction due to reinforcing strips
α = wt/bd = ratio of reinforcement area to area of unit cell
Ers = Young’s modulus of reinforcing strips
Ms = confined modulus of isotropic soil = E(1- ν)/(1+ ν)(1-2ν)
K0 = lateral coefficient of isotropic soil = ν/(1- ν)
G = shear modulus of isotropic soil = E /2(1+ ν)

To apply the above in the CANDE program, input C(1,1) = Ms + R, C(1,2) = MsK0, C(2,2) = Ms, and C(3,3) = G.

3.4 Overburden Dependent Soil Model

The overburden-dependent model is the application of the isotropic elastic model in a series of steps. Each step
represents an increment of soil fill or overburden pressure so that the elastic moduli are modified at each step to
account for an increased stiffness due to increased confining pressure.

Implicit in the model is the assumption that soil stiffness increases with overburden pressure. This assumption holds
true when the soil is in a state of confined compression (one-dimensional straining), wherein increased overburden
pressure further increases the lateral confining pressure and the soil tends to stiffen. On the other hand if the soil is
unconfined, then increased overburden pressure will not stiffen the soil but, on the contrary, stiffness will be reduced
due to shear straining as observed in standard tri-axial tests.

The significant point is that overburden-dependent models are only valid insofar as the soil is predominantly in a state
of confined compression. Generally, gravity loading of the soil promotes states of confined compression; however, in
regions of strong interaction, such as certain areas in the vicinity of the pipe or around other inclusions, the assumption
of confined compression is questionable.

3-8
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

3.4.1 Input data for overburden dependent model

The overburden-dependent model is characterized using the stress-strain data obtained from a confined compression
test as illustrated in Figure 3.4.1-1. In this test, increasing levels of axial stress are applied to the top surface of the soil
specimen and the corresponding axial strains are measured for each stress level, all other normal strains are zero due to
the confining chamber.

Figure 3.4.1-1 Illustration of confined compression test data and secant confined moduli

As illustrated in the figure, the secant confined modulus is defined as, M si = σ i /ε i , and generally increases as
overburden pressure increases. The word secant is used to denote that Msi is the slope of total stress to total strain, not
an incremental relationship. Experimental results from confined compression test are used to compute and list secant
moduli values as function of overburden pressure as depicted in the table below for n data points.

Table 3.4.1-1 Data table of secant moduli versus overburden pressure.

Overburden Pressure Secant confined Modulus Secant Young’s modulus
(vertical stress) Msi = σi/εi Esi = Msi (1+ν)(1-2 ν)/(1- ν)
σ1 Ms1 Es1
σ2 Ms2 Es2
σ3 Ms3 Es3
* * *
* * *
* * *
σn Msn Esn

For most soils the lateral soil pressure exerted on the sides of the chamber typically increases in direct proportion to the
overburden pressure. Thus the lateral coefficient K0, which is the ratio of lateral pressure to vertical stress, remains
practically constant for all levels of vertical stress. Since Poisson ratio ν = K0/(1+ K0), a constant value of Poisson ratio
is used to provide the second elastic parameter for the overburden dependent model. Accordingly as shown in the last
column of the above table, secant values for Young’s moduli are easily computed by multiplying secant confined

3-9
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

moduli by a constant factor β = (1+ν)(1-2 ν)/(1- ν), which is derived from elasticity parameter equivalent relationships
(see Table 3.2.1-1).

To briefly recap, Table 3.4.1-1 is representative of the data needed to exercise the overburden model in a typical soil-
structure problem. CANDE-2007 has “canned tables” of secant Young’s moduli corresponding to increasing values for
overburden pressure for typical soils that may be used in lieu of actual test data.

3.4.2 Overburden model development

To incorporate the overburden pressure data into the incremental stress-strain relationship, it is necessary to convert the
secant moduli data into chord moduli relating increments of vertical stress to increments of vertical strain in going from
load step i to step i +1as illustrated in the figure below.

Figure 3.4.2-1 Incremental stress-strain relationship and chord modulus

Specifically for each plane-strain element in the soil system, the incremental constitutive matrix has the form,

⎛ Δσ x ⎞ ⎛ C11 C12 0 ⎞ ⎛ Δε x ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Δσ y ⎟ = ⎜ C12 C11 0 ⎟ ⎜ Δε y ⎟ Equation 3.4-1
⎜ Δτ ⎟ ⎜ 0 0 C33 ⎟⎠ ⎜⎝ Δγ ⎟⎠
⎝ ⎠ ⎝
where C11, C12, and C13 are chord moduli dependent on the stress state at load step i and step i+1.

Letting σi and εi represent the known vertical stress and strain at load step i in element #n, the increment of vertical
stress in element #n is estimated by determining the vertical pressure contributions of all elements above element #n
that enter the soil system in load step i+1. Tersely, this is stated as,

Δσ ≈ increment of overburden pressure from added soil elements above element #n

The algorithm that computes the estimate for Δσ utilizes the finite element mesh topology and element material
property data to determines which elements entering the system during load step i+1 are above element #n along with
their contributing pressures. The algorithm is laborious but straightforward.

3-10
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Once the estimate for Δσ is obtained, the chord moduli (see above figure) may be determined as,

Δσ
C11 = Equation 3.4-2
(σ i +Δσ)/M i+1 - ε i

C12 = K 0 C11 Equation 3.4-3

1
C33 = (1-K 0 )C11 Equation 3.4-4
2

where, σi , εi = known stress and strain values in element #n at load step i.

M i+1 = secant confined modulus interpolated from data table at pressure σi + Δσ .
K 0 = ν/(1- ν) = constant lateral coefficient (ν = Poisson ratio)

Herein lies the advantage of an overburden-dependent model over the nonlinear models to be described next. That is, to
advance the solution from step i to i+1, the overburden-dependent chord moduli are determined based on the soil
layering within the finite element mesh, not on a trial solution. Thus, there are no iterations within the load step and
convergence is not an issue with the overburden dependent soil model.

3.4.3 Overburden dependent secant moduli data tables in CANDE

For reference, reasonable ranges of Young’s secant moduli are provided in Table 3.4.3-1 along with suggested values
for Poisson ratio. The table offers three broad categories of soil; granular, mixed and cohesive for two compaction
levels, fair and good.

3-11
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Table 3.4.3-1 – Secant Young’s modulus versus overburden pressure

Soil Class→ Granular Mixed Cohesive
Compaction→ Good Fair Good Fair Good Fair
MATNAM→ GGOOD GFAIR MGOOD MFAIR CGOOD CFAIR
Overburden Young’s Young’s Young’s Young’s Young’s Young’s
Pressure Modulus Modulus Modulus Modulus Modulus Modulus
psi psi psi psi psi psi psi
5 1,100 550 600 400 250 150

10 1,300 750 850 550 325 200

15 1,500 850 1,000 600 375 225

20 1,650 1,000 1,100 700 375 250

25 1,800 1,100 1,200 750 400 250

30 1,900 1,150 1,250 800 400 250

40 2,100 1,300 1,350 900 400 250

50 2,250 1,400 1,450 900 400 250

Poisson ratio
(pressure 0.30 to 0.35 0.30 to 0.40 0.33 to 0.40
independent)

The above values are a composite of seven different references as well as experimental data from the original CANDE
investigation (Reference 1). As a whole the data values are conservative (low side) and are suitable for design if actual
soil data is not available.

3-12
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

3.5 Duncan and Duncan/Selig Soil Models

The Duncan and Duncan/Selig soil models are variable-modulus elasticity formulations using stress-dependent
equations for Young’s modulus and bulk modulus. The Duncan form and Duncan/Selig form are very similar, differing
slightly in the expression for the bulk modulus function. Both forms of the soil models are considered to be very
representative of actual soil behavior particularly for representing the stress-dependent behavior of backfill soil in
culvert installations. Basically, the models exhibit stiffening of constitutive moduli when confining stress increases and
softening when shear stress increases.

Duncan’s original work for the Young’s modulus formulation, which is based on a hyperbolic stress-strain relationship,
dates back to 1970 (Reference 19) and remains today as originally developed. In 1978 (Reference 20), Duncan and his
colleagues introduced a variable bulk modulus to serve as the second constitutive parameter, replacing the previous
assumption of a constant Poisson ratio. The bulk modulus function is based on a power law function of confining
pressure. Today, Duncan’s original Young’s modulus formulation together with the power-law bulk modulus function
is referred to as the Duncan soil model.

Selig proposed an alternative form of the bulk modulus function in 1988 (Reference 21). Selig’s bulk modulus function
is based on an observed hyperbolic relationship between volumetric strain and hydrostatic pressure from soil specimens
in hydrostatic compression. Thus, the so-called Duncan/Selig soil model is based on Duncan’s original Young’s
modulus formulation and Selig’s hyperbolic bulk modulus formulation. Selig also performed independent tests to
characterize the parameters of the Duncan/Selig model.

3.5.1 Plane-strain constitutive matrix

From an overall perspective, the Duncan and the Duncan/Selig soil models are used to define the nonlinear components
of an isotropic, elasticity-based constitutive matrix for plane-strain conditions as expressed below.

⎛ Δσ x ⎞ ⎛ C11 C12 0 ⎞ ⎛ Δε x ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Δσ y ⎟ = ⎜ C12 C11 0 ⎟ ⎜ Δε y ⎟ Equation 3.5-1
⎜ Δτ ⎟ ⎜ 0 0 C33 ⎟⎠ ⎜⎝ Δγ ⎟⎠
⎝ ⎠ ⎝

where, Δσ x , Δσ y , Δτ = stress increments for x-direction, y-direction and shear components, respectively
Δε x , Δε y , Δγ = strain increments for x-direction, y-direction and shear components, respectively
C11 , C12 , C33 = nonlinear coefficients dependent on Young’s modulus and bulk modulus functions.

The table below shows the relationship between the constitutive matrix components and the elastic functions.

3-13
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Table 3.5.1-1 Components for constitutive matrix relationship to moduli functions

Components of E = Young’s modulus E = Young’s modulus

Constitutive matrix B = bulk modulus ν = Poisson ratio
(E, B) (E, ν)

C11 = 3B(3B + E) E (1- ν )

9B - E (1+ ν )(1-2ν )
C12 = 3B(3B - E) Eν
9B - E (1+ ν )(1-2ν )
C33 = 3BE E
9B - E 2 (1+ ν )

The middle column in the above table defines the matrix coefficients C11, C12 and C13 in terms of Young’s modulus and
bulk modulus, which are the elastic parameter functions developed in the current Duncan and Duncan Selig models.
The third column defines the matrix coefficients in terms of Young’s modulus and Poisson ratio, which was the 1970
form of the Duncan model. We will return to the implications of the constitutive matrix after the functional forms of
Young’s modulus and bulk modulus are presented.

3.5.2 Duncan Young’s modulus development

Duncan’s formulation is based on experimental observations of soil behavior from standard tri-axial tests. A standard
tri-axial test is conducted by placing a cylindrical soil specimen in a pressure chamber and initially subjecting the
specimen to a uniform hydrostatic pressure, called σ3. Next a steadily increasing axial load is applied to the specimen
producing a net axial stress, called σ1, which includes the hydrostatic pressure. Note that σ3 and σ1 are principal
stresses in the lateral and axial directions, respectively. The difference in principal stresses σ1 – σ3 is called the deviator
stress and is equal to twice the maximum shear stress occuring on 45-degree plane.

As the axial stress increases, axial strain is computed by measuring the axial shortening of the specimen divided by the
specimen length. Note the measured axial strain does not include the initial hydrostatic strain.

The dashed curve in Figure 3.5.2-1 is a plot of deviator stress versus axial strain for a typical tri-axial test. Here, (σ1 –
σ3) f is the measured maximum deviator stress at failure, and Ei is the initial slope representing Young’s modulus at
zero deviator stress. As discussed subsequently, (σ1 – σ3) f and Ei are dependent on the hydrostatic stress level.

3-14
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Figure 3.5.2-1 Deviator stress versus axial strain for tri-axial test and hyperbolic approximation

Duncan hyperbolic function. Duncan’s fundamental insight is that the experimental curve is fairly well approximated
by equating the deviator stress to a hyperbolic function of axial strain as follows,

ε
σ1 - σ3 = Equation 3.5-2
1 ε
+
E i (σ1 - σ3 ) u

where, Ei = initial slope for Young’s modulus (dependent on minimum principal stress)
(σ1 - σ 3 ) u = ultimate deviator stress from hyperbolic model (dependent on minimum principal stress)
ε = axial strain, not including initial hydrostatic strain
As illustrated in by the solid curve in the above figure, the hyperbolic function provides a good representation of the
actual test data up to the point of actual failure, (σ1 – σ3)f . Thereafter the hyperbolic function tends to overshoot the
actual softening behavior as the hyperbolic function asymptotically approaches the limit, (σ1 – σ3)u . In order to
preserve the good curve fit from zero to actual failure, it should be evident that (σ1 – σ3)u cannot, in general, be set
equal to the actual failure deviator stress, (σ1 – σ3)f . Rather, the Duncan model introduces a model parameter called the
failure ratio defined as,
(σ1 -σ3 )f
Rf = ≤ 1.0 Equation 3.5-3
(σ1 -σ3 ) u

Given that (σ1 – σ3)f is characterized from actual experimental data, then Equation 3.5-3 may be used to define (σ1 –
σ3)u for the hyperbolic function.

To complete the hyperbolic model, functional forms for (σ1 – σ3)f and Ei must be defined based on a sequence of tri-
axial tests. Figure 3.5-2 illustrates the typical behavior of a specific soil specimen subjected to a series of tri-axial tests
with increasing confining pressure.

3-15
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Figure 3.5.2-2 Typical behavior of a soil specimen in a sequence of tri-axial tests.

Initial Young’s modulus. Based on a wide variety of soil tests as typified by the above figure, the initial modulus is
observed to increase with confining pressure according to the following power law,

E i = KPa (σ3 /Pa ) n Equation 3.5-4

where, Ei = initial Young’s modulus when deviator stress = 0

K = dimensionless magnitude of initial Young’s modulus
Pa = atmospheric pressure used to nondimensionalize parameters K and n
n = power-law coefficient usually less than 1.0
σ3 = confining pressure or minimum magnitude of principle stress
Mohr-Coulomb failure envelope. As portrayed in the Figure 3.5-2, the deviator stress at failure increases with
confining pressure. Based on the well-known Mohr-Coulomb failure theory, the (σ1 – σ3)f i data points define a failure
envelope in terms of normal stress and shear stress as depicted in the figure below.

Figure 3.5.2-3 Failure envelope in normal and shear stress space

3-16
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Typically, the Mohr-Coulomb failure surface is a straight line defined by two constants, C and φ0, representing the
cohesion intercept and a constant soil friction angle, respectively. As illustrated in the above figure, a more general
form of the Mohr-Coulomb failure surface is used in this development wherein the soil friction angle is a decreasing
function of the confining pressure. Often, this is a better representation of actual soil behavior.

Based on the generalized Mohr-Coulomb theory, the tri-axial deviator stress at failure is characterized as follows,

2C cosφ + 2σ3 sinφ

(σ1 - σ 3 )f = Equation 3.5-5
1- sinφ

where, φ = φ 0 - Δφ log10 (σ 3 /Pa ) = variable soil friction angle Equation 3.5-6

C = cohesion intercept, units of stress
φ0 = initial soil friction angle
Δφ = reduction of soil friction angle for 10-fold increase in σ3

Tangent Young’s modulus. Tangent Young’s modulus is equal to the derivative of axial stress to axial strain under
the conditions of tri-axial testing wherein the confining stress is constant. Since all the terms are now identified in
Equation 3.5-2, we can compute tangent Young’s modulus by forming the derivative dσ1/dε and replacing strain values
with equivalent stresses via Equation 3.5-2 to arrive at,

R f (1- sinφ)(σ1 - σ3 ) 2
E t = E i [1 - ] Equation 3.5-7
2(Ccosφ + σ3sinφ)

where, E t = tangent Young’s modulus dependent on stress state

σ1 = maximum compressive principal stress
σ 3 = minimum compressive principal stress

Equation 3.5.7 is fundamental for both the Duncan soil model and the Duncan/Selig soil model.

3.5.3 Bulk modulus formulations

By definition, the tangent bulk modulus is the change in mean stress divided by the change in volumetric strain, which
is expressed as

dσ m
Bt = Equation 3.5-8
dε vol
where, Bt = tangent bulk modulus, usually stress dependent for soils.
σ m = (σ11 + σ 22 + σ 33 )/3 = mean average stress, also called hydrostatic stress
ε vol = dV/V0 = ε11 + ε 22 + ε 33 = volumetric strain

Duncan bulk modulus function. Duncan used tri-axial test data including measurements of volumetric strain to obtain
experimental values for the tangent bulk modulus computed as,

1
Bmeasured = Δσaxial /Δε vol Equation 3.5-9
3
For a given confining pressure, the change in axial stress and corresponding change in volumetric strain are measured
when the axial stress reaches 70% of the failure stress or the volumetric strain peaks in contraction, whichever occurs
first.

3-17
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Based on a series of tri-axial tests with increasing confining pressure, Duncan proposed a power-law function (very
similar to the initial Young’s modulus function) to describe the tangent bulk modulus as,

Bt = K b Pa (σ 3 /Pa ) m Equation 3.5-10

where, K b = dimensionless magnitude of tangent bulk modulus

Pa = atmospheric pressure used to nondimensionalize parameters Kb and m
m = power-law coefficient usually less than 1.0
σ3 = confining pressure or minimum magnitude of principle stress

Selig bulk modulus function. Selig developed an alternate form of the tangent bulk modulus function based on
hydrostatic tests. In the hydrostatic test the soil specimen is compressed under increasing confining pressure applied
equally in all directions and the change of volume is measured as function of confining pressure as depicted in the
figure below.

Figure 3.5.3-1 Hydrostatic test and Selig’s hyperbolic model

Selig observed that the experimental curves relating mean stress to volumetric strain is reasonably described by a
hyperbolic equation in the form,

Bi
σm = ( )ε vol Equation 3.5-11
1 - ε vol /ε u

where, Bi = initial tangent bulk modulus when volumetric strain = 0.

ε u = ultimate volumetric strain at large hydrostatic stress.

As illustrated in Figure 3.5.3-1, the tangent bulk modulus is determined by forming the derivative dσm/dεvol and again
using Equation 3.5-11 to replace volumetric strain with hydrostatic stress to get the final result,

3-18
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Bt = Bi [1+ σ m /(Bi /ε u )]2 Equation 3.5-12

where, Bt = Selig form of tangent bulk modulus dependent on mean hydrostatic stress.

3.5.4 Summary of Duncan and Duncan/Selig soil models.

Both Duncan and Duncan/Selig employ tangent Young’s modulus given by,

R f (1- sinφ)(σ1 - σ3 ) 2
E t = E i [1 - ] Equation 3.5-13
2(Ccosφ + σ3sinφ)
σ3 n
E i = KPa ( ) Equation 3.5-14
Pa
φ = φ 0 - Δφ log10 (σ 3 /Pa ) Equation 3.5-15

where, E t = tangent Young’s modulus dependent on complete stress state,

Ei = initial tangent dependent on minimum compressive stress,
φ = angle of internal friction dependent on minimum compressive stress,
σ1 = maximum compressive principal stress,
σ 3 = minimum compressive principal stress,
Pa = atmospheric pressure, a constant for units of stress.

Material properties required for the tangent young’s modulus functions are,
K = dimensionless magnitude of initial Young’s modulus,
n = power-law coefficient for initial modulus usually less than 1.0,
C = cohesion intercept for failure, units of stress,
φ0 = initial soil friction angle of failure surface,
Δφ = reduction of soil friction angle for 10-fold increase in σ3,
R f = failure ratio of actual to model failure stress usually les than 1.0.

For the second elasticity function, the choice is between Duncan’s tangent bulk modulus power law or Selig’s tangent
bulk modulus derived from a hyperbolic relationship between hydrostatic pressure and volumetric strain.

Duncan’s tangent bulk modulus power-law function is given by,

Bt = K b Pa (σ 3 /Pa ) m Equation 3.5-16

where, σ3 = minimum magnitude of compressive principle stress,

Pa = atmospheric pressure, a constant for units of stress.

Material properties required for Duncan’s tangent bulk modulus power law are,
K b = dimensionless magnitude of tangent bulk modulus,
m = power-law coefficient usually less than 1.0.

Selig’s tangent bulk modulus function is given by,

3-19
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Bt = Bi [1+ σ m /(Bi /ε u )]2 Equation 3.5-17

where, σ m = mean hydrostatic stress.

Material properties required for Selig’s tangent bulk modulus function are,
Bi = initial tangent bulk modulus when hydrostatic stress is zero,
ε u = ultimate volumetric strain at large hydrostatic stress.

3.5.5 Behavioral characteristics and special considerations

The behavioral characteristics and limitations of the Duncan and the Duncan/Selig soil model are nearly identical and
are enumerated below along with discussions on special treatments for limiting cases. Recall that maximum shear stress
is equal to half the difference in principal stresses, (σ1 – σ3)/2.

• If confining pressure increases while shear stress is held constant, Et and Bt become stiffer. However if
maximum shear increases with constant confining pressure, Et becomes softer while Bt remains constant.

• Shear failure is said to occur when Et approaches zero as shear stress increases to the failure limit. To avoid
numerical problems, the bracketed term in Equation 3.5-13, which varies from 1.0 to 0.0 as shear stress
increases, is limited to the minimum value 0.05 so that Et = 0.0025Ei is the minimum tangent Young’s
modulus, which is effectively a near-zero stiffness.

• Tension failure is said to occur when the minimum compressive stress σ3 becomes tensile wherein most soils
breakdown and lose all stiffness. This is simulated in the soil models by assigning a small limiting value to the
minimum compressive stress, σ3/Pa = 0.1, so that Et and Bt are near-zero stiffness values whenever the
minimum principal stress approaches tension.

• Although Et and Bt are developed as two independent functions, energy considerations require constraints
between the functions (e.g. see Table 3.5-1). The soil models programmed in CANDE satisfy the energy
constraints by enforcing,

Et > 0, and Et/3 < Bt < 8Et.

Lower and upper limits on Bt are equivalent to maintaining Poisson ratio in the range, 0 < ν < 0.48. The
programmed soil models also have the option to use a constant Poisson ratio instead of the tangent bulk
modulus functions.

• Lastly, it is emphasized the Duncan and Duncan/Selig soil models are nonlinear elastic and behave the same in
loading or unloading. There is not a special unloading algorithm with different material characteristics. For the
record, it is noted that some investigators have incorporated ad hoc unloading algorithms into the Duncan soil
model. However such algorithms, which are not based on plasticity concepts, violate the continuity principal
of load path. Two load paths that are arbitrarily close to one another should not result in dramatically different
stress-strain response. Suffice to say, that a theoretically correct unloading model for the Duncan soil model is
yet to be developed.

3.5.6 Implementation of soil models and nonlinear solution strategy

Recall that Equation 3.5-1 is the fundamental plane-strain constitutive matrix relating stress increments to strain
increments from a known stress-strain state at load step i to the unknown stress-strain state at load step i +1.
Accordingly, the matrix coefficients, C11, C12 and C13 are chord moduli as notionally illustrated on the stress-strain
curve in the figure below along with the tangent moduli at the beginning and end of the load step.

3-20
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Figure 3.5.6-1 Chord and tangent moduli on a notional stress-strain curve.

Chord moduli representation. To convert the Duncan and Selig tangent Young’s modulus and tangent Bulk modulus
to chord moduli suitable for defining C11, C12 and C13, the following averaging technique is used,

E c = (1- r)E ti + rE ti+1 Equation 3.5-18

Bc = (1- r)Bti + rBti+1 Equation 3.5-19

where E c , Bc = chord moduli for Young’s modulus and bulk modulus, respectively
E ti , E ti+1 = tangent Young’s moduli at load step i and i+1, respectively
Bti , Bti+1 = tangent bulk moduli at load step i and i+1, respectively
r = averaging ratio usually taken as ½.

The justification for averaging the tangent moduli to obtain the chord moduli follows from the mean-value-theorem of
differential calculus, which implies that the process becomes exact as the size of the load step decreases. Generally the
averaging ratio is taken as r = ½. However as explained in the next paragraph, it is reasonable to set r =1when in situ
soil elements enter the structural system for the first time. For this reason, the averaging ratio is treated as a material
input parameter.

Entering soil elements. Soil elements enter the structural system in one of two categories. The first category applies to
pre-existing or in-situ soil elements in which an initial state of stress exists but is generally not known prior to the
solution. Elements entering in this category are typically part of the initial configuration and are assigned to the first
construction increment.

The second category applies to fill soi1 elements, i.e., soil layers added to the system in a predefined construction
schedule. Here, the initial stress state is non-existent prior to entry into the system. Both categories present special
starting problems for the iteration scheme because the initial stress state is unknown in the first category or nonexistent
in the second category.

3-21
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

For the first category, the initial stress state can be determined iteratively by assuming the pre-existing soil zone is a
construction increment loaded with its own body weight (and, if desired, a consolidation pressure). In this case the
averaging ratio should be set to 1.0, so that, Ec and Bc are equal to the tangent values at the end of the load step because
this corresponds to the existing stress state. When r = 1, the tangent moduli at the beginning of the load step have no
influence on the averaging process. After the first construction increment is complete, the algorithm automatically
changes r = 1/2 so that all subsequent moduli calculations represent chord values in going from a known stress state to
an unknown stress state.

For the second category, soil layers entering the system for the first time have zero initial stress and zero initial stiffness
at the beginning of the load step but gain stiffness at the end of the load step from self-weight and/or compaction load.
In this case, using r = 1/2 implies the effective chord moduli during the load step is equal to one-half the tangent moduli
at the end of the load step, which is considered to be a reasonable assumption. For all load steps following the initial
load step, there is no longer any ambiguity and we set r =1/2 because we know the stress state at the start of each
subsequent load step.

To start the iteration process for entering elements of either category, some initial guess must be made for the end-of-
load-step tangent moduli because the start-of-load step moduli are zero. Thus in order to construct the first trial
stiffness matrix, "dummy" principle stresses are used to get the first trial moduli value. The dummy principle stresses
have no effect on the final converged solution, but they do influence the number of iterations required to achieve
convergence. After an element has entered the system, the initial guess for end-of-load-step tangent moduli for all
subsequent load steps are equated to the last calculated values and the use of dummy principal stresses is not required.

Nonlinear solution summary. Based on the forgoing discussion, a summary on the nonlinear solution strategy is
provided in the flowchart shown in Figure 3.5.6-2. The flow chart depicts the iteration cycle to determine the
constitutive matrix of a soil element wherein the principal stresses from the last iteration are used make closer and
closer estimates of the chord moduli until convergence occurs. Convergence of the algorithm occurs when the chord
moduli obtained from two successive iterations differ by no more than 1% for all elements.

Two additional features of the algorithm not shown in the flow chart are listed below.

1. An under-relaxation scheme to improve the rate of convergence for the tangent moduli comes into play after
the second iteration wherein each estimate of the end-of-load-step tangent moduli is a weighted average of the
current estimate and the previous iteration estimate. This feature takes advantage of the observation that
convergence occurs in an oscillatory manner.

2. A constant Poisson ratio option may be chosen by the user that replaces the tangent bulk modulus formulation.
When the constant Poisson ratio option is exercised, the bulk modulus formulations are bypassed. Otherwise,
the algorithm is essentially the same.

3-22
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Figure 3.5.6-2 Duncan and Duncan/Selig flow chart for computing constitutive matrix.

3-23
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

3.5.7 Recommended Duncan and Duncan/Selig parameters for standard soils

Whenever possible, the Duncan and/or Duncan Selig soil model parameters should be determined directly from tri-
axial test equipment using established curve-fitting procedures. In many instances, however, tri-axial data may be
unavailable, and so, it is convenient to establish standard parameter values for various types of soil and degrees of
compaction.

Table 3.5.7-1 provides parameter values for the Duncan soil model for four soil classifications, each with three levels
of compaction. These standard values, extracted from References 19 and 20, are conservative in the sense that they are
the lower bounds of strength and moduli values observed from numerous tri-axial tests for each soil type. An
independent study at the University of Notre Dame utilizing the same database verified that the table parameters give a
good but conservative representation of the experimental data.

Table 3.5.7-1 Duncan soil model parameters (from Duncan Reference

Soil type and Young’s Tangent Modulus Parameters Bulk Parameters Density
compaction K n C φ0 Δφ Rf Kb m reference
(--) (--) (psi) (deg) (deg) (--) (--) (--) (lb/ft3)
CA105 600 0.40 0.0 42 9 0.7 175 0.2 150
CA95 300 0.40 0.0 36 5 0.7 75 0.2 140
CA90 200 0.40 0.0 33 3 0.7 50 0.2 135
SM100 600 0.25 0.0 36 8 0.7 450 0.0 135
SM90 300 0.25 0.0 32 4 0.7 250 0.0 125
SM85 150 0.25 0.0 30 2 0.7 150 0.0 120
SC100 400 0.60 0.5 33 0 0.7 200 0.5 135
SC90 150 0.60 0.3 33 0 0.7 75 0.5 125
SC85 100 0.60 0.2 33 0 0.7 50 0.5 120
CL100 150 0.45 0.4 30 0 0.7 140 0.2 135
CL90 90 0.45 0.2 30 0 0.7 80 0.2 125
CL85 60 0.45 0.1 30 0 0.7 50 0.2 120

In the above table the soil type is defined as follows: CA = Course Aggregates, SM = Silty Sand, SC = Silty-Clayey
Sand and CL = Silty Clay. The compaction number is percent relative compaction, per AASHTO T-99. As an example,
SM100 means silty sand compacted to 100% relative density per T-99.

In a similar manner Table 3.5.7-2 provides parameter values for the Duncan/Selig soil model for three soil
classifications, each with five levels of compaction. These values were computed by Selig in Reference 21 and are
generally conservative. The bulk modulus parameters were determined from hydrostatic tests which, according to
Selig, tend to be more conservative than those determined from standard tri-axial tests.

3-24
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Table 3.5.7-2 Duncan/Selig soil model parameters (from Selig Reference)

Soil type and Young’s Tangent Modulus Parameters Bulk Parameters Density
compaction K n C φ0 Δφ Rf Bi/ Pa εu reference
(--) (--) (psi) (deg) (deg) (--) (--) (--) (lb/ft3)
SW100 1300 0.90 0.0 54 15 0.65 108.8 0.01 148
SW95 950 0.60 0.0 48 8.0 0.70 74.8 0.02 145
SW90 640 0.43 0.0 42 4.0 0.75 40.8 0.05 140
SW85 450 0.35 0.0 38 2.0 0.80 12.7 0.08 130
SW80 320 0.35 0.0 36 1.0 0.90 6.1 0.11 120
ML95 440 0.40 4.0 34 0.0 0.95 48.3 0.06 135
ML90 200 0.26 3.5 32 0.0 0.89 18.4 0.10 130
ML85 110 0.25 3.0 30 0.0 0.85 9.5 0.14 122
ML80 75 0.25 2.5 28 0.0 0.80 5.1 0.19 115
ML50 16 0.95 0.0 23 0.0 0.55 1.3 0.43 66
CL95 120 0.45 9.0 15 4.0 1.00 21.2 0.13 130
CL90 75 0.54 7.0 17 7.0 0.94 10.2 0.17 125
CL85 50 0.60 6.0 18 8.0 0.90 5.2 0.21 120
CL80 35 0.66 5.0 19 8.5 0.87 3.5 0.25 112

In the above table the soil type is defined as follows: SW = Gravelly sand, ML = Sandy silt, and CL = Silty Clay. The
compaction number is percent relative compaction, per AASHTO T-99. As an example, SW95 means gravelly sand
compacted to 95% relative density per T-99.

In the old CANDE-1989 documentation, Selig presented less conservative values for the tangent bulk modulus
parameters by uniformly increasing the parameter Bi/Pa by a factor of 2.5 and uniformly decreasing εu by the factor
0.71. However, in recent years most researchers prefer the original hydrostatic data given in the above table.

3-25
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

3.6 Extended Hardin Soil Model

The extended Hardin soil model is a variable-modulus elasticity formulation using stress/strain-dependent functions for
the shear modulus and Poisson ratio. Basically, the model exhibits stiffening of constitutive moduli when confining
stress increases and softening when shear strain increases.

Hardin’s original work for the variable shear modulus formulation, which is based on a hyperbolic relationship between
shear stress and shear strain, is summarized in a 1973 Air Force technical report (Reference 22). In 1974 a variable
Poisson ratio function was developed to be the companion elasticity function so that together the two elasticity
functions, shear modulus and Poisson ratio, are referred to as the extended Hardin soil model. Reference 1 documents
the development of the variable Poisson function as well as validation of the shear modulus function against an
independent set of experimental data.

In many ways the extended Hardin model is similar in behavior to the Duncan and Duncan/Selig models, although the
later models are more popular and are supported by a larger database of soil parameters dependent on soil quality and
compaction. However, a unique feature of the extended Hardin model is that the soil parameters for the shear modulus
function are characterized in terms fundamental soil properties including void ratio, plasticity index and percent
saturation of the soil.

3.6.1 Plane-strain constitutive matrix

From an overall perspective, the extended Hardin soil model functions are used to define the nonlinear components of
an isotropic, elasticity-based constitutive matrix for plane-strain conditions as expressed below.

⎛ Δσ x ⎞ ⎛ C11 C12 0 ⎞ ⎛ Δε x ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Δσ y ⎟ = ⎜ C12 C11 0 ⎟ ⎜ Δε y ⎟ Equation 3.6-1
⎜ Δτ ⎟ ⎜ 0 0 C33 ⎟⎠ ⎜⎝ Δγ ⎟⎠
⎝ ⎠ ⎝

where, Δσ x , Δσ y , Δτ = stress increments for x-direction, y-direction and shear components, respectively
Δε x , Δε y , Δγ = strain increments for x-direction, y-direction and shear components, respectively
C11 , C12 , C33 = nonlinear coefficients dependent on shear modulus and Poisson ratio functions.

The table below shows the relationship between the constitutive matrix components and the shear modulus and Poisson
ratio functions that define the extended Hardin model.

Table 3.6.1-1 Components for constitutive matrix relationship to moduli functions

Components of G = shear modulus G = shear modulus

Constitutive matrix ν = Poisson ratio B = bulk modulus
(G, ν) (G, B)

C11 = 2G(1- ν) 4
B+ G
1- 2ν 3
C12 = 2G ν 2
B- G
1- 2ν 3
C33 = G G

3-26
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

The middle column in the above table defines the matrix coefficients C11, C12 and C13 in terms of shear modulus and
Poisson ratio, which are the elastic parameter functions directly developed for the extended Hardin soil model. The
third column defines the matrix coefficients in terms of shear modulus and bulk modulus, which is a more natural pair
of elasticity functions and often easier to implement. The equivalence between column 2 and 3 is given by the elasticity
relationship,

2(1+ ν)
B=G Equation 3.6-2
3(1- 2ν)

The motivation for choosing Poisson ratio as second elasticity function is because it easy to develop a smooth function
that insures Poisson ratio remains within the admissible bounds, 0 ≤ υ ≤ ½. In contrast, the constraints on an
independent bulk modulus function requires that B > 2/3 G, which is more difficult to achieve with a smooth function.

3.6.2 Hardin shear modulus development

Hardin’s formulation is based on experimental observations of soil behavior from standard tri-axial tests. A standard
tri-axial test is conducted by placing a cylindrical soil specimen in a pressure chamber and initially subjecting the
specimen to a uniform hydrostatic pressure, called σ3. Next a steadily increasing axial load is applied to the specimen
producing a net axial stress, called σ1, which includes the hydrostatic pressure. Note that σ3 and σ1 are principal stresses
in the lateral and axial directions, respectively. The difference in principal stresses σ1 – σ3 is equal to twice the
maximum shear stress occuring on 45-degree plane.

As the axial stress increases, axial strain ε1 is computed by measuring the axial shortening of the specimen divided by
the specimen length. Similarly, the lateral strain ε3 is measured at each load step so that the maximum shear stain is
equal to ε1 - ε3. Note that ε1 and ε3 do not include the initial hydrostatic strain and the algebraic signs must be strictly
observed in computing differences. The above concepts are summarized below.

1
τ= (σ1 - σ 3 ) Equation 3.6-3
2
γ = ε1 - ε 3 Equation 3.6-4

τ = G s γ (implied nonlinear relationship) Equation 3.6-5

where, τ = accumulated shear stress from tri-axial test

γ = accumulated shear strain from tri-axial test
G s = secant shear modulus, a function of stress and strain

The curve in Figure 3.6.2-1 is an idealized plot of shear stress versus shear strain for a typical tri-axial test. Here, Gmax
is the initial slope at zero shear strain and τmax is the maximum shear strain at failure. As discussed subsequently, Gmax
and τmax are dependent on the hydrostatic stress level.

3-27
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Figure 3.6.2-1 Idealized shear stress-strain curve for tri-axial test

Hardin hyperbolic function. Hardin’s fundamental insight is that for any tri-axial test, the shear stress-strain curve is
fairly well approximated by equating the shear stress to a hyperbolic function of shear strain as follows,

G max
τ= γ Equation 3.6-6
1+ γ h
γ α
γh = (1 + ) Equation 3.6-7
γr γ 0.4
exp( )
γr
γ r = G max /C1 Equation 3.6-8

where, G max = initial slope for shear modulus (dependent on hydrostatic stress)
γ h = hyperbolic strain function
γ r = reference shear stain
α = dimensionless soil parameter, related to soil type and percent saturation.
C1 = soil parameter with units of stress, related to void ratio, percent saturation, and plasticity index.

As shear strain increases to infinity, the shear stress in Equation 3.6-6 approaches an asymptotic limit representing the
maximum shear stress (or shear failure stress) given by,

τ max = G max γ r Equation 3.6-9

3-28
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Initial shear modulus. Based on a wide variety of soil tests as typified by the above figure, Hardin discovered that the
initial shear modulus increased in proportion with the square root of the initial tri-axial confining pressure as expressed
below.

G max = S1 σ m Equation 3.6-10

where, G max = initial shear modulus when shear strain = 0

S1 = soil parameter related to void ratio (units in psi )
σ m = 1/3(σ11 + σ 22 + σ33 ) = hydrostatic stress (units psi)

Equations 3.6-6 through 3.6-10 form the complete mathematical description of the Hardin secant shear modulus
function, which is characterized by three soil model parameters, C1 , S1 and α. The direct method of determining the
soil model parameters is to conduct a series of tri-axial tests on the particular soil being investigated. Alternatively,
Hardin has developed an indirect method for determining soil parameters based on fundamental soil properties. Both
the direct and indirect methods are presented below.

Direct method of soil parameter identification. Clearly the most accurate way to determine the shear modulus
parameters, C1 , S1 and α, is to perform a series of tri-axial tests on the soil under investigation. For each confining
pressure σ3, it is required to plot an experimental stress-strain curves similar to Figure 3.6.2-1 and then follow the steps
below.

1. Construct the initial tangent at zero shear stain and denote its value as Gmax. Based in Equation 3.6-10, the
parameter S1 may be computed as follows.

S1 = G max / σ3 Equation 3.6-11

Ideally each confining pressure would produce the same value for S1. However since the model is not perfect,
the final value for S1 should be the average value determined from all initial confining pressures.

2. Using the measured value of Gmax from step 1 along with the experimentally observed max shear stress τmax ,
the parameter C1 may be computed by combining Equations 3.6-8 and 3.6-9 as follows.

C1 = G max 2 /τ max Equation 3.6-12

In an ideal world, the computed value of C1 should be the same for all confining pressures. However since the
model is not perfect, the final value for C1 should be the average value determined from all initial confining
pressures.

3. The third and last soil parameter α controls the shape (concavity )of the shear stress-strain curve in Figure
3.6.2-1. Since all parameters of the secant shear modulus model are known except α, Equation 3.6-6 may be
applied at a single experimental data point (τ*, γ*) to provide an algebraic equation to solve for α. A natural
data point to select is midway in the stress range, τ* = ½ τmax along with the corresponding experimentally
measured shear strain γ*.

Indirect method of soil parameter identification. A particularly useful result of Hardin’s work is that he developed
an indirect method for determining soil parameters based on fundamental soil properties. This was accomplished by
correlating the results of many tri-axial tests for many types soil with basic soil properties including void ratio, percent
saturation, and plasticity index.

3-29
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Listed below are the expressions he developed to quantify the soil parameters C1 , S1 and α for three broad classes of
soil, granular, mixed and cohesive.

F2 R 2
C1 = (units psi) Equation 3.6-13
0.6 - 0.25(PI)0.6

S1 = 1230 F (units psi ) Equation 3.6-14

⎡3.2 for granular soil ⎤

α = ⎢ 2.54(1+0.02S) for mixed soil ⎥ (dimensionless) Equation 3.6-15
⎢ ⎥
⎢⎣1.12(1+0.02S) for cohesive soil ⎥⎦

(2.973 - e) 2
where, F= for all soil types
1+e

⎡1100 for granular soil ⎤

R= ⎢ ⎥
⎣1100 - 6.0 S for mixed and cohesive soil ⎦

and, e = void ratio (0 ≤ e ≤ 1)

S = percent saturation ( 0 ≤ S ≤ 100)
PI = plasticity index (0 ≤ PI ≤ 1)
Thus knowing the type of soil, void ratio, percent saturation, and plasticity index, Hardin’s secant shear modulus
function is completely characterized without the need of a tri-axial test.

3.6.3 Poisson ratio development

Poisson ratio may be back calculated from tri-axial test data providing that principal stresses and principal strains are
measured during the test. Reference 1 presents a study showing back-calculated values of Poisson ratio for a sequence
of tri-axial tests with increasing confining pressure. Two significant findings from this study are summarized below.

1. For each individual tri-axial test with a specified confining pressure, the measured Poisson ratio increases as
shear strain increases. Starting at a low value of approximately 0.01, the measured Poisson ratio
asymptotically approaches an upper limit, nominally equal to 0.49, as the shear strain approaches the shear
failure limit.

2. When confining pressure is increased, the starting and ending values of Poisson ratio remain the same as noted
above, however the rate of increase of Poisson ratio with shear strain increases as confining pressure increases.
Said another way, the measured value of Poisson ratio at a given level of shear strain is inversely related to the
confining pressure. The above two observations are illustrated in Figure 3.6.3-1.

3-30
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Figure 3.6.3-1 Illustration of measured Poisson ratio from tri-axial tests.

Poisson ratio function. A Poisson ratio function that simulates the above two observations is shown below, wherein γp
is a scaled measure of shear stain.

ν min + γ p ν max
ν= Equation 3.6-16
1 + γp

q
γp = γ Equation 3.6-17
γr

where, ν = Poisson ratio function.

γ p = scaled shear strain measure.
γ r = Hardin reference shear strain (= Gmax/C1 from Equation 3.6-8)

and, ν min = Poisson function parameter denoting minimum value (dimensionless constant)
ν max = Poisson function parameter denoting maximum value (dimensionless constant)
q = Poisson function parameter controlling curve shape (dimensionless constant)

3-31
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Upon inspecting the Poisson ratio function, it is clear that when shear strain is zero, ν = νmin, and when shear strain
becomes very large, ν = νmax. Thus, the Poisson ratio function satisfies the first observation noted above. The second
observation is achieved by defining γp with Hardin reference strain γr in the denominator as shown in Equation 3.16-7.
Since γr is proportional to Gmax and Gmax is an increasing function of confining pressure, γp is reduced for tests with
higher confining pressures so that the Poisson ratio function behaves as illustrated in the above.

It is important to realize that the predicted and observed variation in Poisson ratio as portrayed in Figure 3.6.3-1 is only
for the case of a standard tri-axial loading environment. When Equation 3.6-16 is used to predict the behavior of
Poisson ratio in a uniaxial-strain loading environment (K0-test), the Poisson ratio prediction remains nearly constant as
axial strain (or shear strain) increases. This is because the steadily increasing confining pressure increases the value of
Gmax at nearly the same rate as the increase in strain so that γp remains nearly constant in accordance with observed
behavior.

Parameter identification. The Poisson ratio function requires identifying values for three model parameters, νmin, νmax
and q. The study presented in Reference 1 measured Poisson ratio values from five tri-axial tests with confining
pressures ranging from 25 psi to 250 psi. The soil specimens were dry, dense sand, and the model parameters identified
as,
ν min = 0.10
ν max = 0.49
q = 0.26
Choosing values for νmin and νmax are straightforward observations from the experimental data, and the selection of the
shape parameter q is determined by standard curve fitting techniques. The study concludes that the Poisson ratio
function with the above parameter values is in excellent agreement with the measured Poisson ratio data points for all
five tri-axial tests.

Of course, it is always best to conduct tri-axial tests on new soils to identify the parameters of the Poisson ratio
function. However, experience has shown that νmin, νmax and q are not overly sensitive to the type of soil and soil
properties, thus if no other data is available the above parametric values are fairly reasonable for all soil types.

3.6.4 Summary of extended Hardin soil model functions

The extended Hardin soil model is composed of the secant shear modulus function and the Poisson ratio function both
of which are functions of maximum shear strain and hydrostatic pressure.

The secant shear modulus function is given by the following function and sub functions.

G max
Gs = = shear modulus function Equation 3.6-18
1+ γ h
γ α
γh = (1 + ) = hyperbolic strain function Equation 3.6-19
γr γ 0.4
exp( )
γr
γ r = G max /C1 = reference shear stain Equation 3.6-20

G max = S1 σ m = initial shear modulus when shear strain = 0 Equation 3.6-21

where, γ = maximum shear strain

σ m = 1/3(σ11 + σ 22 + σ33 ) = hydrostatic stress (units psi)

3-32
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

and, S1 = soil parameter related to void ratio (units in psi )

α = soil parameter, related to soil type and percent saturation (dimensionless)
C1 = soil parameter related to void ratio, percent saturation, and plasticity index (units psi)

Parameter identification for C1 , S1 and α may be accomplished directly with tri-axial tests as outlined in the discussion
associated with Equations 3.6-11 and 3.6-12. Alternatively the parameters may be quantified by the indirect method
outlined with the discussion associated with Equations 3.6-13, -14, and –15.

The Poisson ratio function is given by the following function and sub functions.

ν min + γ p ν max
ν= = Poisson ratio function Equation 3.6-22
1 + γp
q
γp = γ = scaled shear strain measure Equation 3.6-23
γr

where, γ r = Hardin reference shear strain (= Gmax/C1 from Equation 3.6-20)

and, ν min = Poisson function parameter denoting minimum value (dimensionless constant)
ν max = Poisson function parameter denoting maximum value (dimensionless constant)
q = Poisson function parameter controlling curve shape (dimensionless constant)

Parameter identification for νmin, νmax and q may be accomplished with tri-axial tests as outlined in the discussion in
Section 3.6.3. If no independent test data is available, the default values, νmin, = 0.1, νmax = 0.49, and q = 0.26 are
reasonable for all soil types.

The overall behavior of the extended Hardin soil model mimics the actual behavior of soil. The model exhibits a
stiffening behavior when confining pressure dominates shear strain, and conversely a softening behavior when shear
strain dominates confining pressure.

Lastly, it is emphasized the extended Hardin soil model is a nonlinear elasticity model and behaves the same in loading
or unloading. Like the Duncan and Duncan/Selig soil models, additional research needs to be done in order to
incorporate plasticity-like concepts into the model so that unloading behavior can be simulated without violating
energy theorems and continuity principles.

3.6.5 Implementation of soil models and nonlinear solution strategy

To implement the extended Hardin model into the plane-strain constitutive matrix given by Equation 3.6-1, it is more
convenient to use the bulk modulus function with the shear modulus function (G, B) rather than the Poisson ratio
function (G ,ν) as was indicated in Table 3.6.1-1. The bulk modulus function is related to the extended Hardin functions
by elasticity identity relationships as,

2(1+ ν)
Bs = G s Equation 3.6-24
3(1- 2ν)

where, Bs = secant bulk modulus function

G s = secant shear modulus function (Equation 3.6-18)
ν = Poisson ratio function (Equation 3.6-22)

3-33
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Letting σ1, σ2, σ3 be the total principal stresses and ε1, ε2 and ε3 the corresponding total principal stains, then, the secant
moduli (Gs, Bs) relate total stress quantities to total strain quantities as follows.

τ = Gs γ Equation 3.6-25
σ m = Bs φ Equation 3.6-26

where, τ = (σ1 -σ 3 )/2 = maximum shear stress

γ = ε1 - ε 3 = maximum shear strain
σ m = (σ1 + σ 2 + σ3 )/3 = mean stress or hydrostatic stress
φ = ε1 + ε 2 + ε 3 = volumetric strain

Chord moduli representation. To progress from load step i to load step i+1, the moduli must be expressed as
incremental chord moduli, which are related to secant moduli as illustrated in the figures below.

Figure 3.6.5-1 Chord modulus relationship to secant modulus for shear modulus.

3-34
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

Figure 3.6.5-2 Chord modulus relationship to secant modulus for bulk modulus.

The incremental relationships between load step i and i+1 as portrayed in the above figures are defined below.

Δτ = G c Δγ Equation 3.6-27

Δσ m = Bc Δφ Equation 3.6-28

G s ( i+1) γ i+1 - τi
Gc = Equation 3.6-29
γ i+1 - γ i

Bs ( i+1) φi+1 - σ m(i)

Bc = Equation 3.6-30
φi+1 - φi

where, G c = chord shear modulus from load step i to i+1

Bc = chord bulk modulus from load step i to i+1
Δτ = τi+1 - τi = shear stress increment
Δγ = γ i+1 - γ i = shear strain increment
Δσ m = σ m(i+1) - σ mi = mean stress increment
Δφ = φi+1 - φi = volumetric strain increment.

Nonlinear solution strategy. The solution strategy is based on a direct iterative approach, also called trial and error.

3-35
Chapter 3 – Soil Models CANDE-2007 Solution Methods and Formulations

We begin with the understanding that we have a converged solution at load step i so that we know the total stress and
total strain state, σi and εi , respectively. Our objective is to advance the solution to load step i+1 following the steps
below.

1. To obtain the first trial solution, assume the values for Gc and Bc remain the same as they were for load step i
for each individual element.

2. Construct the element stiffness matrix using Gc and Bc to define the current constitutive matrix for each
element (Equation 3.6-1)

3. Assemble the global stiffness matrix and load increment vector for step i+1. Solve the system for a trial
solution and recover current estimates for increments of stress and strain vectors. Update trial vectors for total
stress and strain as indicated below.

Δε = incremental strain vector determined from strain-displacement relationship.

Δσ = incremental stress vector determined from current constitutive matrix

εi+1 = εi + Δε
σi+1 = σi + Δσ

4. Based on the known stress-strain state at step i and the estimated stress-strain state at step i+1, compute new
estimates for Gc and Bc .

G s ( i+1) γ i+1 - τi
Gc = (see Equation 3.6-18)
γ i+1 - γ i

Bs ( i+1) φi+1 - σ m(i)

Bc = (see Equations 3.6-18,-22,-24)
φi+1 - φi

5. If Gc and Bc computed in step 4 are sufficiently close to the previous estimates, say within 1% difference, then
the load step has converged, the solution is saved, and control shifts back to step 1 to advance to the next load
step. Otherwise, the iterative process continues wherein the incremental solution is discarded and control shifts
back to step 2 to repeat the load increment using the new values for Gc and Bc to get a new trial solution.

Typically the above algorithm converges for all elements in three to ten iterations for each load step, depending on the
scope of the problem. Convergence means that static equilibrium and the constitutive matrix are simultaneously
satisfied.

3-36
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

4 INTERFACE ELEMENT
This chapter presents the development of a simple friction-contact interface element that simulates frictional slippage,
separation, and re-bonding of two bodies initially mating at a common interface and subsequently deforming under an
arbitrary static loading schedule. Constraint equations between initially mating pairs of nodes and the general principle
of virtual work are used to formulate the interface element for a finite element solution methodology. The practical
application of the interface elements is to mimic the behavior of the interface between a culvert and backfill soil by
simulating frictional slippage of the soil over the pipe wall during the backfilling operations. Similarly, the model is
useful for simulating the slippage of backfill soil along trench walls for steep walled trench installations. The model is
also useful for enforcing a pinned connection between pipe element groups.

4.1 Introduction
CANDE’s interface element dates back to its original development in 1976, which is documented in Reference 1 and
published in the open literature in Reference 24. Today, the interface element remains as originally developed with a
few minor modifications to enhance convergence. Engineering literature contains a variety of interface element
formulations; however, they can generally be classified as either a stiffness approach (e.g. directionally stiff elements)
or a constraint approach typically formulated with the use of Lagrange multipliers. The constraint approach is used in
this development, however rather than using Lagrange multipliers, the development is based on a novel extension of
virtual work, which in itself became a contribution to FEM technology. Motivations for adopting the constraint
approach include, avoidance of numerical round-off problems inherent in the stiffness approach, direct control of
interface forces and relative movements, accuracy, and ease of FEM implementation.

The interface element is capable of responding to a general step-by-step loading history, such that tensile separation,
frictional sliding, or complete bonding (or re-bonding) is possible during any load step. Two nodes, one on either side
of an interface, defined by the interface angle, are used to establish the interface element along with a third “dummy
node” to represent interface forces. By selectively applying constraint equations to the interface nodes in an
incremental-iterative solution procedure, the behavior of a contact-friction interface is simulated.

Two computational advantages are inherent in the interface element. First, unlike the traditional Lagrange-multiplier
method that requires a minimization principle, the virtual work formulation is not restricted to conservative systems.
Secondly, the formulation results in an element 'constraint matrix' and element 'load vector' analogous to a typical
element stiffness matrix and load vector. Thus, the interface element can be assembled into the global equations by
standard techniques.

4-1
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

4.2 Virtual Work and Constraint Equations

In Chapter 1, the traditional virtual work formulation is presented without constraints resulting in the classical,
stiffness-based finite equations expressed in a scalar virtual work statement as,

δuˆ T (KΔuˆ - ΔP) = 0 Equation 4.2-1

where, Δuˆ = incremental unknown nodal displacements

K = current global stiffness matrix (generally nonlinear)
ΔP = incremental load vector.
Since δuT is an arbitrary virtual displacement, the satisfaction of Equation 4.2-1 requires that KΔu = ΔP, usually solved
iteratively to find displacements.

If Equation 4.2-1 represented a finite element model of two separate bodies initially in contact at node pairs as
suggested in Figure 4.2-1, then the bodies can deform independently resulting in overlapping and/or separation along
the interface. At the other extreme, if the node pairs at the interface are constrained to move together, a completely
bonded response is obtained. Between these two extremes, a slipping response can be obtained by constraining only the
displacements normal to the interface at each node pair.

Figure 4.2-1 Idealization of two bodies with initially mating node pairs along interface

Body B

Body A
Node
pairs

4-2
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

4.2.1 Constraint equations general form

To achieve the above, the immediate objective is to incorporate an arbitrary set of constraint equations into a global
virtual work statement. Formally, any set of algebraic nodal point constraint equations can be expressed as:

CΔuˆ - Δaˆ = 0ˆ Equation 4.2-2

where, C = constraint coefficient matrix

Δuˆ = subset of nodal degrees of freedom being constrained
Δaˆ = vector of specified constants (for example, displacement gaps)
0̂ = vector of zeroes, one zero for each constraint equation.
To fix ideas by means of a simple example, suppose it is desired to constrain one node in body A to be glued (fixed) to
the partner node in body B with no displacement gap between the node pair. In this case, Equation 4.2-2 would be
written in expanded notation as,

⎛ u Ax ⎞
⎜ ⎟
⎛ -1 0 1 0 ⎞ ⎜ u Ay ⎟ ⎛ 0 ⎞ ⎛ 0 ⎞
⎜ ⎟ -⎜ ⎟=⎜ ⎟
⎝ 0 -1 0 1 ⎠ ⎜ u Bx ⎟ ⎝ 0 ⎠ ⎝ 0 ⎠
⎜⎜ ⎟⎟
⎝ u By ⎠
In the above example we are simply imposing the two constraint equations, uAx = uBx and uAy = uBy, which forces the x
and y displacement component of the node in body A to be equal to those of the paired node in body B.

Associated with each constraint equation, there exists an unknown internal constraint force λ enforcing the constraint.
For convenience, Equation 4.4-2 can be expressed in a scalar (work) form as:

δλˆ T (CΔuˆ - Δa)

ˆ =0 Equation 4.2-3

where, λ̂ = vector of interface constraint forces, one force component for each constraint equation.

Since δλˆ is an arbitrary variation of the constraint force vector, Equation 4.2-3 is satisfied if and only if the constraint
equations are satisfied.

As yet the virtual work of constraint forces has not been established or defined. To this end, we argue that the internal
constraint forces produce internal virtual work when the constraint is given a small virtual movement (variation) as
follows.

ˆ T Δλˆ = δuˆ T CT Δλˆ

VWconstraint = δ(CΔuˆ - Δa) Equation 4.2-4

In the above definition for the incremental virtual work of constraints, we assume the system is equilibrium at load step
i and moves to a new equilibrium position at load step i+1 consistent with the incremental constraint equations, which
in turn generate incremental constraint forces, Δλ. When the constraints are given a small virtual disturbance
(variation), the constraint forces perform virtual work as they move with the virtual disturbance. Physically, this is
analogous to imposing constraints with stiff springs between node pairs wherein the internal spring force corresponds
to the constraint force. However, in this case the constraint forces are primary unknowns.

4-3
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

4.2.2 Global virtual work statement with constraints

With the above understandings, we complete the formulation by adding the internal virtual work of constraints
(Equation 4.2-4) to the standard virtual work statement in Equation 4.2-1 and append the constraint requirement in
Equation 4.2-3 to get the general virtual work statement including constraints.

T
⎛ δuˆ ⎞ ⎛ ⎛ K CT ⎞ ⎛ Δuˆ ⎞ ⎛ ΔP ⎞ ⎞
⎜⎜ ˆ ⎟⎟ ⎜⎜ - ⎟= 0
⎜ ⎝ C 0 ⎟⎠ ⎜⎜ Δλˆ ⎟⎟ ⎜⎝ Δaˆ ⎟⎠ ⎟
Equation 4.2-5
⎝ δλ ⎠ ⎝ ⎝ ⎠ ⎠

Since the virtual displacements and virtual constraint forces are arbitrary, the virtual work statement leads to the
following global system of equations.

⎛ K CT ⎞ ⎛ Δû ⎞ ⎛ ΔP ⎞
⎜ ⎟ ⎜⎜ ⎟⎟ = ⎜ ⎟ Equation 4.2-6
⎝ C 0 ⎠ ⎝ Δλˆ ⎠ ⎝ Δâ ⎠

The coupled set of global matrix equations is to be solved for nodal displacements and constraint forces. Note that the
separate partitioning of the global system into Δu and Δλ is a mere formality and not required in actual global
assembly.

4.2.3 General element constraint matrix and load vector

Recalling that the global stiffness equations are formed from the summation element stiffness matrices and load vectors
based on the element’s contributions to total virtual work, we can in the same manner construct an element constraint
matrix and load vector based on the constraint element’s contribution to the virtual work as shown below.

T
⎛ δuˆ e ⎞ ⎛⎛ 0 Ce T ⎞ ⎛ Δuˆ e ⎞ ⎛ 0ˆ ⎞ ⎞
δVcontstraint = ⎜ ⎜⎜ ⎟⎜ ⎟-⎜ ⎟⎟
⎜ δλˆ ⎟⎟
Equation 4.2-7
element
⎝ e⎠
⎜ ⎝ Ce
⎝ 0 ⎠ ⎜⎝ Δλˆ e ⎟⎠ ⎜⎝ Δaˆ e ⎟⎠ ⎟⎠

Since the above expression represents the contribution of a general constraint element to the global virtual work
statement given by Equation 4.2-5, the element-level constraint matrix and load vector are established as follows.

⎛0 Ce T ⎞
C* = ⎜ ⎟ Equation 4.2-8
⎝ Ce 0 ⎠

⎛ 0̂ ⎞
f̂ = ⎜
⎜ Δâ ⎟⎟
Equation 4.2-9
⎝ e⎠

where, C* = symmetric element constraint matrix, dimension (n+m) x (n+m)

Ce = constraint coefficient matrix for element, dimension m x n
f̂ = incremental load vector of zeros and displacement gaps, dimension n+m
Δaˆ = specified incremental displacement gaps, dimension m
m = number of constraint equations associated with element (m = 2 for interface element)
n = number of constrained degrees of freedom within element (n= 4 for interface element)
Just as the global stiffness matrix is assembled by a summation of element stiffness matrices k*, the global constraint
matrix is assembled by a summation of element constraint matrices C*. Similarly, the global constraint load vector is
assembled by a summation of element load vectors f̂ .

4-4
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

4.3 Interface Element Matrix and Load Vector

To specialize the foregoing to friction-contact interfaces, we assume that any interface can be modeled as a sequence of
node pairs (Figure 4.2-1), so that attention can be focused on a single node pair that forms an interface element.

4.3.1 Definition of interface element

The interface element is composed two nodes on either side of a common interface plus an interior node (a stand alone
node K) that represents interface forces. Figure 4.3.1-1 shows an exploded view of the interface element.

Figure 4.3.1-1 Interface element exploded view defining normal and tangential directions

Here, the interface element is defined in the local n, s coordinates, rotated φ degrees from the global x, y coordinate
system. Node I belongs to the body on one side of the interface and Node J belongs to the body on the other side of the
interface. Node I and J are initially assigned to the same x, y coordinate location prior to loading. The positive normal
direction is in the direction of travel from node I to node J. Nodal displacement components in the n-s system are
marked with a prime to distinguish them from the unprimed components in the x-y system.

Physical properties assigned to the interface element are listed below.

μ = coefficient of friction (Coulomb type)

β = tensile rupture force (typically = 0)
φ = angle of the interface normal with respect to x-axis

4-5
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

The coefficient of friction and tensile rupture force are discussed subsequently, however it should be evident that if
these properties are assigned very large numbers the interface will remain bonded (fixed) regardless of the interface
angle and internal forces.

4.3.2 Incremental and total responses (notation)

As with all developments in this manual, the interface element is an incremental relationship from load step i to load
step i+1. Letting the symbol q represent any interface response, we adopt the following notation for incremental
relationships.

q i+1 = q i + Δq Equation 4.3-1

where, qi+1 = total response at load step i+1

q i = total response at load step i
Δq = incremental response from step i to i+1

The following list defines interface response notation represented by the symbol q

u′I = displacement of node I in the normal direction.

u ′J = displacement of node J in the normal direction
v′I = displacement of node I in the tangent direction.
v′J = displacement of node J in the tangent direction.
λ n = normal force between node I and J
λs = tangent force between node I and J
u I = displacement of node I in the global x direction.
u J = displacement of node J in the global x direction
vI = displacement of node I in the global y direction.
v J = displacement of node J in the global y direction.
d n = interface gap distance in normal direction
ds = interface gap distance in tangent direction

Interface gap distances are illustrated in the figure below for load step i wherein the two bodies have separated at nodes
I and J producing normal and tangential gaps between the nodes.

4-6
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

Figure 4.3.1-1 Interface element in separated state at load step i.

The total normal gap and the total tangential gap at load step i are determined by the difference of nodal displacement
components in the n-s coordinates as given below.

d n i = u ′J i - u ′I i Equation 4.3-2a

d s i = v′J i - v′I i Equation 4.3-2b

4.3.3 Element constraint matrices and load vectors for three interface states.
We now identify three 'interface states' that may be imposed during a load step depending on the current responses of
the interface element. The three states are called fixed, slip, and free. In each case we will define the element constraint
matrix and load vector.

Fixed State. The fixed state means that nodes I and J are constrained to move during the load step such that specified
incremental gaps will occur in the normal and tangent direction (usually the incremental gaps are specified to be zero).
In the local n-s coordinates system the fixed state is expressed with the following two constrain equations.

-Δu′I + Δu′J = Δa Equation 4.3-3

-Δv′I + Δv′J = Δb Equation 4.3-4

where, Δa = specified normal gap increment (often zero)

Δb = specified tangential gap increment (often zero)
Since the global equations require that displacement components be expressed in x-y coordinates, the primed
displacement components are transformed to x-y displacement components by standard coordinate transformation
techniques so that the above constraint equations become,

4-7
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

⎛ Δu I ⎞
⎜ ⎟
⎛ -cosφ -sinφ cosφ sinφ ⎞ ⎜ Δv I ⎟ ⎛ Δa ⎞
⎜ ⎟ =⎜ ⎟ Equation 4.3-5
⎝ sinφ -cosφ -sinφ cosφ ⎠ ⎜ Δu J ⎟ ⎝ Δb ⎠
⎜ ⎟
⎝ Δv J ⎠

Here the 2x4 matrix is the constraint coefficient matrix Ce that is imbedded in the symmetric element constraint matrix
C* defined by Equation 4.2-8. The right hand side is the specified incremental gap vector for the element load vector
defined in Equation 4.2-9. If Δa and Δb are both specified zero, then the normal gap and tangential gap do not change
during the load step, i.e. both nodes experience identical displacement increments. If, on the other hand, we wish to
return the nodes to their initial position (no gaps), we specify Δa and Δb equal to the total displacement gaps defined in
Equations 4-3-2a,b.

The fixed state is summarized in the top portion of Table 4.3.3-1, which explicitly defines the element constraint matrix
C* and the element load vector. These may be assembled in the global system just like element stiffness matrix.

Table 4.3.3-1 Element constraint matrix C* (6x6) and load vector f̂ (6x1) for three interface states.
f̂
State Δu I Δv I Δu J Δv J Δλ n Δλ s (load)
0 0 0 0 -c s 0
0 0 0 0 -s -c 0
0 0 0 0 c -s 0
Fixed 0 0 0 0 s c 0
-c -s c s 0 0 Δa
s -c -s c 0 0 Δb

0 0 0 0 -c 0 -sΔT
0 0 0 0 -s 0 cΔT
0 0 0 0 c 0 sΔT
Slip 0 0 0 0 s 0 -cΔT
-c -s c s 0 0 Δa
0 0 0 0 0 1 ΔT

0 0 0 0 0 0 cΔN - sΔT
0 0 0 0 0 0 sΔN + cΔT
0 0 0 0 0 0 -cΔN + sΔT
Free 0 0 0 0 0 0 -sΔN - cΔT
0 0 0 0 1 0 ΔN
0 0 0 0 0 1 ΔT

c = cosφ, and s = sinφ

Slip State. The slipping state is defined by constraining the normal displacement increments of nodes I and J such that
a specified incremental gap will occur in the normal direction during the load increment. The tangential displacement
constraint is suppressed and replaced with a specification for the incremental interface force in the tangent direction.
Specifically, the slip state is expressed with the following constraint equation and force equation in the n-s system.

-Δu′I + Δu′J = Δa Equation 4.3-6

Δλ s = ΔT Equation 4.3-7

4-8
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

where, Δa = specified normal gap increment (often zero)

ΔT = specified increment of tangential force between nodes I and J
Since other types of elements share nodes I and J, the n-s displacement components must be transformed to the
common x-y system for global assembly and solution. Thus the above constraint equation is transformed to x-y system
as shown below.

⎛ Δu I ⎞
⎜ ⎟
T Δv I
( -cosφ -sinφ cosφ sinφ ) ⎜ ⎟⎟ = Δa
⎜ Equation 4.3-8
Δu J
⎜ ⎟
⎝ Δv J ⎠

With regard to the force equation (Equation 4.3-7), it may be retained in the n-s coordinate system because the dummy
node K containing the interface forces is not shared with any other element.
Whenever we suppress a constraint and specify Δλs = ΔT, we must also assign a positive ΔT force to node I and a
negative ΔT force to node J thereby effecting the interface force on the nodes. Also we must transform these forces to
x-y system for proper assembly in the global load vector as shown in the load vector column in Table 4.3.3-1.

The slipped state is summarized in the central portion of Table 4.3.3-1. Here the C* matrix contains one constraint
equation and force equation (λs = T), wherein the matrix size remains 6 x 6 for computational convenience. As
discussed above, the load vector includes force components at nodes I and J from the specified interface force, ΔT.

To simulate a friction-contact interface, Coulomb-type friction is assumed so that the maximum possible interface
tangent force in advancing from load step i to i+1 is a function of the normal compressive force and given by,

Fi+1 = μ λ n i+1 sgn(λ s i+1 ) Equation 4.3-9

where, Fi+1 = maximum Coulomb frictional force at step i+1 (signed)

μ = coefficient of friction
i+1
⎧⎪+1, for λ s i+1 > 0 ⎫⎪
sgn(λ s )= ⎨ i+1 ⎬
⎩⎪ -1, for λ s < 0 ⎭⎪
With the above understanding, a frictional slipping interface requires that the specified incremental tangent force from
load step i to i+1 satisfy the equilibrium condition as written below.

ΔT = Fi+1 - λ s i Equation 4.3-10

i
where, λ s = known tangential interface force at load step i.

Of course, ΔT is recomputed by iterating within the load step until two successive values of Fi+1 are acceptably close,
say within 1%.

Free State. The free state is defined by suppressing both constraint equations and specifying the normal and tangent
interface force increments as shown below.

Δλ n = ΔN Equation 4.3-11

Δλ s = ΔT Equation 4.3-12

4-9
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

where, ΔN = specified increment of normal force between nodes I and J

ΔT = specified increment of tangential force between nodes I and J
Again, the above force equations may be expressed in the n-s coordinate system because the dummy node K containing
the interface forces is not shared with any other element. In addition, the specified interface forces ΔN and ΔT must be
assigned to nodes I and J, positive for node I and negative for node J. Also these forces must be transformed to x-y
system for proper assembly in the global load vector.

The free state is summarized in the bottom portion of Table 4.3.3-1. Here the C* matrix is all zeros except C* (5,5) =
C* (6,6) = 1 that enforces the equations Δλn = ΔN and Δλs = ΔT. Again, the matrix size remains 6 x 6 for
computational convenience. Lastly, the load vector includes x-y force components at nodes I and J from the specified
interface forces.

4-10
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

4.4 Nonlinear Solution Strategy for Interface

The problem is posed as follows. We assume we know the interface solution at end of load step i, which includes
converged solutions for the following accumulated interface responses.

λ n i , λ s i = total interface normal and tangent interface forces at end of load step i.
d n i , d s i = total interface normal and tangent displacement gaps at end of load step i.

We seek to determine an incremental solution ( Δλn, Δλs, Δdn, Δds) to achieve a valid solution for the total interface
responses at the end of load step i+1 computed as follows.

λ n i+1 = λ n i + Δλ n Equation 4.4-1

i+1 i
λs = λ s + Δλ s Equation 4.4-2
i+1 i
dn = d n + Δd n Equation 4.4-3
d s i+1 = d s i + Δd s Equation 4.4-4

By a valid solution we mean the physical compatibility and equilibrium conditions of the contact-friction interface are
satisfied. In general the valid solution must be determined iteratively wherein a particular state (fixed, slip, or free) is
assumed and solved to obtain a trial solution. The trial solution is used to determine if the assumed state is correct, and
if not, what state is more likely to be correct. At the same time, the trial solution is used to estimate new parameters for
the trial load vector associated with trial interface state as listed below.

• Fixed state load parameters = Δa, Δb

• Slip state load parameters = Δa, ΔT
• Free state load parameters = ΔN, ΔT

Recall that Δa and Δb are specified incremental displacement gaps in normal and tangential directions, respectively,
and ΔN and ΔT are specified incremental interface forces in normal and tangential directions, respectively.

4.4.1 Selecting a new trial interface state

Table II offers an exhaustive set of physical criteria to test the validity of an assumed state. The table may be viewed as
a 3 x 3 decision matrix wherein the previously assumed state forms rows and the new candidate states form columns.

Table 4.4.1-1 Decision matrix for selecting new trial interface state during iteration cycle from load step i to i+1

New trial state → Fix Slip Free

↑ ↑ ↑
Previous trial state ↓

(1) Fix → λ n i+1 < β and λ s i+1 < Fi+1 λ n i+1 < β and λ s i+1 > Fi+1 λ n i+1 > β

(2) Slip → λ n i+1 < β and Δd s Fi+1 < 0 λ n i+1 < β and Δd s Fi+1 > 0 λ n i+1 > β

(3) Free → d n i+1 < 0 -- d n i+1 > 0

4-11
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

To illustrate the decision matrix, if the previously assumed state was fixed, then row 1of the decision matrix is entered
from the left. The fist query determines if the fixed state is correct by checking if the current normal interface force is
compressive (or less than the tensile rupture force β) and the magnitude of the tangential force is less than the current
frictional resistance calculated from Equation 4.3-9. If both of these inequalities are satisfied than the fixed state is
correct, and the fixed state is assumed for the next iteration. Otherwise, we proceed to query the slip conditions to see if
the tangential force exceeds the frictional resistance while the interface normal force is compressive. If both of these
conditions are satisfied, then we select the slip sate for the next iteration. The only remaining possibility is that the
normal force exceeds the tensile resistance, implying the interface is prone to separate so that we choose the free state
for the next iteration.

If the previous state was slip, we enter row 2 of the decision matrix to determine the most likely state for the next
iteration. The first query determines whether or not we should shift to a fixed state based on the conditions that the
normal force remains compressive, but there is a change in the direction of relative slip determined by the sign of the
product of the slip increment, Δds, and the signed value of frictional resistance from Equation 4.3-9. This somewhat
subtle argument is based on the observation that relative slip movement cannot reverse its direction until the passive
frictional force totally reverses its direction. Therefore, a fixed state for the next iteration is assumed. On the other
hand, if the product of the slip increment and the frictional resistance is positive (that is, the have the same sign), then
we assume the slip state remains correct for the next iteration. The only remaining possibility is that the normal force
exceeds the tensile resistance implying the interface is prone to separate so that we choose the free state for the next
iteration.

Lastly if the previous state was free, we enter row 3 to determine whether or not to remain in the free state or shift to a
fixed state for the next iteration. We shift to a fixed state if the current normal gap is less than zero meaning that the
two bodies have incorrectly penetrated each other at this point. Otherwise, we remain in the free state for the next
iteration. Note that row 3 does not provide a direct path from a free state to a slip state. This does not mean a slip state
cannot be reached from a free state; it simply means a slip state must be reached by the iterative path, free to fix to slip.

Recapping, the decision matrix provides the choice for the next assumed state, thereby dictating the form of the element
constraint matrix C* defined in Table 4.3.3-1. Before we can obtain the next trial solution, we must determine the
constraint load parameters as discussed next.

4.4.2 Computing load vector parameters for next iteration

Given that we know the interface state (fixed, slip or free) from the converged solution at load step i and the new trial
interface state from the above decision matrix, then the trial parameters for the new load vector may be determined by
referring to Table 4.4.2-1. That is, go directly to the unique row/column position in the table where row = known state
at step i, and column = current trial state for step i+1.

Table 4.4.2-1 Specified incremental parameter values for trial constraint load vector

New trial state Fix Slip Free

for load step i+1 → (a,b) (a,T) (N,T)

Load step i ↓
Δa = 0 Δa = 0 ΔN = -λ n i
Fix Δb = 0 ΔT = Fi+1 - λ s i ΔT = -λ s i
Δa = 0 Δa = 0 ΔN = -λ n i
Slip Δb = 0 ΔT = Fi+1 - λ s i ΔT = -λ s i
Δa = -d n i Δa = -d n i ΔN = 0
Free ΔT = 0
Δb = Δd s d n i /Δd n ΔT = Fi+1

4-12
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

For example, if a slip state existed at the end of load step i and the next iteration for step i+1 is to be assumed free, then
we locate slip/free table position and set ΔN and ΔT equal to the negative value of the interface forces existing at the
end of load step i. As a consequence, the trial solution will result in net zero values for the interface forces, which must
be the case for a free interface state.

Recall that the specification for ΔT under the slip column has already been addressed in the development of Equation
4.3-10. The remaining entries in Table 4.4.2-1 are generally self-explanatory except, perhaps, for the free/fix table
position. Here the implication is that the node pair began in a free state at the end load step i, however interface
penetration is observed in the decision matrix so that we are assuming a fixed state (re-bonding) for the next iteration.
To this end, the normal gap is closed by specifying Δa equal to the negative of existing normal gap, which is
physically correct, resulting in zero normal gap at the end of load step i+1. With regard to specifying the slip gap
increment, we observe that the ratio existing-gap/incremental gap must be less than 1, otherwise penetration would not
have been observed. Thus, it is reasonable to assume that the slip gap increment can be specified by reducing the
incremental slip gap in the same proportion, which explains the logic for the Δb specification.

4.4.3 Algorithm summary and convergence

The following step-by-step algorithm summarizes the solution strategy and convergence criteria. As always it is
assumed that a valid solution exists for a load step i ,and we seek an incremental solution to advance the solution to
load step i+1.

1. At the beginning of each new load step assume each interface element remains in the same state (fixed, slip or
free) as determined by the converged solution of the previous load step. Therefore, the element constraint
matrix is initially known (Table4.3.3-1) and the constraint element load vector is set to zero for the first
iteration. (For the very first load step, all constraints elements are initially assumed to be in the fixed state)

2. Assemble the global system matrix (see Equation 4.2-6) and solve the system for incremental displacements
and incremental interface forces. Form a temporary trial solution for load step i+1by summing the interface
responses as shown in Equations 4.4-1 to 4.4-4.

3. Using the decision matrix in Table 4.4.1-1 evaluate each constraint element to see if it changed state with
respect to the previous iteration. Assign new constraint matrices to all elements that changed state and keep a
record of how elements changed state.

4. Using Table 4.4.2-1 compute a new load vector for each element. For each constraint element currently in a
slipping state, compare the current value of frictional resistance (Fi+1 in Equation 4.3-9) with the value
determined in the previous iteration and record the number of elements that have relative differences higher
than 1%.

5. Check for convergence.

• Convergence is achieved if all constraint elements remain in the same state for two successive
iterations, and the frictional resistance of all slipping elements is within 1 % tolerance on successive
iterations. When this occurs the load step is converged and we proceed to step 6.

• Otherwise if convergence is not achieved, we through away the trial solution, and return to step 2 for
another iteration.

6. Update the solution by permanently adding the incremental solution to the previous converged solution and
return to step 1 to solve for the next load step until all load steps have been processed.

In closing this chapter, the reader is referred References 1 and 24 for practical applications of the using constraint
elements to successful predict the performance real-world culvert installations.

4-13
Chapter 4 – Interface Element CANDE-2007 Solution Methods and Formulations

(This page intentionally left blank)

4-14
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5 LARGE DEFORMATIONS AND BUCKLING

This chapter presents the development of beam theory with large deformations that includes beam bending, column
deformation, material nonlinearity, and incremental loading. The development is an extension on the small-deformation
beam theory presented in Section 2.1 but removes the limitations of small deformation theory.

5.1 Updated Lagrange Formulation

The major assumptions and limitations are listed below:

1. Two-dimensional framework in a plane strain formulation.

2. Bernoulli-Euler beam kinematics without shear deformation.
3. Material nonlinearity is a function of normal stress and strain.
4. Incremental virtual work formulation with incremental stress-strain relationships and loading.
5. Updated Lagrange strain-displacement relationship appropriate for large rotations and small stretch.

5.1.1 Coordinates and incremental relationships.

Consider a beam element originally in an undeformed state that has experienced a sequence of load steps (1, 2, 3, …i)
and is in equilibrium after load step i. Let the nodes at either end of the element be labeled “a” and “b”, and let their
current location in the global X-Y coordinate system be denoted as (Xai, Yai) and (Xbi, Ybi), respectively.

Next, construct a local x-y coordinate system with the origin at node a. The x-axis extends from node a to node b along
the base of the beam, and the y-axis is orthogonal to the x-axis with y = 0 at the bottom of the beam cross section. From
this reference position we apply load step i + 1 and seek to determine the incremental structural responses. The local
displacement increments in displacing from load step i to step i + 1 will be denoted as Δu in the x direction and Δv in
the y direction.

Equilibrium of all the forces and moments acting on the element at load step i + 1 will be in reference to the x-y
coordinate system of the previous load step i, not the element’s original position as assumed in small deformation
theory.

5.1.2 Bernoulli-Euler beam kinematics.

Based on the assumption that cross-sectional planes remain plane in bending and axial deformation, the well-known
Bernoulli-Euler assumption for displacement increments Δu and Δv at any station x along the beam length and at any
point y in the beam’s cross section may be expressed as:

Δu(x, y) = Δa(x) + (y*-y)Δv’ Equation 5-1

Δv(x) = Δv(x) Æ a function of x only Equation 5-

2

where, Δu(x,y) = displacement in x direction from column and bending deformation

Δa(x) = uniform displacement in x direction from column, independent of y
Δv(x) = displacement in y direction as function of x, independent of y
Δv’ = d(Δv)/dx, derivative of Δv with respect to x (local deformation slope)
y* = a reference plane in the beam cross section at y = y*, yet to be specified
( )’ = d( )/dx, prime symbol denotes derivative of any quantity with respect to x

5.1.3 Total Lagrangian strain for large rotations.

The total normal strain (x direction) at load step i at any station x along the beam length and at any point y in the
beam’s cross section may be described by a truncated form of total Lagrangian strain given by,

5-1
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Ε i(x,y) = ui’ + ½(vi’)2 Equation 5-3a

All quantities are in reference to the original undeformed position, i = 0. It is evident that ui’ is the standard definition of
small strain in the x direction. The nonlinear term, ½(vi’)2, is a second order correction term that adds a uniform
stretching strain to the axial strain as a result of beam rotation. The correction term may be visualized as the stretch that
is induced in the leg of a right triangle as it is rotated into the position of the hypotenuse through an angle approximated
by vi’. The larger the value of vi’ the more inaccurate is the approximation of the angle.

If we express an increment of the total Lagrangian strain from step i to i+1 we obtain:

ΔΕ(x,y) = Ε i+1(x,y) - Ε i(x,y) = Δu’ + ½( Δv’)2 + vi’Δv’ Equation 5-3b

The last term in Equation 5-3b contains vi’, which is not an incremental term but rather the approximated angle at step i
in reference to the undeformed position at i = 0. Thus the incremental form of total Lagrangian strain, ΔΕ (x,y),
produces the same error as Equation 5-3a when dealing with large rotations. Note that when determining the strain
increment for the first load increment starting from i = 0, we have vi’ = v0’ = 0, because the reference configuration has
no displacements or rotations.

5.1.4 Updated Lagrangian strain increments.

The so-called updated Lagrange formulation provides improved accuracy for calculating strains by updating the x and y
coordinates of the beam element’s position to step i prior to solving for step i + 1. In other words the beam element’s
position at step i is taken as the reference position for equilibrium and for measuring additional displacements resulting
from the loads that are imposed at step i + 1, which means vi’ = 0 for each load step from i to i + 1.

With the above understanding, we can rewrite Equation B-3 in the updated Lagrangian description as:

Δε (x,y) = Δu’ + ½( Δv’)2 Equation 5-4a

where the symbol Δε is used to represent updated Lagrangian strain increments in order to distinguish it from
increments of total Lagrangian strain, ΔΕ.

Inserting the Bernoulli-Euler kinematical equations for Δu into the updated Lagrangian strain increment equation, we
arrive at the incremental strain-displacement relationships for beams under large rotations, given as,

Δε(x,y) = Δa’+ (y*-y) Δv’’+ ½(Δv’)2 Equation 5-4b

Equation 5-4b is the new equation for this development. That is, the remainder of the basic equations for the nonlinear
stress-strain model, internal force resultants, and virtual work remain valid as expressed in the small-deformation
formulation. However, the interaction of these basic equations with Equation 5-4b results in new nonlinear governing
equations for beam deformations.

5.1.5 Incremental stress-strain model.

The nonlinear stress-strain laws used in CANDE are specific to each pipe material such as corrugated metal and
reinforced concrete, which are detailed in previous chapters. However, all these stress-strain models conform to the
same generic form that may be expressed as

Δσ = Ec Δε Equation 5-5

where, Δσ = increment of axial stress from load step i to i + 1, (Δσ = Δσ (x,y))

Δε = increment of axial strain from load step i to i + 1, (Δε = Equation 5-4b)
Ec = chord modulus of total stress-strain curve from load step i to i + 1

5-2
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

The chord modulus is dependent on the pipe material and the history of stress and strain throughout all loading steps. It
is determined iteratively during each load step by repeating the solution process until the value of Ec converges for each
element to a small tolerance of error.

Replacing Δε in Equation 5-5 with Equation 5-4b, the fundamental stress-displacement relationship may be expressed
as,

Δσ = Ec(Δa’+ (y*-y) Δv’’+ ½(Δv’)2) Equation 5-6

Next, we will use Equation 5-6 to define the internal thrust and internal moment acting on the cross section at any
station x.

5.1.6 Internal thrust and moment increments.

As is customary, we define the thrust increment as the integral of Δσ over the cross section area A, and the moment
increment as the integral of Δσ (y*-y) over the cross section area A. Specifically,

ΔN = ∫A Δσ dA = ∫A Ec(Δa’+ (y*-y) Δv’’+ ½(Δv’)2) dA Equation 5-7a

ΔM = ∫A Δσ (y*-y) dA = ∫A Ec(Δa’+ (y*-y) Δv’’+ ½(Δv’)2) (y*-y) dA Equation 5-7b

where, ΔN = thrust increment from load step i to i + 1, (ΔN = ΔN (x))

ΔM = moment increment from load step i to i + 1, (ΔM = ΔM (x))
dA = b(y)dy, a differential area of cross section where b(y) is the width with y measured from bottom.

The arbitrary reference position y* is now chosen so that ∫A Ec (y*-y) dA = 0, which simplifies Equations 5-7a,b, and
gives the location of y* measured from the bottom as,

y* = ( ∫A Ec y dA) / (∫A Ec dA) Equation 5-8

The remaining area integrals in Equation 5-7a,b relate to effective axial stiffness and effective bending stiffness and are
defined as,

EA* = ∫A Ec dA = Effective axial stiffness of beam element Equation 5-9

EI* = ∫A Ec (y*-y)2dA = Effective bending stiffness of beam element Equation 5-10

All of the integrals in Equations 5-8, 5-9, and 5-10 are computed numerically for every iteration within each load step
until convergence is achieved.

With the above definitions for y*, EA* and EI*, the thrust and moment increments (Equations 5-7a,b) simplify to the
following equations,

ΔN = EA*(Δa’ + ½(Δv’)2) Equation 5-11

ΔM = EI* Δv’’ Equation 5-12

Note that the moment-increment expression is the familiar bending stiffness-curvature relationship used in small
deformation theory. However, the thrust-increment expression, in addition to the familiar column stiffness-deformation
relationship, also contains a nonlinear stiffness term related to rotation, ½(Δv’)2.

5-3
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5.1.7 Virtual work for beam-column element.

The general concept of virtual work is fully developed in Chapters 1 and 2. Here the attention is focused on the internal
virtual of beam-column elements.

The internal virtual work for the beam-column element under consideration may be expressed as the following volume
integration at any load step i + 1:

δUi+1 = ∫V δε i+1σ i+1 dV Equation 5-13

where, δεi+1 = virtual strain field arising from the imposed virtual displacements
σi+1 = stress field existing in beam element at load step i +1
∫V (…) dV = volume integration over the beam element

The virtual symbol δ means a small variation of the strain field. Similar to variational calculus, δ operates on the
displacement variables like a total derivative. Thus operating on the updated Lagrangian strain, the virtual strain field is
determined as:

δεi+1 = δ[a’ + (y*-y) v’’ + ½(v’)2]i+1 = δa’ + (y*-y)δv’’ + v’i+1δv’ Equation 5-14

The displacement variables prefaced with the symbol δ are arbitrary motions and need not be associated with the values
at load step i+1. However, the term v’i+1 is associated with load step i+1 and is equivalent to, v’i+1 = Δv’, since v’i = 0 by
virtue of the updated Lagrange formulation.

The incremental form of the element internal virtual work is defined by;

δΔUe = δUi+1 - δUi = ∫V (δε i+1σ i+1 - δε iσ i )dV Equation 5-15

Observing that δε i+1= δε i + Δv’δv’, the above becomes;

δΔUe = ∫V {(δa’+(y*-y)δv’’)Δσ + Δv’δv’σ i+1 }dV Equation 5-16

Separating the volume integral into area and length integrals, dV = dAdx, we arrive at,

δΔUe = ∫x{δa’ ∫AΔσ dA + δv’Δv’ ∫Aσi+1 dA + δv’’ ∫AΔσ (y*-y)dA} dx Equation 5-17

Here it is observed that the two area integrals involving Δσ have already been identified as ΔN and ΔM in Equations 5-
7a,b. The area integral involving σi+1 will now be designated as Ni+1, which is the total thrust at step i+1 referred to the
beam element’s coordinate system at step i. Thus Equation 5-17 is written as;

δΔUe = ∫x{ δa’ΔN + δv’ Δv’ Ni+1 + δv’’ΔM } dx Equation 5-18

Since we have already established expressions for ΔN and ΔM in terms of displacements in Equations 5-11 and 5-12,
we replace the unknown force increments ΔN and ΔM in Equation 5-18 with the equivalent unknown displacements.
However with regard to Ni+1, which is nonlinear because it is coupled with Δv’, we choose to leave Ni+1 as expressed to
facilitate the nonlinear solution strategy. With the above replacements, Equation 5-18 becomes,

δΔUe = ∫x{δa’EA*Δa’ + δa’EA* ½(Δv’)2 + δv’Δv’Ni+1 + δv’’EI*Δv’’} dx Equation 5-19

For the sake of clarity, we will express Equation 5-19 as the sum of four terms,

δΔUe = δΔUa + δΔUr + δΔUg + δΔUb Equation 5-20

where,

δΔUa = ∫x{δa’EA*Δa’} dx Equation 5-20a

5-4
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

δΔUr = ∫x{δa’EA* ½(Δv’)2}dx Equation 5-20b

δΔUg = ∫x{δv’Δv’ Ni+1 }dx Equation 5-20c

δΔUb = ∫x{δv’’EI*Δv’’}dx Equation 5-20d

Of the four terms in Equation 5- 20, the first term is the familiar column stiffness term and the fourth term is familiar
bending stiffness term identical to small deformation theory. The middle two terms are nonlinear and are associated
with large deformations. As we shall see in the subsequent FEM development, the second term will be treated as a load
vector and the third term will lead to the so-called geometric stiffness matrix.

5-5
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5.2 Finite Element Development

5.2.1 Finite element interpolation functions.

Equation 5-20 is the incremental internal virtual work of a beam column element with large deformation without
numerical approximations. To express this in a FEM formulation we introduce interpolation functions for the
displacements that become exact as the element lengths become small. The interpolation functions are expressed in
terms of unknown nodal variables shown in the sketch below.

y
va vb
θa ua ub θb
x ( )
L = length
Node b
Node a
where, Δua, Δub = incremental nodal displacements in x direction at nodes a and b
Δva, Δvb = incremental nodal displacements in y direction at nodes a and b
Δθa, Δθb = incremental nodal rotations in counter-clockwise direction at nodes
a and b

The incremental axial displacement function is approximated with a linear interpolation function expressed in vector
notation as;

Δa(x) = < φ1 φ2 > < Δua Δub>T Equation 5-21

where, φ1(x) = 1 – x/L

φ2(x) = x/L
L = element length

The incremental vertical displacement function is approximated by a cubic polynomial, known as a Hermetian
interpolation function, and is expressed in vector notation by,

Δv(x) = < γ1 γ2 γ3 γ4 > < Δva Δθa Δvb Δθb>T Equation 5-22

where, γ1(x) = 1 – 3 (x/L)2 + 2 (x/L)3

γ2(x) = L (1 – x/L)2 x/L
γ3(x) = 3 (x/L)2 – 2 (x/L)3
γ4(x) = L (1 – x/L) (x/L) 2

5.2.2 Element matrices and vectors

Taking the necessary derivatives of the approximating functions for Δa(x) and Δv(x), inserting them into the four
internal virtual work terms of Equation 5-20, and integrating with respect to x over the element length L, we arrive at
the following evaluations.

For the internal virtual work of the axial stiffness term we have,

δΔUa = < δΔua δΔub> Ka < Δua Δub> T

where the axial stiffness matrix is defined as,

5-6
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Ka = (EA*) ∫x [< φ’1 φ’2 >T < φ’1 φ’2 >]dx Equation 5-23a

and upon evaluating the integral, the axial stiffness matrix is determined as,

Ka = (EA*/L) 1 -1 Equation 5-23b

-1 1

For the internal virtual work of the bending stiffness term we have,

δΔUb = < δΔva δΔθa δΔvb δΔθb> Kb < Δva Δθa Δvb Δθb>T

where the standard bending stiffness matrix is defined as,

Kb = (EI*) ∫x [< γ’’1 γ’’2 γ’’3 γ’’4 >T < γ’’1 γ’’2 γ’’3 γ’’4 >]dx Equation 5-24a

and upon evaluating the integral, the standard bending stiffness matrix is determined as,

12/L2 6/L -12/L2 6/L

Kb = (EI*/L) 6/L 4 -6/L 2 Equation 5-24b
-12/L2 -6/L 12/L2 -6/L
6/L 2 -6/L 4

For the internal virtual work of the geometric stiffness term we have,

δΔUg = < δΔva δΔθa δΔvb δΔθb> Kg < Δva Δθa Δvb Δθb>T

where the geometric stiffness matrix is defined as,

Kg = (Ni+1) ∫x [< γ’1 γ’2 γ’3 γ’4 >T < γ’1 γ’2 γ’3 γ’4 >]dx Equation 5-25a

and upon evaluating the integral, the geometric stiffness matrix, which is dependent on the current thrust scalar is
determined as,
6/5L 1/10 -6/5L 1/10
Kg = (Ni+1) 1/10 2L/15 -1/10 -L/30 Equation 5-25b
-6/5L -1/10 6/5L -1/10
1/10 -L/30 -1/10 2L/15

For the remaining internal virtual work term (rotational stretch vector), we have

δΔUr = < δΔua δΔub> re

wherein the rotational stretch vector is expressed as,

re = ρ (EA*/2L) <-1 1>T Equation 5-26a

where, ρ = ∫x (Δv’)2dx = < Δva Δθa Δvb Δθb> Kg* < Δva Δθa Δvb Δθb>T

Kg* = Kg without the multiplier (Ni+1)

Thus, the rotational stretch vector, which is dependent on the nonlinear scalar ρ, is determined as,

re = ra = ρ (EA*/2L) -1 Equation 5-26b

rb 1

5-7
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

The above four terms of the element’s internal virtual work, δΔUe, may be brought together into a single 6 x 6 matrix
along with 6-component vectors where the six nodal unknowns are listed in the following order < Δua Δva Δθa Δub Δvb
Δθb>. Thus the finite element equivalent of Equation 5-20 is expressed as,

δΔUe = < δ Δua δΔva δΔθa δΔub δΔvb δΔθb> ( Ke < Δua Δva Δθa Δub Δvb Δθb> T + re )

where, Ke = total effective element stiffness matrix (6 x 6) Equation 5-27

re = < ra 0 0 rb 0 0 >T = rotational stretch vector (1 x 6) Equation 5-28

Of course, it is understood that the components of Ka, Kb, and Kg are added into the proper locations of the Ke matrix
corresponding to the ordering of nodal variables.

5.2.3 Transformation and global assembly.

The last step before adding the element’s contribution of δΔUe into the entire system is to transform the nodal variables
from local coordinates to global coordinates.

Let β = angle from global X-axis to local x-axis, then the local nodal variables may be expressed as global nodal
variables by,

Δue = T ΔuE Equation 5-29

where, Δue = < Δua Δva Δθa Δub Δvb Δθb> T local nodal variables for element

ΔuE = < ΔuA ΔvA ΔθA ΔuB ΔvB ΔθB> T global nodal variables for element

C S 0 0 0 0
T = -S C 0 0 0 0
0 0 1 0 0 0
0 0 0 C S 0
0 0 0 -S C 0
0 0 0 0 0 1

and, C = cos(β) and S = sin(β)

With the above transformation matrix, the element stiffness matrix and rotational stretch vector may be expressed in
global coordinates as,

KE = TT Ke T Equation 5-30

r E = TT r e Equation 5-31

It should be clear that lower-case subscripts are used for element quantities that are in expressed in local coordinates,
whereas upper-case subscripts are used for element quantities that are in expressed in global coordinates.

The effevtive element stiffness matrix KE and the rotational stretch vector rE are in the proper form for global assembly
wherein KE is added to the global effective stiffness matrix and rE is brought to the right-hand-side of the equations to
be subtracted from the load vector.

5-8
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5.3 Solution Strategy

The overall solution process is a direct iterative method as outlined below. The viewpoint is that we have a converged
solution at load step i, and we seek a solution for load step i+1.

5.3.1 Iterative methodology

In obtaining a solution from load step i to i+1, we begin by assuming Ni+1 = Ni and ρ = 0 so that element matrices and
vectors may be computed for each element. After global assembly, the current set of equations may be expressed for
iteration k as

KGk ΔuGk = ΔPGk Equation 5-32

where, KGk = global stiffness matrix for iteration k, i.e., collection of elements ∑ KEk

ΔuGk = global vector of all unkown nodal variiable increments for iteration k

ΔPGk = global vector of load increments for iteration k, including negative rotational-stretch vector.

k = iteration counter; 1, 2, 3, …

The set of linear alebraic equations represented by Equation 5-32 is solved by Gauss Elimination for the dispacement
increments ΔuGk. This trial solution is used to form a new and better estimates for Ni+1 and ρ, which leads to revised
values for KGk and ΔPGk for the next iteration. When two successive iterations produce the same results within an
acceptable tolerance, the solution has coverged for load step i + 1, and all quantities are updateded as,

q i+1 = q i + Δq Equation 5-33

where q stands for all structural responses such as displacements, stresses, strains, moments, thrusts, and so on. In
addition, the updated Lagrange formulation implies that the global coordinate locations of each node are updated so
that q also repesents each (X,Y) nodal coordinate that are updated by the corresponding displacement increments.

5.3.2 Recovery of element forces

When load step i+1 converges to a solution, we want to recover the incremental internal forces at the ends of the
element, that is, increments of thrust, shear and moment at the element’s nodes. We can obtain these force increments
in reference to “configuration i coordinates” by making use of virtual work.

Let, ΔTa, ΔTb = Thrust force increments (xi direction) at nodes a and b

ΔQa, ΔQb = Shear force increments (yi direction) at nodes a and b

ΔMa, ΔMb = Moment increments at nodes a and b (invariant to x-y coordinates)

The incremental virtual external work of these end forces is given by;

δΔWe = < δ Δua δΔva δΔθa δΔub δΔvb δΔθb> < ΔTa ΔQa ΔMa ΔTb ΔQb ΔMb>T

With the element’s incremental end loads defined, we can now express the net incremental virtual work of any element
to be equal to zero; i.e.,

δΔVe = δΔUe - δΔWe = 0

Using equation Equations 5-27 and 5-28 to express δΔUe, we arrive at the following equations
to determine the incremental end forces and moments:

5-9
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

⎛ ΔTa ⎞ ⎛ ⎞ ⎛ Δu a ⎞ ⎛ ra ⎞
⎜ ΔQ ⎟ ⎜ ⎟ ⎜ Δv ⎟ ⎜ 0 ⎟
⎜ a⎟ ⎜ ⎟⎜ a ⎟ ⎜ ⎟
⎜ ΔM a ⎟ ⎜ K a + Kb + K g ⎟ ⎜ Δθ a ⎟ ⎜ 0 ⎟
⎜ ⎟=⎜ ⎟⎜ ⎟+⎜ ⎟ Equation 5-34
⎜ ΔTb ⎟ ⎜ ⎟ ⎜ Δu b ⎟ ⎜ rb ⎟
⎜ ΔQ b ⎟ ⎜ ⎟ ⎜ Δv b ⎟ ⎜ 0 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟
⎝ ΔM b ⎠ ⎝ ⎠ ⎝ Δθ b ⎠ ⎝ 0 ⎠

Thus the end forces can be recovered by multiplying the complete element stiffness matrix times the incremental
displacements. For the end thrust forces we also add on the rotational stretch forces ra and rb . All of the above
operations are carried out in reference to local the coordinate axis at step i.

5.3.3 Update coordinates

Before proceeding to the next load step, it is neccessary to update all quatities to the new starting configuation. In
particular we need to update the thrust associated with the geometric stiffness matrix. Let N* represent the
accumulation of all past thrust increments which is expressed in the x-direction of the previous step (i,e., N* = Ni+1).
Similary let Q* represent the accumulation of all past shear increments which is expressed in the y-direction of the
previous step The goal is to express in N* in the x-direction and Q* in the y-direction of the current position. The
increment of rotation, say α, that the element experienced in moving from the previous step to the current step may be
computed as:

Lnew = (Lold+ Δub - Δua)2 + (vb –va)2)1/2 (new length of element) Equation 5-35a

sinα = (vb –va)/ Lnew Equation 5-35b

cosα = (Lold + Δub - Δua)/ Lnew Equation 5-35c

Thus we can rotate the thrust and shear to the current configuration as

N*α = cosα N* + sinα Q* Equation 5-36a

Q*α = -sinα N* + cosα Q* Equation 5-36b

For the iteration process during the next load step, we use the following estimate for thrust in the geometric stiffness
matrix;

Ni+1 = N*α + ΔN Equation 5-37

where N*α is a fixed value and ΔN is determined iteratively. For the first iteration we assume ΔN = 0. In the absence of
material nonlinearity, this process always converges in three load steps.

After all load steps are complete, the problem is finished and the load-deformation path of the structural system may be
visualized. Note that this method of analysis by itself does not directly provide a critical buckling load like a linearized
buckling (Euler) analysis. Rather, the actual buckling load is observed by noting the peak load (e.g., soil height) when
the system fails due to unbounded displacements. In order to predict the buckling capacity at the end of each load
increment, we augment the large-deformation analysis with a linearized buckling solution as discussed next.

5-10
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5.4 Buckling Capacity

At the end of each converged load step, we have stored in memory the current components for the standard stiffness
matrix and the geometric stiffness matrix, which, when taken together, form the global stiffness matrix that is expressed
in Equation 5-32. Thus after convergence of load step i+1, we may write the global stiffness matrix in two parts as,

KG = KS + KGeo(Ni+1) Equation 5-38

where, KS = global standard stiffness matrix (small deformations)

KGeo(Ni+1) = global geometric stiffnes matrix written as a function of element thrust values

The known values for thrust Ni+1 as expressed in the above geometric stiffness matrix, have been determined iteratively
during the solution for step i + 1. Thus, Ni+1 represents the current set of thrust values in the various beam-column
elements at load step i+1.

If we temporarily assumed the matrices in Equation 5-38 remain constant except the geometric stiffness is scaled by a
load parameter “c”, then we can determine what magnitude increase in the set of thrust values Ni+1 that would cause
unlimited deformations, i.e., critical buckling. This will occur when the combined standard and geometric stiffness
becomes singular (determinant = 0), which is dependent on the value of c.

To this end, let the global stiffness matrix be written as,

KG(c) = KS + c KGeo(Ni+1) Equation 5-39

and pose the problem as, find the value(s) c such that,

Det KG(c) = 0 Equation 5-40

This, of course, is the classic eigenvalue problem for buckling bifurcation wherein the lowest eigenvalue “c” is a direct
measure of the current buckling capacity. If, for example, the value of c is found to be 2.50, then the current set of
thrust values Ni+1 could be increased by any factor up to 2.5 before buckling would occur. Thus we can say that the
value of “c” is the predicted buckling capacity or equivalently the safety factor against buckling.

In CANDE-2007, a determinant search process finds the value of “c”. Starting with the guess c = 1, the determinant
expressed in Equation 5-40 is evaluated. If the determinant is positive, meaning the matrix is structurally stable, then
the value of c is steadily increased until the determinant is found to be negative, meaning the matrix structurally
unstable. Knowing the correct value c lies in an interval bounded by the last two estimates, a binary-chop method is
used to keep refining the estimate for “c” until it is accurate to two decimal places.

In summary, this chapter presents a complete derivation of a beam theory with large deformations that includes beam
bending, column deformation, material nonlinearity, and incremental loading. The large deformation responses (forces,
displacements, stresses and strain) are output at the end of each load step along with a prediction of the buckling
capacity.

5-11
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5.5 Illustration – Simply Supported Beam

This illustration compares the updated Lagrange CANDE-2007 solution and buckling prediction with a closed-form
solution for an axially load, simply supported beam with initial constant end moment. The closed-form solution is
theoretically equivalent to a one-step updated Lagrange solution (i.e., total Lagrange method) so that a direct
comparison may be made if we limit the CANDE solutions to a single load step. Subsequently, we will obtain CANDE
solutions with multiple load-steps to demonstrate a more realistic load-deformation behavior in contrast with the single-
step solutions.

Thus, the specific objectives are:

1. Develop a closed form solution for a simply supported beam with an initial end moment and an increasing
axial load based on total Lagrange formulation. Note a closed form Updated Lagrange solution is not tractable.

2. Compare CANDE load-deformation path and buckling prediction with the closed-form solution wherein the
CANDE solution is obtained with a series of single load steps with increasing axial load (i.e., a series of total
Lagrange solutions).

3. Show realistic load-deformation paths from CANDE solutions based on multiple steps to successively increase
the axial load while updating the coordinates at the end of each load step (i.e., an updated Lagrange solution).

5.5.1 Development of closed-form solution

The idealized beam, shown in the figure below, is a flexible in bending with bending stiffness EI and inextensible in
axial elongation. The right end is pinned supported, and the left end is on a roller with a constant moment Mo and an
axially increasing thrust force, Pi .

Figure 5.5.1-1 Simply supported beam with increasing axial load and fixed end moment.

Mo

V+dV
111 O
V L M+dM Δ
N N+dN
M
v v+dv
x dx

A differential segment of the test beam is shown above for formulating the governing differential equation,
where, M, dM = moment and differential
V, dV = shear and differential
N, dN = thrust and differential
v, dv = vertical displacement and differential
x = horizontal coordinate
(.)’ = d(.)/dx denotes derivative with respect to x

Horizontal and vertical equilibrium of the differential segment requires:

dN ≡ N’dx = 0 (Thrust is constant) Equation 5-41

5-12
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

dV ≡ V’dx = 0 (Shear is constant) Equation 5-42

Moment equilibrium of the differential segment is approximated by letting v’ represent the tangent of the total rotation
so that the differential equation remains linear as expressed below:

M’ + Nv’ + V = 0 Equation 5-43

The Bernoulli-Euler moment-curvature relationship, wherein EI is the beam bending stiffness, is given by;

M = -EI v’’ Equation 5-44

Taking the first derivative of Equation 3 and the second derivative of Equation 4 and combining the results we arrive at
(recalling that N’ = 0 and V’ = 0):

-EI v’’’’ + N v’’ = 0 Equation 5-45

Since N is constant, the boundary condition in Figure 1 dictates that N = - P, where P is the applied axial force, positive
in compression, and we define the positive quantity q as,

q2 = P/EI Equation 5-46

Using the above in Equation 5-45, we arrive at the governing ordinary differential equation written as,

v’’’’ + q2v’’ = 0 Equation 5-47

The general solution to Equation 5-47 is given by,

v(x) = C1 + C2x + C3 cosqx + C4sinqx Equation 5-48

Where C1, C2, C3, and C4 are constants determined from the four boundary conditions and written as,

v(0) = 0 Equation 5-49a

v’’(0) = -M0/EI Equation 5-49b
v(L) = 0 Equation 5-49c
v”(L) = 0 Equation 5-49d

After determining the constants, the final closed-form solution may be expressed as,

v(x) = (M0/P){ -1.0 + x/L + cosqx - (cosqL/sinqL)sinqx} Equation 5-50

Upon inspecting the last term in Equation 5-50, it is evident that the predicted vertical displacement becomes infinite
when the devisor sinqL = 0. This occurs when the load P reaches the first critical value, Pcritical , such that qL = π. Or,
making use of Equation 5-46, we arrive at the well-known critical buckling load for a simply supported beam,

Pcritical = EI(π /L)2 Equation 5-51

The major approximation in the above derivation occurs in Equation 5-43 wherein the tangent of the rotation angle
formed by the differential element with respect to the undeformed position is approximated by the derivative of the
vertical displacement. This approximation is exactly equivalent to the total Lagrangian approach (truncated from)
presented earlier in this chapter. Of course, this approximation becomes less and less accurate as the angle increases.

5.5.2 Example test problem for closed-form solution

To numerically illustrate the load-deformation behavior of Equation 5-50 as well as the critical buckling load given by
Equation 5-51, hypothetical beam parameters are defined in the table below.

5-13
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Table 5.5.2-1 Numerical values for test problem.

Beam Parameter Parameter value
Total Length, L 12 inches
Bending Stiffness, EI 1000 lbs-in2
Initial end moment, M0 1 in-lb
Axial load, P (variable) 0 to Pcritical
Pcritical (Eq 5-51) 68.539 lbs

The following figure shows a plot of the applied load P versus mid-span deflection where the applied load is presented
as percentage of the critical load (68.54 lbs)

Figure 5.5.2-1 Load-deformation curve from closed-form solution of test problem.

LOAD vesus Deflection

Closed form

120%
Applied load % of Critical .

100%

80%

60%

40%

20%

0%
0.00 0.50 1.00 1.50 2.00
Deflection at mid span, inches

Deflections are observed to remain small until the applied load is near 80% of the critical load. After the 80% load
level, the simple-theory solution predicts rapidly increasing displacements that become infinite as the load reaches
100% critical. The numerical values of displacements are less and less accurate as the displacements (and hence the
rotations) increase in magnitude. The totally false prediction of an infinite displacement as the load approaches 100%
of critical is a characteristic inaccuracy of this total Lagrangian formulation, which is limited to small rotations on the
order of 5%. Another characteristic inaccuracy of total Lagrange is that for loads greater then 100% of critical is that
predicted displacements are opposite in sign, which of course, is also obliviously unrealistic.
Although the total Lagrange formulation does not give realistic predictions for the displacement values for loads near
the critical load, the prediction of the critical load level where rapid changes in displacement occur is quite accurate.
This finding, which is demonstrated in the next section, is significant because it gives credence to the buckling load
prediction method used in CANDE for load schedules that are significantly below the critical load.

5.5.3 CANDE model of test problem

The corresponding CANDE model is shown in the figure below with parameters defined in the following table. Twelve
beam-column elements are used to simulate the test problem. The same beam properties previously defined for the

5-14
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

closed-form solution are assigned to the CANDE model except the CANDE model also includes axial deformation,
which is not included in the closed-form formulation. In order to minimize the effect of axial deformation, the cross-
sectional area is assigned a value two orders of magnitude larger than the moment of inertia so that bending
deformation will dominate.

Figure 5.5.3-1 Twelve element CANDE model of test problem.

P
Mo = 1 1

L = 12 in

Table 5.5.3-1 Numerical values for CANDE model.

Beam Parameter Parameter value
Total Length, L 12 inches
Young’s modulus, E 1000 psi
Moment of inertia, I 1.0 in4
Cross-section area, A 100.0 in2
Initial end moment, M0 1 in-lb
Axial load, P (variable) 0 to Pcritical
Pcritical (Eq 5-51) 68.539 lbs

5.5.4 CANDE simulating total Lagrange approach.

Although CANDE is developed for an updated Lagrange methodology, it is easy to simulate a total Lagrange solution
by specifying only one load step so that the original configuration remains the reference configuration and coordinates
are not updated. By solving a series of problems (each with one load step) the axial load P can be successively
increased from zero to Pcritical, while the initial end moment is held constant for each problem.

As shown in the table below, the predicted deflections from this series of CANDE solutions are found to be nearly
identical to the closed form solution with agreement to three significant digits. The lower part of the table shows
CANDE’s prediction for the critical buckling load at each axial load level.

Table 5.5.4-1a. Mid-span Deflections (inches) for increasing axial load % Critical Load
Axial load as
percent of Critical 0% 29.18% 58.36% 87.54% 94.84% 99.21% 99.94%
Closed-form 0.0090 0.0128 0.0220 0.0742 0.1796 1.1803 16.357
Deflections

CANDE
(single-step) 0.0090 0.0128 0.0220 0.0742 0.1795 1.1800 16.170
Deflections

Table 5.5.4-1b. CANDE prediction of Buckling load (lbs) at each axial load level
Predicted Buckling
Load N/A 68.6 68.4 68.4 68.3 68.8 69.2

The rather remarkable agreement of deflection predictions and buckling load predictions suggests that the CANDE
formulation for the geometric stiffness matrix, the rotational stretch vector and the nonlinear solution strategy is

5-15
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

working properly. Typically CANDE underwent three solution iterations per load level to achieve convergence.
Buckling predictions typically required 8 determinant searches per problem to find the buckling load factor. The error
tolerance on the determinant search algorithm is set at 2%.

5.5.5 CANDE Solution for updated Lagrange approach.

In this more accurate approach, the CANDE solution is obtained by running a single problem wherein the axial load is
increased on each load step with P starting at 0.0 and steadily increasing in value up to and well beyond Pcritical .

The only difference between the previous total Lagrange methodology and the updated Lagrange methodology is that
the beam’s reference position is updated to the deformed position of the previous load step and the existing internal
forces (thrust and shear) are transformed to be aligned with the deformed position.

Figure 5.5.5-1 shows load-deformation curves for mid-span deflection for the total and updated Lagrange methods.
Recall that the total Lagrange solution as predicted by closed-form solution or by CANDE give nearly identical results.

Figure 5.5.5-1. Comparison of load-deformation curves from updated and total Lagrange.

Load versus Deflection

160%
140%
Applied load % of Critical .

120%
100% CANDE Updated
Lagrange
80%
CANDE Total
60% Lagrange
40%
20%
0%
0.00 1.00 2.00 3.00 4.00 5.00 6.00
Deflection at mid span, inches

Some pertinent observations from the above figure are noted below:
1. The size of the load increments for the updated Lagrange solution are indicated by the points shown in the
figure. Smaller steps are required in the vicinity of the critical load where large displacements occur.
2. For axial loads below 80% of the Pcritical, the deflections as predicted by total Lagrange are close but not
identical to Updated Lagrange, differing at most by 4% at the 80% load level
3. When the axial load approaches Pcritical, both the updated and total Lagrange solutions exhibit buckling-like
behavior by showing rapid increases in displacement for small increases load.
4. The large displacements predicted by the total Lagrange solution are not realistic for loads close to Pcritical and
incorrectly infers that there is no stable equilibrium position as the deflections become infinite. In contrast the
updated Lagrange solution predicts realistic displacement and finds a stable position at about 110% of Pcritical..

5-16
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5. Although not shown, the buckling load predictions for updated and total Lagrange methods are very close for
all load levels below Pcritical.. The updated Lagrange prediction for the buckling load when the axial load is very
near Pcritical is about 2% higher than that predicted from the total Lagrange solution.

In summary for the simply supported beam example, the CANDE solution matches the closed form solution when
restricted to one-step and loading. Under multi-step loading CANDE solutions are physically meaningful at load levels
near and above the critical buckling load whereas the closed-form solution is not. In all cases, the prediction of
buckling capacity is accurate.

5-17
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5.6 Illustration -- Soil-Structure Interaction

This illustration examines the effects of large deformations and buckling capacity in the presence of other nonlinearities
for a corrugated steel culvert installation. The primary purpose is to compare and contrast small deformation solutions
versus large deformation solutions in the presence of linear versus nonlinear corrugated metal behavior, linear versus
nonlinear soil model behavior, and bonded versus frictionless soil-structure interface behavior.

Specifically, three separate studies are investigated for a corrugated steel pipe under progressively deep soil fill for the
modeling conditions identified in the table below.

Deformation theory Corrugated Metal Soil Model Interface

Study # Small Large Linear Elast-plastic Linear Duncan/Selig Bonded Frictionless
9 9 9 9
1 9 9 9 9
9 9 9 9
9 9 9 9
9 9 9 9
2 9 9 9 9
9 9 9 9
9 9 9 9
9 9 9 9
3 9 9 9 9
9 9 9 9
9 9 9 9

Each study contains four cases comparing small deformation solutions and large deformation solutions for linear and
nonlinear corrugated metal models. Study #1 investigates the four cases under the assumption of a linear soil model and
bonded interface. Study # 2 investigates the four cases under the assumption of a Duncan/Selig soil model and bonded
interface. Study # 3 investigates the four cases under the assumption of a linear soil model and a frictionless interface.

5.6.1 CANDE soil-structure model and parameters

All the CANDE simulations are represented by finite element mesh shown in Figure 5.6.1-1 and have the following
common features:

• Pipe is 60-inch corrugated steel, 3x1-inch corrugation, 0.109-inch thickness

• Soil is homogenous without bedding and weighs 120 pcf.
• Maximum soil height is 160 feet above the crown
• Truncated mesh height is 10 feet above crown (LEVEL 2 PIPE)
• Soil loading in excess of 10 feet is placed as equivalent increments of overburden pressure, where each
increment represents 10 feet of soil.
• The first load increment includes the entire soil mass with 10 feet of fill followed by 15 increments of
overburden pressure.

5-18
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Figure 5.6.1-1. CANDE mesh of corrugated steel pipe (symmetric half)

Overburden
Pressure

10 ft

Soil Density
120 pcf
CMP 60-inch
3x1, 0.109-inch thick

Parameter values for the linear and nonlinear cases are identified in the table below.

Table 5.6.1-1. Linear versus nonlinear modeling assumptions.

System Component Linear form Non-linear form

Deformation theory Small deformation theory Large deformation theory

(Standard engineering) (Updated Lagrange)
Corrugated metal stress-strain Linear Elastic: Elastic-Plastic:
relationship: Young’s mod = 29,000,000 psi Modulus = 29,000,000 psi
Yield stress = 33,000 psi
Soil stress-strain relationship: Linear Elastic: Duncan/Selig Hyperbolic model for
Young’s mod = 1000 psi ML90
Poisson ratio = 0.3
Soil-culvert interface Fully Bonded Frictionless slippage with tension
assumption: separation allowed

In each study, comparison of the four cases are presented as vertical deflection versus fill height, maximum thrust stress
versus fill height, and buckling safety factor versus fill height. These quantities along with plastic penetration constitute
the design criteria for corrugated metal pipe.

5.6.2 Study # 1 -- Linear soil and bonded interface

The first study assumes the culvert installation is defined by a linear elastic soil and a bonded soil-structure interface
with the following four modeling cases:

5-19
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

(1) linear steel material and small deformation theory

(2) linear steel material and large deformation theory
(3) nonlinear steel material and small deformation theory
(4) nonlinear steel material and large deformation theory

Deflection. The following figure shows the vertical deflection of the corrugated steel pipe for four modeling cases.

Figure 5.6.2-1 Load-deflection plot – vertical deflection as percent of diameter (Study #1)

180 Load-Deform ation Plots

160

140
Height of Cover Soil (feet) .

120

100

80

60 #1 - All Linear
#2 - Large Deform
40 #3 - Metal plastic

20
#4 Large D+Plastic

0
0 2 4 6 8 10 12 14
Percent Deflection

As expected, large deformation theory (case 2) produces deflections that progressively increase compared to the linear
model (case 1). Although not shown, the large-deformation deflection curve becomes nearly flat at 300 feet of fill.

Case 3 shows that material nonlinearity by itself produces deflections that exceed case 2 after about 90 feet of fill. The
dogleg bend in the deflection curve is caused by complete plastic hinging at crown, springline, and invert occurring at
80 feet of fill. Outer fiber compressive yielding first begins at the springline at 30 feet of fill, becomes a complete
plastic hinge at 70 feet of fill, and finally forms a collapse mechanism at 80 feet of fill. After 80 feet fill the soil
provides the elastic stiffness to resist complete collapse.

Case 4, which includes both material nonlinearity and large deformations, is presumably the most realistic case for
applying the design criteria for deflection. Assuming 5% deflection is the allowable design limit, then 80 feet of fill is
the maximum safe burial depth.
Thrust stress. The figure below shows the load-thrust plots for the same four cases wherein the maximum thrust occurs
at the springline.

5-20
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Figure 5.6.2-2 Load-Thrust stress plots of corrugated metal pipe (Study #1)

Load - Thrust Relationship

180

160

140
#1 - All Linear
Height of Soil Cover (feet) .

#2 - Large Deform
120 #3 - Metal plastic
#4 - Large D+Plastic
100

80 Full plastic penetration at

crown, springline and invert
60 for cases 3 and 4

40

20

0
0 5000 10000 15000 20000 25000 30000 35000 40000
Maximum Thrust Stress (psi)

Comparing plots for cases 1 and 2, we observe that the large-deformation solutions are only slightly less than small
deformation solutions. The same finding applies to comparing cases 3 and 4 so that it may be concluded that the thrust
response is fairly insensitive to the large deformations.

In contrast to the above, when we compare cases 1 and 3 (or cases 2 and 4) we see that the thrust stress response is very
sensitive to complete plastic penetration of the corrugated cross-section, causing an abrupt change of behavior. The
abrupt change of thrust behavior in cases 3 and 4 is characterized by significantly smaller increases in thrust stress as
increments of overburden pressure are applied. This behavior is to be expected because the combination of thrust and
moment has produced complete plastic penetration of the corrugated cross-section at the crown, springline and invert
so that the pipe now has minimal hoop stiffness (like slotted joints) which forces the soil to carry the bulk of the
additional overburden pressure.

For the reasons described above, cases 3 and 4 do not produce a maximum thrust stress that approaches the yield stress
value of 33,000 psi even at 160 feet of fill. This result reinforces the notion that the design criteria for corrugated metal
should include a limit on the allowable plastic penetration because thrust stress alone can be misleading as evidenced in
the above figure.

Buckling Stress and Safety Factor. The last comparison for Study # 1 is the buckling load and associated safety factor.
Table 5.6.2-1 lists CANDE’s prediction for the thrust stress level that will produce a singular system matrix for each
load increment. The buckling capacity is the maximum thrust stress value multiplied by the computed buckling factor
“c”. Of course, the CANDE predictions, which are based on large deformation theory, only apply to cases 2 and 4.

For reference Table 5.6.2-1 also shows the simplified AASHTO prediction for buckling stress (LRFD specifications
12.7.2.4). The AASHTO equation for buckling stress is only dependent on the pipe diameter, modulus and cross-
section geometry, it is not dependent on soil stiffness (assumed the same for all installations), nor is it dependent on
metal yield stress. Thus, the predicted AASHTO buckling stress is the same for cases studied herein and is computed to
be 42,867 psi, which is 30% greater than yield stress. Another reference value is the often-used Chelapatti-Allgood
prediction, which predicts the buckling thrust equal to 54,275 psi. This prediction is based on the buckling of an elastic
ring embedded in elastic springs (soil) and subjected to hydrostatic pressure.

5-21
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Table 5.6.2-1 Buckling stress predictions

Predicted Thrust Stress Level For Buckling-like Response (psi)
Fill Height
(feet) CANDE – (Case 2) CANDE – (Case 4) ASSHTO Chelapatti-Allgood
Linear steel Nonlinear steel Prediction
10 100,385 100,385 42,867 54,257
20 100,195 100,195 42,867 54,257
30 103,025 101,980 42,867 54,257
40 100,648 97,322 42,867 54,257
50 100,025 80,563 42,867 54,257
60 103,398 58,011 42,867 54,257
70 101,341 42,049 42,867 54,257
80 101,104 15,989 42,867 54,257
90 102,561 0 42,867 54,257
100 102,238 0 42,867 54,257
160 102,057 0 42,867 54,257

For the case of linear steel, CANDE’s prediction for buckling stress remains nearly constant and is roughly twice the
value of the AASHTO or Chelapati-Algood approximations. However for the case of nonlinear steel, the CANDE
predictions for buckling capacity begin to steadily decrease as the steel begins to yield at about 30 feet of fill. When the
complete collapse mechanism has formed at 80+ feet of fill, the buckling capacity approaches zero. Recall that
CANDE’s prediction of buckling stress is based on the current system stiffness matrix that becomes less and less robust
as steel yielding increases with each load step.

Figure 5.6.2-3 shows the buckling safety factors as a function of fill height. Safety factors are defined as the buckling
stress capacity listed in Table 2 divided by the corresponding thrust stress shown in Figure 5.6.2-2. Note that the
loading axis in the figure has been switched to the horizontal in order to illustrate the decrease in safety as fill height
increases.

For the case of linear steel, it is evident that the AASHTO buckling safety factor is more conservative than CANDE
prediction for all stages of fill height. However when comparison is made for cases including nonlinear steel, the
AASHTO safety factor predictions become less conservative after 80 feet of fill. That is the AASHTO safety factor
tends to stay constant beyond 80 feet of fill due to the nearly constant thrust stress from the nonlinear steel solution.

In contrast the CANDE safety factor rapidly decreases due to the rapid loss of buckling capacity such that, after 70 feet
of fill, CANDE provides a more conservative (and accurate) prediction of buckling safety than does the AASHTO
prediction,

From the above study it is concluded that the AASHTO prediction for global is buckling is not always conservative.

5-22
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Figure 5.6.2-3 Buckling safety factor versus fill height for corrugated steel pipe (Study # 1)

Buckling Safety Factors

10.00

9.00 AASHTO-Linear Steel

CANDE-Linear steel
8.00 AASHTO- Nonlinear Steel
CANDE-Nonlinear Steel
7.00
Safety Factor .

6.00

5.00

4.00

3.00

2.00

1.00

0.00
0 20 40 60 80 100 120 140

Height of Soil Cover

5.6.3 Study # 2 -- Nonlinear soil and bonded interface

This study assumes a nonlinear soil model while retaining the bonded soil-structure interface to define the test problem.
Specifically the nonlinear soil is called the Duncan/Selig hyperbolic formulation and represents silty sand compacted to
90% relative density (CANDE reference ML90).

Deflection. Figure 5.6.3-1 shows the vertical deflection of the corrugated metal pipe for four modeling cases composed
small and large deformation theory and linear and nonlinear model for the corrugated steel. In this study the plot for
case # 1 is not a linear response as it was in the previous study. Rather the curve exhibits a decreasing defection as fill
height increases. This, of course, is the influence of the soil model, which becomes stiffer under as confining pressure
increases. Note, the soil model becomes less stiff under increasing shear strain, however the overall effect in this case
is a stiffening of the soil mass.

Comparative observations between the deflection plots for four cases in are similar to Study #1. Specifically, large
deformation theory (case 2) produces deflections that progressively increase as compared to the small deformation
model (case 1). The dogleg bends in the deflection curves of cases 3 and 4 are caused by complete plastic hinging at
crown, springline, and invert occurring at 80 feet of fill. Outer fiber compressive yielding first begins at the springline
at 30 feet of fill, becomes a complete plastic hinge at 70 feet of fill, and finally forms a collapse mechanism at 80 feet
of fill. After 80 feet of fill the ever-stiffening soil provides the stiffness to resist complete collapse.

Case 4, which includes both material nonlinearity and large deformations, is presumably the most realistic case for
applying the design criteria for deflection. Assuming 5% deflection is the allowable design limit, then 90 feet of fill is
the maximum safe burial depth.

5-23
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Figure 5.6.3-1 Load-Deflection plots of corrugated metal pipe (Study # 2)

Load-Deformation plots

180

160

140
Height of Cover Soil (feet) .

120

100

80
#1 - All Linear
60 #2 - Large Deform

40
#3 - Metal plastic
#4 - Large D+Plastic
20

0
0 2 4 6 8 10 12 14
Percent Deflection

Thrust stress. Figure 5.6.3-2 shows the load-thrust plots for the same four cases. In all cases the maximum thrust stress
occurs at the springline. The thrust stress responses in this figure are very similar to Study #1. Overall, the effect of the
nonlinear soil model is to reduce thrust stress values by about 5%, otherwise the shape of the curves and behavior is the
same. Again we conclude that the thrust response is fairly insensitive to large deformations.

As discussed in the previous study, the abrupt change of thrust behavior in cases 3 and 4 is to be expected because the
combination of thrusts and moments has produced complete plastic penetration of the corrugated cross-section at the
crown, springline and invert so that the pipe now has minimal hoop stiffness (like slotted joints) which forces the soil to
carry the bulk of the additional overburden pressure.

As mentioned before, cases 3 and 4 do not produce a maximum thrust stress that approaches the yield stress value of
33,000 psi even at 160 feet of fill. This result reinforces the notion that the design criteria for corrugated metal should
include a limit on the allowable plastic penetration because thrust stress alone can be misleading.

5-24
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Figure 5.6.3-2 Load-Thrust stress plots of corrugated steel pipe (Study # 2).

Load-Thrust Relationship

180
#1 - All Linear
160 #2 - Large Deform
#3 - Metal plastic
140
#4 - Large D+Plastic
Height of Soil Cover (feet) .

120

100

Full plastic penetration at crown,

80
springline and invert for cases 3 and 4.

60

40

20

0
0 5000 10000 15000 20000 25000 30000 35000 40000
Maximum Thrust Stress (psi)

Buckling Stress and Safety Factor. The last comparison for Study # 2 is the buckling load and associated safety factor.
Table 5.6.3-1 lists CANDE’s prediction for the thrust stress level that will produce a singular system matrix for each
load increment. Of course, the CANDE predictions, which are based on large deformation theory, only apply to cases 2
and 4.

As explained in Study # 1, the table also contains the simplified AASHTO prediction for ASSHTO buckling stress and
the Chelapatti-Allgood prediction.

Table 5.6.3-1. Buckling stress predictions for Study # 2

Predicted Thrust Stress Level For Buckling-like Response (psi)
Fill Height
(feet) CANDE – (#2) CANDE – (#4) ASSHTO Chelapatti-Allgood
Linear steel Nonlinear steel Prediction
10 107174 107174 42,867 54,257
20 105029 105029 42,867 54,257
30 106067 109260 42,867 54,257
40 111191 102047 42,867 54,257
50 116490 98317 42,867 54,257
60 123310 88218 42,867 54,257
70 131538 67710 42,867 54,257
80 136658 37562 42,867 54,257
90 142060 0 42,867 54,257
100 150731 0 42,867 54,257
160 175659 0 42,867 54,257

5-25
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

For the case of linear steel, CANDE’s prediction for buckling capacity increases steadily at each load step because the
soil stiffness is increasing without a loss in structural stiffness. However for the case of nonlinear steel, the CANDE
predictions for buckling stress begin to steadily decrease as the steel begins to yield at about 40 feet of fill. When the
complete collapse mechanism has formed at 80+ feet of fill, the buckling capacity approaches zero. Recall that
CANDE’s prediction of buckling stress is based on the current system stiffness matrix that becomes less and less robust
as steel yielding increases with each load step.

The following figure shows the buckling safety factors as a function of fill height. Safety factors are defined as the
buckling stress capacity listed in the above table divided by the corresponding thrust stress shown in Figure 5.6.3-2.
Note that the loading axis in the figure has been switched to the horizontal in order to illustrate the decrease in safety as
fill height increases.

Figure 5.6.3-3. Buckling safety factor versus fill height for corrugated steel pipe (Study #2)

Buckling Safety Factors

10.00
AASHTO-Linear Steel
9.00 CANDE-Linear steel
AASHTO- Nonlinear Steel
8.00
CANDE-Nonlinear Steel
7.00

6.00
Safety Factor

5.00

4.00

3.00

2.00

1.00

0.00
0 20 40 60 80 100 120 140
Height of Soil Cover (feet)

In the above figure we can make the same observations as we did in Study # 1. Namely for the case of linear steel, the
AASHTO buckling safety factor is more conservative than CANDE prediction for all stages of fill height. However
when comparison is made for cases including yielding steel, the AASHTO safety factor predictions become less
conservative after 80 feet of fill. That is the AASHTO safety factor stays relatively constant due to the nearly constant
thrust stress from the nonlinear steel solution.

In contrast to AASHTO, the CANDE safety factor rapidly decreases due to the rapid loss of buckling capacity such
that, after 80 feet of fill, CANDE provides a more conservative (and accurate) prediction of buckling safety than does
the AASHTO prediction. Again, it is concluded that the AASHTO prediction for global is buckling is not always
conservative.

5-26
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

5.6.4 Study # 3 -- Linear soil and frictionless interface

Like the first study, Study # 3 uses a linear soil model, however the soil-structure interface is now assumed to be
frictionless. The frictionless interface also infers that the culvert will separate from the soil if tension forces should
occur.

Deflection. Figure 8 shows the vertical deflection of the corrugated metal pipe for three modeling cases: (the case of
steel yielding with small deformations is no longer of interest)

(1) linear steel material and small deformation theory

(2) linear steel material and large deformation theory
(3) yielding steel material and large deformation theory

Figure 5.6.4-1 Load-Deflection plots of corrugated steel pipe (Study # 3)

Load-Deformation plots

180

160

140
Height of Cover Soil (feet) .

120

100

80
#1 - All Linear
60
#2 - Large Deform
40 #3 - Large D+Plastic
20

0
0 2 4 6 8 10 12 14
Percent Deflection

The load-deflection curves for Study 3 are similar to the corresponding curves in Study 1except that all displacements
in the above figure are somewhat larger. In the case of combined large deformations and steel yielding, the deflections
become excessive at 100 feet fill. This is because at 90 feet of fill the steel at the crown and the invert begin to buckle
inward and separates from the soil due to tension break in the interface model. This causes the large displacement
observed at 100 feet of fill in the above figure.

Thrust stress. Figure 5.6.4-2 shows the load-thrust plots for the same three cases. In all cases the maximum thrust stress
occurs at the springline. Again, the thrust-load plots are similar to the corresponding plots in Study # 1 except that all
thrusts are somewhat lower than the completely bonded case. This result is ascribed to loss of shear traction in the
frictionless case. For case # 3, the combination of thrust and moment form a collapse mechanism at about 80 feet of fill
so that further increase in thrust stress is retarded

5-27
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Figure 5.6.4-2. Load-Thrust stress plots of corrugated metal pipe (Study # 3).

Load-Thrust relationship
180

160

140
Height of Soil Cover (feet) .

Steel buckles away from soil at

120
crown and invert for case 3 at 90
feet ( interface seprates).
100
Full plastic penetration at
crown, springline and invert for
80
case 3 at 80 feet.

60

40 #1 - All Linear
#2 - Large Deform
20
#3 - Large D+Plastic
0
0 5000 10000 15000 20000 25000 30000 35000 40000
Maximum Thrust Stress (psi)

Buckling Stress and Safety Factor. The last comparison for Study # 3 is the buckling capacity and associated safety
factor. Table 5.6.4-1 lists CANDE’s prediction for the thrust stress level that will produce a singular system matrix for
each load increment. The table also contains the simplified AASHTO prediction for ASSHTO buckling stress and the
Chelapatti-Allgood prediction.

Table 5.6.4-1 Buckling stress predictions for Study #3

Fill Height Predicted Thrust Stress Level For Buckling-like Response (psi)
(feet)
CANDE – (#2) CANDE – (#3) ASSHTO Chelapatti-Allgood
Linear steel Nonlinear steel Prediction
10 84219 84219 42,867 54,257
20 84318 84318 42,867 54,257
30 83282 83663 42,867 54,257
40 82244 77324 42,867 54,257
50 83348 64154 42,867 54,257
60 83256 43962 42,867 54,257
70 81915 27274 42,867 54,257
80 81140 4196 42,867 54,257
90 83076 0 42,867 54,257
100 82295 0 42,867 54,257
160 83145 0 42,867 54,257

For the case of linear steel, the predicted buckling stress remains relatively constant at a nominal value of 83,000 psi,
which is about 20% less than the corresponding predictions in Study # 1. The 20% reduction is attributed to the
observation that the frictionless interface does not generate the same amount of shear stiffness in the soil as does the
fully bonded condition. Note that the Chelapati-Algood model is completely devoid of shear stiffness, which is one
reason that their buckling predictions are on the low side.

5-28
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

As observed in the previous studies, the CANDE predictions for buckling stress in the presence of yielding steel begin
to steadily decrease as the steel begins to yield between 30 and 40 feet of fill. When the complete collapse mechanism
has formed at 80+ feet of fill, the buckling capacity approaches zero.

Figure 5.6.4-3 shows the buckling safety factors as a function of fill height. Safety factors are defined as the buckling
stress capacity listed in the above table divided by the corresponding thrust stress shown in Figure 5.6.4-2. Again it is
observed that in the case of yielding steel, the CANDE safety factor rapidly decreases due to the rapid loss of buckling
capacity. After 60 feet of fill, CANDE provides a more conservative (and accurate) prediction of buckling safety than
does the AASHTO prediction.

Figure 5.6.4-3 Buckling safety factor versus fill height for corrugated metal pipe (Study #3)

Buckling Safety Factors

10.00
AASHTO-Linear Steel
9.00
CANDE-Linear steel
8.00
CANDE-Nonlinear Steel
7.00
Safety Factor .

6.00

5.00

4.00

3.00

2.00

1.00

0.00
0 20 40 60 80 100 120 140
Height of Soil Cover (feet)

5.6.5 Comparing Studies #1, #2 and #3 with design criteria

We conclude this chapter by comparing the maximum allowable fill heights for the three studies based on working-
stress design criteria for corrugated steel pipe. Restricting the comparison to the cases of large deformation with steel
yielding, which are presumably the most realistic CANDE solutions, the three studies are differentiated as follows:

• Study #1 = linear soil with bonded interface,

• Study #2 = nonlinear soil with bonded interface
• Study #3 = linear soil with frictionless interface

The design criteria and safety factors are given in the table below,

5-29
Chapter 5 – Large Deformations and Buckling CANDE-2007 Solution Methods and Formulations

Table 5.6.5-1 Working stress design criteria for corrugated metal pipe

Design Criterion Safety Factor Allowable limit

Thrust yielding 2.0 Thrust stress = ½(yield stress)

Global buckling 2.0 Thrust stress = ½(buckling prediction)
Plastic penetration 1.5 Amount of wall yielding = 2/3 of depth
Vertical deflection 1.0 Deflection = 5% of diameter

Using the above allowable limits and referring to the previous plots for the cases of large deformation with steel
yielding, the maximum fill height may determined for each design criterion. The results for each study are recorded in
the table below.

Table 5.6.5-2. Allowable fill heights for each design criterion (feet)
Thrust Vertical
Study # Yielding Global Buckling Plastic Penetration Deflection

1 73.0 ft 73.3 ft 55.2 ft 79.5 ft

2 79.4 ft 81.1 ft 49.7 ft 87.7 ft
3 88.9 ft 70.3 ft 53.9 ft 70.9 ft

Plastic penetration is the controlling design criterion for each study, which limits the maximum burial depth on the
order of 50 feet. It is interesting to note that all three studies (installation conditions) have both strengths and
weaknesses with regard to the design criteria.

As a final comment, the CANDE program performed in a robust manner to obtain converged solutions with several
nonlinear models being exercised simultaneousl

5-30
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

6 DESIGN CRITERIA AND LRFD DESIGN METHODOLGY

6.1 Introduction
The probabilistic-based design philosophy known as load and resistance factor design (LRFD) has been adopted into
the AASHTO LRFD Bridge Design Specifications (Reference 6) wherein the specifications for buried structures are
primarily contained in Section 12. One of the tasks in the NCHRP CANDE Upgrade project was to incorporate LRFD
methodology for all culvert materials into CANDE-2007 in addition to the traditional working stress design (WSD)
methodology.

This chapter provides a succinct interpretation of the AASHTO LRFD specifications for buried structures with regard
to defining factored loads and establishing design criteria. Tabularized values are summarized in this chapter for load
factors and load modifiers to quantify dead loads, earth loads, and live loads, which are required for a LRFD analysis to
produce the factored demands. The contents of this chapter are published in the Transportation Research Record No.
2028, Reference 23.

Also presented are tabularized descriptions of the design criteria and the corresponding factored capacities, which are
defined and quantified herein for corrugated metal, reinforced concrete, and plastic pipe materials. In some cases, the
design criteria have been generalized to provide mechanistically meaningful statements that are compatible with
comprehensive analysis methods like CANDE.

Although this chapter is focused on LRFD design methodology, the fundamental design criteria presented herein are
also used for CANDE’s WSD methodology. The WSD approach uses service loads and unfactored capacities to
compute safety factors defined as capacity divided by service-load demand. The LRFD approach uses factored loads
and factored capacities to compute ratios of factored demand divided by factored capacity for strength related design
criteria, and service-load demand divided by performance limits for service related design criteria.

6.2 Objective and Scope

The objectives of this chapter are as follows:

• Summarize the loading definitions and load factors for LRFD methodology that are applicable to the
comprehensive solution approach embodied in the CANDE-2007 program.

• Provide a clear and concise set of design criteria for buried structures for corrugated metal, reinforced concrete
and plastic pipe that are mechanistically consistent with AASHTO LRFD specifications, but are not dependent
on simplified design equations.

• Identify areas where improvements in the current AASHTO specifications are desirable and offer suggestions
for a more generalized statement of the design criteria.

LRFD methodology for buried structures includes both service limit states and strength limit states. One consequence
of the two limit states when using a comprehensive solution method is that two separate solutions may need to be
obtained, one with service loading and one with factored loading. The definitions of service loading and factored
loading are provided in the following discussion along with tables summarizing the specified load factors. To augment
this discussion, guidance is provided on applying these factored loads in the context of a comprehensive finite element
solution methodology.

6-1
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

6.3 Service Loads

A common example of a design criterion that is associated with service loading is the allowable deflection of flexible
culverts. In the case of rigid culverts, a common example is the allowable concrete crack width. Later, the exact form of
these and other design criteria will be discussed and quantified. For present purposes it is only necessary to understand
that a solution at the service load level is required in order to evaluate the service limit design criteria. For a culvert
system to successfully pass the service load design criteria, the predicted response such as the percent deflection must
be less than or equal to the allowable limit.

The service load level is the actual design load experienced by the culvert system as if one were performing a working
stress analysis. Service loads include the following conditions:

1. Dead load of structure

2. Earth load on the soil
3. Live load pressure on the structure transmitted through soil
4. Live load multiplier due to vehicle impact
5. Live load multiplier due to multiple lane presence
6. Other design load considerations such as hydrostatic or thermal loading

In terms of finite element modeling, the structure dead load (body weight of the culvert) is typically applied in the first
load step after consideration of in situ soils, followed by a sequence of load steps composed of soil element layers,
which are placed around and above the culvert up to the design cover height. Each soil layer is loaded with its own
body weight according to the soil density associated with the region. Live loads are typically applied in the last load
step(s) on the soil surface as a pressure boundary condition, representing the tire footprint pressures from the design-
truck. AASHTO specifications also require that service-load tire pressures be increased by multipliers to account for
dynamic impact and the effects of multiple lane presence, which are prescribed as follows:

• Impact multiplier = 1 + 0.33( 1.0 – H/8.0) ≥ 1.0 Equation 6-1

where, H is height of soil cover above the crown in feet.

• Multiple-lane-presence multiplier = 1.2 Equation 6-2

The multiple lane presence factor is 1.2 for a single loaded lane and considers that a single truck has a high likelihood
of being overloaded. While LRFD allows reduced values of the multiple presence factor when more than one lane is
loaded, calculations for culverts under live load show that the single loaded land condition virtually always controls the
design.

As a side comment, it is noted that the prescription of the basic live load tire pressure presents a special problem for
two-dimensional, plane strain finite element programs because the tire foot print area extends to infinity in the out-of-
plane direction, thereby overestimating the actual loading on the culvert from as compared to the actual footprint width
of finite dimensions. One way to correct for this problem is to reduce the surface pressure using the procedure
described in Reference 12 and Reference 13. Other methods are currently being invested under NCHRP Project 15-29.

6-2
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

6.4 LRFD Loads for Strength-limit States

For strength limit states such as the yield strength, ultimate strain, or global buckling capacity of the soil-structure
system, the LRFD methodology assigns net multiplying factors to the service loads. The resulting structural responses,
which are at higher levels of stress than the service load responses, are called the factored demands. Concurrently,
resistance factors, whose values are typically 1.0 or less, are multiplied by the strength values and the resulting
products are called factored capacities. The test of a successful culvert design using the LRFD methodology is that the
factored capacities are greater or equal to the corresponding factored demands for all the strength-state design criteria.
In a simplistic sense, the ratio of the effective load factor to the resistance factor is comparable to the safety factor
concept used in working stress design methodology. However, it is generally accepted that LRFD methodology offers a
more logical design assessment than WSD methodology.

The net multiplying factor applied to service loads is product of two terms called the load factor and the net load
modifier. Stated explicitly, the net factored load at any load step may be written as:

Factored load = γ η (Service load) Equation 6-3

where, γ is the specified load factor, and η is the net load multiplier.

6.4.1 Load Factors.

The load factor term, symbolically denoted as γ, is currently specified in AASHTO LRFD specifications as shown in
the table below. A general observation that should fit well with reader’s intuition is that γmax (on average) increases in
value from the dead load case to the earth load case to the live load case. This increase in load factor value roughly
corresponds with our lack-of-confidence in knowing the actual load value in these three cases.

TABLE 6.4.1-1 Load factors and load modifiers for buried structures
Culvert Type Dead Load Culvert Earth fill Loading Live Load
(DC) (EB) (LL)
γmax γmin ηDC γmax γmin ηEB γmax ηLL
Reinforced 1.25 0.9 1.05 1.3 0.9 1.05 1.75 1.00
Concrete pipe
Reinforced 1.25 0.9 1.05 1.3 0.9 1.05 1.75 1.00
Concrete box
Reinforced 1.25 0.9 1.05 1.3 0.9 1.05 1.75 1.00
Concrete arch
Corrugated metal 1.25 0.9 1.05 1.95 0.9 1.05 1.75 1.00
pipe or arch
Corrugated metal 1.25 0.9 1.05 1.50 0.9 1.05 1.75 1.00
box
Plastic pipe 1.25 0.9 1.05 1.95 0.9 1.05 1.75 1.05
(HDPE or PVC)

Symbols:
γmax = maximum load factor dependent on load case and culvert type.
γmin = minimum load factor dependent on load case and culvert type.
ηDC = net load modifier for DC load case = {(ductility)(redundancy)(importance)}
ηEB = net load modifier for EB load case = {(ductility)(redundancy)(importance)}
ηLL = net load modifier for LL load case = {(ductility)(redundancy)(importance)}

Two observations about the load factors in Table 1 that require further explanation are; (1) there are two values, γmax
and γmin, listed for each load case, and (2) the load factors are dependent on the type of culvert material and shape.

With regard to the first observation, the maximum load factor, γmax, is generally the value used in an analysis to
produce the worst-case structural response. However during the sequence of load steps in a comprehensive solution

6-3
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

method, it is conceivable that for some load steps the consistent application of γmax may not produce a worst case
scenario for a given design criteria. For example, if the earth-fill layers of soil are heavily compacted along the sides of
the culvert so that lateral pressure causes an inward movement, the resulting change in moments in the culvert may
become less than the initial moment caused by the dead load of culvert in step 1, and also counter to the sign of the
moments that will be added into the culvert when subsequent earth-fill layers are placed on top of the culvert. Thus in
the spirit of seeking the worst-case LFRD loading scenario, it may be reasonable to use γmin for those load steps
wherein earth-load layers are placed along the sides of the culvert.

The second observation, i.e., the load factor values are dependent on culvert type and shape, does not have a rational
explanation except to say the AASHTO specifications represent work-in-progress. It is certainly reasonable to assign
different resistance factors to different structural materials; however it is difficult to argue why the load factors, such as
those for vertical earth load, should be different. Perhaps at sometime in the future, the load factors will be uniform for
all buried structures. However until that time comes, Table 1 serves as the reference for load factors.

6.4.2 Load Modifiers.

The so-called net load modifier, symbolically denoted as η, is a product of three sub-factors related to ductility,
redundancy and operational importance that are described in AASHTO LRFD Specifications 1.3.2.1. For buried
structures it is customary to assign the sub-factors associated with ductility and operational importance to be 1.0.
However for the redundancy sub-factor the LRFD specification suggests that value should be 1.05 for buried structures
under dead loads and earth loads, but not live loads.

Based on the above discussion, the values in Table 6.4.1-1 for ηDC and ηEB are recommended to be 1.05 when the
maximum load factor γmax is being used. On the other hand, if the minimum load factor γmin is assigned to a particular
load step, then the corresponding load modifier is the inverse value, e.g. load modifier = 1/1.05 = 0.95.

The load modifier for operational importance is a function of the specific structure being designed, thus, while often set
to 1.0, other values may be appropriate.

The above loading conditions are not a recipe for all culvert problems because worst loading scenario depends on the
structure type, installation, and the governing limit state. However, the beauty of comprehensive solution procedures
like CANDE is that it easy to re-run the same problem with different load factors to find the worst loading condition.

6.4.3 LRFD load factors and nonlinear soil models.

A long standing issue with regard to LRFD analysis is whether or not nonlinear soil models should be driven by
stresses and strains from factored loading or from service loading. Although there is no question that factored loads are
required to produce the factored demands to evaluate the LRFD design criteria against the factored capacities, some
engineers argue that the factored soil stresses and strains cause the nonlinear soil models to become overly stiff due to
the increased confining pressure from factored loads, especially for the case of deep burial wherein nonlinear soil
models like the Duncan/Selig hyperbolic soil models generally result in substantially increased soil stiffness due to
“artificially” increased confining stress from the earth load factor. Others argue that the LRFD philosophy implies that
factored loads have some “real” probability of occurring so that the increased soil stiffness is the proper analysis
approach. Of course, this dilemma only applies to nonlinear soil models because linear soil models are insensitive to
the level of stress.

Rather than adopting only one method or the other in the CANDE-2007 program, the Duncan/Selig soil model is
programmed to operate by either method at the user’s option. Specifically, the input variable called NON in the
Duncan/Selig input instructions allows the user to choose whether "service stresses" (NON = 0) or "factored stresses"
(NON =1) are used to compute the Duncan/Selig stiffness values for Young's modulus and Bulk modulus. Thus even
though factored loads are input into CANDE and the solution output are actual factored structural responses, for the
case NON = 0, the stiffness calculations performed within the Duncan/Selig subroutine are based on service stresses at
each load step. This is achieved by dividing the factored stress and strain increments by the corresponding load factor
for that load step and maintaining a running total of the service stresses and strains to drive the nonlinear soil model.

6-4
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

6.5 Design Criteria

In the following paragraphs, the AASHTO LRFD design criteria are regrouped into three material types called
corrugated metal, reinforced concrete, and plastic pipe. The design criteria are summarized in tables followed by a
discussion of each design criterion. The tables list the resistance factors, factored capacities and the corresponding
reference to AASHTO LRFD Specification. We believe the tabularized design criteria accurately reflect the intentions
of the AASHTO specifications; however in some cases, denoted with an asterisk, we have generalized the criterion to
be mechanistically compatible with comprehensive solution methods like CANDE-2007.

6.5.1 Corrugated metal

Corrugated steel and corrugated aluminum have the same general AASHTO design criteria, differing only by the
numerical value of design criterion limits such as the yield stress. Table 2 is applicable to both types of corrugated
metal. The criterion for plastic penetration is a generalization of the AASHTO specifications and is discussed below.

Table 6.5.1-1 Corrugated Metal LRFD Design Criteria.

Design Criterion Resistance Factored Definition and
(Strength limits) Factor Capacity AASHTO Reference
(1) Thrust stress (psi) φ1 = 1.0 φ1F1 F1 = metal yield strength
(Equation 12.7.2.3-1.)
(2) Global Buckling (psi) φ2 = 1.0 φ2F2 F2 = buckling thrust stress
(Equations 12.7.2.4-1-2)
(3) Seam strength (psi) φ3 = 0.67, 1.0 φ3F3 F3 = seam strength
(Section 12.7.2.5)
(4) Plastic Penetration *(%) φ4 = 0.90 φ4F4 F4 = fully plastic (100%)
(Replaces 12.9.4.3.1)
Performance Limits
( Service Load)
(5) Allowable deflection *(%) -- P5 P5 = 5%, (2% LS)

1. The thrust stress limit value, F1, is typically 33,000 psi for steel and 24,000 psi for aluminum. The thrust stress
criterion is based on AASHTO LRFD equation 12.7.2.3-1.

2. The thrust stress level that causes global buckling, F2, may be conservatively approximated from the
simplified AASHTO LRFD equations 12.7.2.4-1-2. A more accurate alternative is to utilize the new large
deformation formulation with buckling capacity prediction available in CANDE-2007

3. If seams are not present, the factored seam strength value φ3F3 may be set equal to F1. If longitudinal seams
are present, F3 should represent the measured seam capacity and φ3 = 0.67

4. Failure from full plastic penetration occurs when the cross section becomes fully plastic due to plastic hinging,
which occurs when the internal moment approaches the plastic moment capacity (F4 = 100%). The
recommended LRFD resistance factors are 0.90 for steel and 0.85 for aluminum, respectively. This criterion is
a generalization of the original AASHTO LRFD equation 12.9.4.3.1 for structural plate box culverts.

AASHTO specifications are silent on limiting moments or plastic hinging for corrugated metal structures other
than structural plate box structures. As implied in the above table, we believe plastic penetration design
criterion should applicable to all corrugated metal structures including pipes, arches and long spans.

5. The service load limit for allowable deflection, P5, is generally taken as 5% for all corrugated metal structures
except long spans, which are usually limited to 2% of the vertical rise. Although AASHTO Bridge

6-5
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

Specification does not currently specify deflection limits, experience has shown that deflections greater than
5% indicate that the backfill type and/or compaction are below standard.

6.5.2 Reinforced concrete

The AASHTO design criteria for reinforced concrete are the most challenging criteria to generalize because the criteria
are masked in simplified solution methods and are differentiated for various culvert shapes. In Table 6.5.2-1 and the
ensuing discussion, it is believed that the proposed design criteria faithfully represent the intent and in some cases
improve the clarity of the criteria as presented in the AASHTO LRFD specifications.

Table 6.5.2-1 Reinforced Concrete LRFD Design Criteria

Design Criterion Resistance Factored Definition and
(Strength limits) Factor Capacity AASHTO Reference
(1) Steel yielding (psi) φ1 = 0.90 φ1F1 F1 = reinforcement strength
(Alternative to 12.10.4.2.4a-1)
(2) Concrete crushing* (psi) φ2 = 0.75 φ2F2 F2 = concrete strength
(New proposed criterion)
(3) Shear failure (lb/in) φ3 = 0.90 φ3F3 F3 = concrete shear strength
(Multiple AASHTO equations)
(4) Radial tension failure (psi) φ4 = 0.90 φ4F4 F4 = radial tension strength
(Adapted from 12.10.4.2.4c-1)
Performance Limits
( Service Load)
(5) Allowable crack width* (in) -- P5 P5 = 0.01 inch
(Recommendation)

1. The steel yielding stress limit value, F1, is nominally 60,000 psi for deformed bars and 65,000 psi for smooth
wire fabric. This criterion is a generalization of AASHTO LRFD Equation 12.10.4.2.4a-1 and more
fundamental then the AASHTO equation for steel area. Note that the AASHTO equation specifies the required
steel area for the steel-yielding criterion based on a simplified solution method.

2. The outer-fiber concrete crushing stress limit value, F2, is the ultimate compressive strength fc’. A concrete
compression criterion is explicitly stated in the AASHTO LRFD specifications because it is assumed that
reinforcing steel is designed to yield prior to the load level causing the concrete to begin crushing. Here it is
proposed that concrete crushing should be retained as a direct design criterion and the specified resistance
factor for concrete crushing be less than specified the resistance factor for steel yielding, with values of 0.75
and 0.90, respectively.

3. The shear capacity, F3, is the shear force causing diagonal tension failure at a given cross-section. If the shear
demand exceeds the shear capacity, then transverse reinforcement (stirrups) is required. The value of F3 is
specified differently by AASHTO depending on structural shape and burial depth.

• For concrete pipes and arches, the shear strength is specified by Equations 12.10.4.2.5, which
yield a variable value for F3 dependent on the values for moment, thrust and shear around the
culvert wall.

• For boxes and 3-sided structures with 2 or more feet of soil cover, the shear strength is specified
by Equations 5.14.5.3-1 wherein the value for F3 is dependent on moment and shear (not thrust).

• For boxes and 3-sided structures with less than 2 feet of soil cover, the shear strength is specified
by equations in Section 5.8.3.3 that, in some cases, depend on the nature of the traverse
reinforcement (stirrups).

Clearly there is a need for unifying the AASHTO LRFD specifications for the shear strength capacity because
rational mechanics suggests that shear strength capacity should be a function of the cross-section properties

6-6
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

and state of loading, not on the culvert shape or depth of burial. Hopefully this remark may stimulate a future
research study to develop a more fundamental criterion, which is the root cause of diagonal cracking. However
until that time, we propose that AASHTO Equation 12.10.4.2.5 is the best predictor of shear strength capacity
for culverts because of the large experimental data base upon which it was developed (Reference 14).

4. The radial tension failure stress, F4, is related to the tension strength of concrete and the structure span. The
criterion is adapted from AASHTO LRFD equation 12.10.4.2.4c-1. It is restated here in psi units as:

F4 = 37.92(fc’/1000)1/2 Frt Equation 6-4

where Frt is a scale factor dependent on structure span, which is specified in the paragraphs following Equation
12.10.4.2.4c-1.

The corresponding factored demand is the average radial tensile stress that the curved inner cage exerts on the
inside concrete cover thickness. This is automatically determined in CANDE-2007 by dividing the interior
cage steel force (maximum tensile force per unit length) by the radius of curvature of the steel cage.

5. The allowable crack width at service loading is generally taken as 0.01 inches in accordance with AASHTO
LRFD specification 12.10.3. An empirical formula to predict crack width, developed by Heger and McGrath
(Reference 15), has been adopted by AASHTO in the form of Equation 12.10.4.2.4d-1. For purposes of this
paper, the authors reworked the equation in a more fundamental form, which linearly relates crack width to the
tension stress in the reinforcing steel once it exceeds the surrounding concrete tensile strength. The revised
equation is:

CW = (tbsl / 2n)1/3{fs – 0.0316C1(h/d)2√fc’/ρ}/3000 Equation 6-5

where, CW is crack width in inches, ≥ 0

fs is tension steel stress in ksi units
fc’ is concrete strength in ksi units
All other symbols are section properties defined in Section 12.10.4.2.4d

CANDE-2007 predicts the crack width at each point around the culvert by feeding the computed tension steel
stress fs into Equation 5.6 thereby producing a semi-empirical prediction.. This procedure has been well
calibrated for crack widths of approximately 0.01 in. and becomes increasingly uncertain as the crack width
varies from this value. As an alternative, CANDE 2007 provides the option to use the older Gergely-Lutz
formula instead of Equation 6-5.

6.5.3 Plastic pipe

The design criteria for plastic pipe, summarized in Table 6.5.3-1, are straightforward interpretations of the AASHTO
LRFD specifications. Said another way, the AASHTO fundamental design criteria are easily understood and not
confounded with the simplified methods that are offered to predict the factored demands. The main difficulty in
applying the plastic pipe design criteria is deciding whether to use the long-term or short-term material properties.
Typically an engineer would select short-term material properties if live loads were the major concern. On the other
hand, for a deep burial situation long-term properties are appropriate. In some cases, it may be prudent to solve the
problem twice, once with short-term properties and once with long-term properties. This is a relatively easy task with
CANDE-2007.

6-7
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

Table 6.5.3-1 Plastic Pipe LRFD Design Criteria.

Design Criterion Resistance Factored Definitions and
(Strength limits) Factor Capacity AASHTO Reference
(1) Thrust failure (psi) φ1 = 1.0 φ1F1 F1 = plastic yield strength
(Equation 12.12.3.5.1)
(2) Global buckling (psi) φ2 = 1.0 φ2F2 F2 = thrust stress for buckling
(Equation 12.12.3.5.2-1)
(3) Combined strain (in/in) φ3 = 1.0 φ3F3 F3 = fiber strain limit
(Equation 12.12.3.5.4a-1)
Performance Limits
( Service Load)
(4) Allowable -- P4 P4 = 5%
displacement (%) (Recommendation)
(5) Allowable tensile -- P5 P5 = 0.05 (for HDPE)
strain (in/in) (Table 12.12.3.3-1)

1. The thrust stress limit or ultimate strength, F1, is dependent on the type plastic and the load duration. Types of
plastic identified by AASHTO include high-density polyethylene (HDPE) and polyvinyl chloride (PVC) with
characteristic material properties for short-term or long-term loading. Nominal values, taken from the
ASSHTO LRFD specifications and elsewhere, are shown in Table 6.5.3-2. Thrust load capacity is also
influenced by profile geometry, which can reduce section capacity due to susceptibility to local buckling as
discussed below.

Table 6.6.3-2 Recommended plastic properties for short and long-term loading
Effective Young’s Modulus Ultimate strength
(PE) (PU)
Type of plastic Short-term Long-term Short-term Long-term
(ksi) (ksi) (ksi) (ksi)

HDPE – 110.0 22.0 3.00 0.90

PVC – 400.0 140.0 6.00 2.60

PP – 135.0 27.0 3.10 1.00

2. The thrust stress level that causes local global buckling, F2, may be conservatively approximated from
the simplified AASHTO LRFD equations 12.7.2.4-1-2. A more accurate alternative is to utilize the
new large deformation formulation with buckling capacity prediction available in CANDE-2007.

3. The combined strain limit value (combined thrust and bending), F3, is specified by AASHTO LRFD Equation
12.12.3.5.4a-1 and is equal to 1.5 times the long-term strength divided by the long-term modulus.
Accordingly, HDPE ~ 0.06 in/in, PVC ~ 0.028 in/in. and PP ~ 0.045 in/in.

4. The service load value for allowable deflection, P4, is generally taken as 5% of the diameter for all plastic
pipes; however, the deflection limit is not directly specified in the AASHTO LRFD Design Specifications.

5. The service load value for maximum allowable tensile strain, P5, is specified in AASHTO LRFD table
12.12.3.3-1 and is 0.05 in/in for HDPE and 0.035 or 0.05 in/in for PVC depending on cell class.

Local Buckling. Local buckling is not a direct design criterion for plastic pipe, but it does influence the demands and
capacities of the design criteria listed above. Local buckling is a nonlinear phenomenon caused by compressive forces

6-8
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

that induce the sub elements of profile wall pipe to deform out-of-plane in a wrinkled pattern. Similar to corrugated
metal pipe wherein some amount of outer fiber yielding is permitted, some amount of local buckling is permitted in
plastic profile pipe. Chapter 2 provides a description of the CANDE’s nonlinear model to account for local buckling
based on Section 12.12.3.5.3 of AASHTO LRFD specifications. The procedure simulates the effect of local buckling
by reducing the effective area of the sub elements dependent upon the average level of compressive strain in each sub
element. Thus, local buckling increases the demand and/or decreases the capacity of design criteria listed above.

6.6 Illustration of LRFD Factors and Evaluation with CANDE

This example is for a concrete arch with straight legs installed in a trench-like installation with backfill soil placed in
eight layers to a final cover height of 2 feet above the crown. A live load is placed over the arch centerline on the ninth
and last load increment. To evaluate this problem under LRFD design methodology, CANDE’s Level-2 arch mesh
option is used to automatically construct a symmetric finite element mesh with nine construction increments with the
load factors taken from Table 6.4.1-1 and computed as described in the following paragraphs.

6.6.1 Construction increment number 1.

As shown in Figure 6.6.1-1, the first construction increment includes the surrounding in situ soil (red), concrete footing
(yellow), and reinforced concrete arch culvert (black). The in-situ soil and the footing are assigned appropriate stiffness
values, but with zero body weight. That is, since the in-situ soil and footing are in place before the arch is constructed,
the arch is only loaded with its own self-weight. Thus, the arch is assigned a body weight of 150 pcf, representing body
weight of concrete. LRFD analysis requires the user to define the load factor for each construction increment. In this
case, the first load step represents dead load (γ = 1.25 and η = 1.05) so that the net load factor = 1.25*1.05 = 1.31.

6.6.2 Construction increment numbers 2 through 8

Figure 6.6.2-1 shows construction increments 2 through 8, which are all associated with earth loading. The 2nd
increment is shown as two rows of orange elements, the 3rd increment is shown in yellow, the 4th in green and so on up
to the final cover height. Each of these layers of soil is assigned the actual body weight (120 pcf) of the fill soil. Also
each increment is assigned the same load factor based on earth loading, computed as (γ = 1.30 and η = 1.05) = 1.37.
CANDE applies the load factor to the earth loads in steps 2 through 8 so that the accumulated solution is the factored
demand at the end of each step.

Figure 6.6.1-1 Concrete arch installation for first construction increment (net load factor =1.31)

6-9
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

Figure 6.6.2-1 Concrete arch installation for construction increments 2-8 (net load factor =1.31)

6.6.3 Construction increment number 9 (live load).

The last construction increment represents a live load as illustrated by the red circle in the figure below. The load is
applied in the 9th step without additional soil loading as a downward force (boundary condition) representing the actual
service load of a design truck, but reduced for plane-strain considerations. In this case the net load factor is computed
as (γ = 1.75 and η = 1.05) = 1.84. CANDE applies this load factor to the service load as prescribed in the boundary
condition. Thus CANDE’s final accumulated solution at the end of load step 9 is the total factored demand.

Figure 6.6.3-1 Concrete arch installation for ninth construction increment (net load factor =1.84)

6-10
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

6.6.4 Final evaluation of LRFD design criteria

The final evaluation of the reinforced concrete arch is shown Figure 6.6.4-1in terms of the design criteria for reinforced
concrete. CANDE prints out this evaluation at each load step but our interest is focused on the last load step because it
represents the worst case. Here we observe that the concrete arch is safe for the proscribed loading conditions because
all demand-to-capacity ratios are less than 1.0. In this example, the steel yielding design criterion is called the
controlling design criterion because it has the highest ratio. The controlling node number in the evaluation summary
identifies the location in the concrete arch where each design criterion has the maximum demand-to-capacity ratio.
Although not shown here, the evaluation summary also includes a performance assessment including the predicted
crack width under service load conditions.

Figure 6.6.4-1 LRFD evaluation summary for concrete arch after load step 9 (copy from CANDE)

LRFD SUMMARY EVALUATION FOR GROUP 1, LOAD STEP 9

DESIGN-CRITERION CONTROL FACTORED FACTORED RATIO

NODE DEMAND CAPACITY VALUE

STEEL YIELDING (psi) 1 46450.1 54000.0 0.860

CONCRETE CRUSHING (psi) 12 1934.2 3000.0 0.645

SHEAR FAILURE (lbs/in) 11 460.5 782.7 0.588

RADIAL-TENSION FAIL (psi) 1 7.8 54.6 0.142

(Print out from CANDE-2007)

6-11
Chapter 6 – Design Criteria and LRFD Design Methodology CANDE-2007 Solution Methods and Formulations

T(This page intentionally left blank)

6-12
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

7 BANDWIDTH MINIMIZATION
A bandwidth minimizer is installed in CANDE-2007 that reduces the maximum bandwidth of the system of equations
characterizing the finite element mesh topology. Bandwidth minimization is achieved by strategically renumbering the
node numbers that were originally input by the user. Node renumbering is accomplished heuristically by swapping
node numbers, a pair at a time, starting with the node number with the highest bandwidth and exchanging it with
another node number that optimally reduces the maximum bandwidth at the first location while not generating a
bandwidth at the second location that is greater or equal to the original maximum bandwidth.

7.1 Background
For any given finite element problem solved by CANDE, the number rows in global stiffness matrix is dictated by the
number of nodal unknowns, which defines the number of equations. In contrast, the number of rows in the global
stiffness matrix is dependent on the maximum bandwidth, which depends on the nodal numbering pattern, not the
number of equations. By judiciously numbering the nodal pattern, the maximum bandwidth can de reduced. The
motivations for reducing the maximum bandwidth are, (1) to reduce storage space for the global stiffness matrix whose
size is equal to the number of equations times the maximum bandwidth, and (2) to reduce the time required to solve the
system of equations, which is proportional to the number of equations times the square of the bandwidth.

For Level 2 operations, which use “canned” finite element meshes, a bandwidth minimizer is not necessary because the
nodal numbering pattern has been optimized once and for all time. On the other hand for Level 3 operations, which
implies a user-defined mesh topology, a bandwidth minimizer can be useful. Although experienced users usually
construct nodal numbering patterns that are reasonably close to an optimum pattern, there are situations in which the
nodal numbering pattern may differ substantially from optimum. For example, when a pre-existing finite element mesh
is altered to include more elements and nodes in a local area, the added nodes are likely to have node numbers that are
numerically much larger than the existing nodes in this neighborhood, thereby resulting in an excessively large
bandwidth.

The bandwidth minimization problem may be formally stated as follows. Given a sparse, n x n, matrix A with
components aij, find a permutation of rows and columns (node renumbering pattern) such that all the non-zero
components of matrix A reside in a band that is close as possible to the main diagonal. Or expressed mathematically,
Minimize{ i-j for i, j = 1 to n: aij ≠ 0}.

Beginning in the 1950s, the bandwidth minimization problem received a lot of attention from mathematicians due to
the emergence of the computer for solving large systems of equations representing a wide range of applications. The
classical Cuthill-McKee algorithm, published in 1969, employs the breadth-first search method to construct a level
structure in the context of graph theory. A similar but more efficient algorithm, known as Gibbs-Poole-Stockmeyer,
was published in 1976, and remains very popular.

Although the above and many other general-purpose bandwidth minimization algorithms are available in the literature,
the following development is unique and takes advantage of the finite element assembly methodology to systematically
reduce the bandwidth. To the best of the author’s knowledge the bandwidth minimization algorithm presented herein is
novel.

7-1
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

7.2 Objective and Scope

Within the context of the CANDE finite element program and mesh topology, the objectives are;

1. Develop a robust algorithm to convert an arbitrary node-numbering input pattern into a new node-numbering
pattern that minimizes the maximum bandwidth.

2. Display all printed and graphical output from the CANDE solutions in terms of the original node-numbering
pattern so that the user does not need to learn the new node number scheme, i.e., transparent to user.

Like other finite element programs, CANDE’s finite element methodology has three properties that are extremely
useful for the bandwidth minimization algorithm:
• First, the global system matrix is symmetric so that bandwidth minimization is restricted to one side of the
matrix diagonal.
• Second, each node number represents a group of nodal unknowns (e.g., x-displacement, y-displacement, and
rotation) each with the same bandwidth. Thus bandwidth minimization is reduced to reordering the node-
number pattern as opposed to the entire set of equations.
• Third, all the information required to minimize the bandwidth is contained in the so-called element
connectivity matrix, which simply identifies the node numbers associated with each element.

The above advantages are exploited in the following development.

7-2
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

7.3 Bandwidth Minimization Methodology in CANDE

7.3.1 Element connectivity matrix

The element connectivity matrix contains the node numbers that are attached to each element and is symbolically
denoted and dimensioned as NOD(NET,NDX), where NET is the total number of elements and NDX is the maximum
number of nodal locations as required by the various element types. In the CANDE program, for example, NDX = 4 to
accommodate the 4-noded quadrilateral elements, which has more nodes than the other types of elements as shown in
the table below.

Element Type NOD(NE,1) NOD(NE,2) NOD(NE,3) NOD(NE,4)

(connected nodes)
Beam-column Node-1 Node-2 0 0
(nodes=2)
Triangle Node-1 Node-2 Node-3 0
(nodes=3)
Interface Node-1 Node-2 Node-3 0
(nodes=3)
Quadrilateral Node-1 Node-2 Node-3 Node-4
(nodes=4)

Letting NE be a counter from 1 to NET, then the components of element connectivity matrix are identified as:
• NOD(NE,1) = node number assigned to location 1 of element NE
• NOD(NE,2) = node number assigned to location 2 of element NE
• NOD(NE,3) = node number assigned to location 3 of element NE (depending on element type)
• NOD(NE,4) = node number assigned to location 4 of element NE (depending on element type)

If, for example, element number 10 is a beam element with the two end-nodes numbered 33 and 43, then NOD(10,1) =
33 and NOD(10,2) = 43. The remainder of the nodal list for element 10 is equal to 0.

7.3.2 Algorithm Cycle.

With the above information as a starting point, bandwidth minimization is accomplished by swapping node numbers, a
pair at a time, starting with a node number that has the highest bandwidth and exchanging it with another node number
that optimally reduces the maximum bandwidth at the first node location while not generating a bandwidth at the
second location that is greater or equal to the original maximum bandwidth.

The process is presented in the following four steps. These steps are repeated iteratively until no additional node pairs
can be found to further reduce the bandwidth.

Step 1. Use the current NOD connectivity matrix to create a new matrix denoted and dimensioned as MT(NDT,3)
where NDT is the total number of nodes in the mesh. Letting N represent each node number in turn, the three
components of the MT matrix are defined as follows:

• MT(N,1) = Maximum node number among all nodes that are attached to any element that is also attached to
node N. MT(N,1) is referred to as NDMAX.

• MT(N,2) = Minimum node number among all nodes that are attached to any element that is also attached to
node N. MT(N,2) is referred to as NDMIN.

• MT(N,3) = Maximum of {ILEFT or IRITE}, where ILEFT = N - NDMIN, which is the bandwidth to the left
of node N, and IRITE = NDMAX – N, which is the bandwidth to the right of node N. MT(N,3) is referred to
as MAXB, which is the largest bandwidth associated with node N.

7-3
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

The computational key to establishing the MT matrix is to start with an outer do-loop for each node number (referred to
as the pivot node). Insert an imbedded do-loop for the element numbering and use the NOD connectivity array. If any
of the element’s connectivity nodes match the pivot node, then record the maximum and minimum nodal values found
in that element’s connectivity array. Replace the maximum and/or minimum nodal values by node numbers from
subsequent element connectivity arrays if the node values are more extreme than the previously recorded values. At
completion of the element loop, the values of NDMAX, NDMIN and MAXB for the current pivot node are stored in
the MT matrix.

Step 2. Search the MT(N,3) column to find the maximum bandwidth of all nodes. Denote and save this value as
MAXALL, (i.e., MAXALL = Maximum of {MT(N,3), N=1 to total number of nodes}. It is often the case that several
nodes share the same maximum bandwidth value, MAXALL. Accordingly, each of these node numbers is a candidate
for the current nodal position to be called node-1. The primary goal is to optimally reduce the bandwidth at the node-1
position by interchanging the node-1 number with an as yet undetermined node number, called node-2. Of course, the
insertion of the node-1 number into the node-2 position must also be considered as explained in the next step.

Step 3. The purpose of step 3 is to select an optimum pair of nodes, called node-1 and node-2, such that the
interchange of their node numbers reduces the local bandwidth at node-1, while maintaining the bandwidth at node-2 to
be less than MAXALL. A candidate number for node-1 is labeled N1 and a candidate number for node-2 is labeled N2.
We begin with an outer do-loop for all nodes using the node counter N1, and we exclude all N1 node numbers such that
MT(3,N1) ≠ MAXALL. Thus, the N1 nodes that are not excluded are candidates for node-1. For node-2 candidates, we
use an inner do-loop wherein we restrict our search for N2 node numbers in the sequential range;

N2(start) = MT(1,N1) – MAXALL + 1

N2(end) = MT(2,N1) + MAXALL – 1

This restricted range of N2 contains the only node numbers that will reduce the bandwidth at the N1 position.

If the candidate node pair, N1 and N2, were interchanged such that number N2 is assigned to position 1 and number N1
is assigned is assigned to position 2, then the new bandwidths at these two positions can be respectively determined as;

MAXB1 = Maximum of {N2-MT(N1,2) or MT(N1,1)-N2}

MAXB2 = Maximum of {N1-MT(N2,2) or MT(N2,1)-N1}

The next step within the inner loop is to test whether or not the current candidate pair, N1 and N2, is a better choice
than the best of the previous candidate pairs. This test is passed by satisfying two simultaneous requirements:

(1) MAXB1 is smaller than that produced by any previous candidate pairs, and

(2) MAXB2 is less than MAXALL

When all candidate node pairs have been tested, there are two possible outcomes, which dictate one of two possible
courses of action:

(1) A final node pair, node-1 and node-2, has been identified that reduces the bandwidth. In this case we proceed
to Step 4 to update matrices and vectors to perform the node swapping.

(2) No node pair could be found to reduce the current maximum bandwidth (MAXALL). In this case, we exit
from the bandwidth minimization process because the job is done.

Step 4. Given a viable solution for node-1 and node-2, the purpose of this step is to update the element connectivity
matrix with the current node swap. It is also useful to update a history vector that keeps track of all the nodal changes,
called an “is-was” vector.

7-4
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

To update the connectivity matrix, we perform an outer loop search with element counter NE, and an inner loop search
with node-location counter K. At any point during the update process, let node-x be the nodal value representing
NOD(NE,K). The update is easily achieved by two successive if-then statements:

If node-x equals node-1, then set NOD(NE,K) = node-2

If node-x equals node-2, then set NOD(NE,K) = node-1

When the search loops are completed, the NOD connectivity matrix is completely updated and we may return to Step 1
to repeat the entire process with the new NOD matrix, which provides new starting conditions.

However before returning to Step 1, it is useful to update a history vector that relates the current node numbering to the
original node numbering defined by the user. To this end, denote and dimension a nodal vector ISWAS(NPT) and
initialize ISWAS(N) = N, for N progressing from 1 to NPT. This initialization infers that new node numbers N are
identical to the old node numbers ISWAS(N).

To update for each node swap, let K1 = ISWAS(node-1) and K2 = ISWAS(node-2), where K1 and K2 are the original
node numbers that are currently represented by node-1 and node-2, respectively. Then nodal swap is easily updated and
saved as

ISWAS(node-1) = K2

ISWAS(node-2) = K1

When all cycles of node swapping is complete (job done), the ISWAS vector will contain the final correspondence
between the new node number N and the old node number ISWAS(N). Said another way, ISWAS(N) is the original
node number that identifies the same nodal location that is now identified as N. As discussed in the next section, the
ISWAS vector is useful for updating other nodal vectors such as nodal coordinates that must be converted to the new
nodal numbering pattern.

This marks the end of Step 4 and the algorithm requires that we return to Step 1 for another cycle. The termination of
cycles is decided in Step 3.

7.3.3 Post algorithm-cycle updates.

When the algorithm cycle is exited at Step 3, the maximum bandwidth has been minimized, and the updated NOD
connectivity matrix and the ISWAS vector are complete. However before the finite element solution process can begin,
other nodal data that is defined in the original nodal numbering scheme must be converted to the new node numbering
scheme such as nodal coordinate vectors and boundary condition vectors prescribing displacements and/or forces at
nodes. The ISWAS vector is used to accomplish these conversions.

For vectors such as nodal coordinates, say X and Y, that implicitly reference the old nodal number by means of the
vector index number N, it is necessary to first make the conversion to a temporary vector, say T, over a do loop N = 1
to NDT as shown below:

T(N) = X(ISWAS(N))

After this do-loop is completed, the X vector is converted to the new node numbers by a second do-loop over the same
index range:

X(N) = T(N)

The two-step conversion is required so that information in X vector is not prematurely overwritten, which would occur
in a one-step conversion.

7-5
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

For vectors that explicitly define old node numbers such as the list of nodes where boundary conditions are applied, say
vector NB, the conversion can be made in one do-loop as,

NB(I) = NB(ISWAS(I))

where I = 1, to the number of boundary conditions.

After the last conversion of old node numbers to new numbers is completed, the ISWAS vector as previously defined is
no longer useful. In preparation for the finite element solution phase, it is convenient to redefine the ISWAS vector to
be the inverse of its original definition. That is, ISWAS (N) is the new node number corresponding to the old node
number N. The inverse is accomplished in a two-step process using a temporary vector T, similar to the conversion of
nodal coordinates. Specifically, for N = 1 to NDT:

T(ISWAS(N)) = N

Followed by a second do-loop to store inverse definition back into the ISWAS,

ISWAS(N) = T(N)

In summary, the entire process described above is easily executed in a single subroutine (called BMIZER in CANDE).
The output from this subroutine includes the updated NOD matrix, the updated nodal vectors (X, Y, NB, etc.) and the
ISWAS vector in its second definition, that is, ISWAS(N) is the new node number associated with old node number N.
From this point on, the finite element assembly and solution routines will be working the new node numbering pattern.

7-6
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

7.4 Transparency of Node Numbers in Solution Output.

The second objective of this development is to list the solution output with reference to the original node numbers. A
prime example is the tabular listing of displacements where, for example, the output displays values for N, U(N), V(N),
where U(N) and V(N) are the x and y displacements at node N. Since U(N) and V(N) are in reference to the new node-
numbering pattern, we must redefine the displacement indexes if N represents the sequential numbers of the old
numbering pattern.

The proper indexes for the displacements are contained in the ISWAS vector (2nd definition). Therefore, to display the
information by a write statement imbedded in a do-loop for N = 1 to NPT, the write statement would read as,

WRITE(*,*) N, U(ISWAS(N)), V(ISWAS(N))

Output of any other nodal data is handled in the same manner.

Fortunately, element output data such as stress and strain measures are referenced to element numbers, not node
numbers. Thus element output data is the same for either the old or new nodal number pattern.

7-7
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

7.5 Illustration of Bandwidth Minimization

A simple finite element mesh layout, resembling a house on stilts, is shown in Figure 7.5-1 and is used to illustrate the
bandwidth minimization algorithm. The mesh contains a total of eight elements composed of 4 beam elements, 2
quadrilateral elements, and 2 triangular elements. The initial node-numbering pattern, subsequently referred to as the
old node numbers, is shown in the figure with red numbers. A quick inspection of Figure 1 reveals that the initial node-
numbering pattern is far from optimum because of the intentional disparity between node number values surrounding
each node point. Thus, there is room for improvement from the bandwidth minimization algorithm.

Figure 7.5-1. Mesh topology for house-on-stilts showing elements and original node pattern

7 8

4
10 11

5 6

8 9
2
5 4

6 7
1 3

1 2

B e a m e le m e n t n u m b e r s : 1 , 2 , 3 & 4
Q u a d r ila t e r a l e le m e n t n u m b e r s : 5 & 6
T r ia n g le e le m e n t n u m b e r s : 7 & 8
N o d e n u m b e r in g p a tt e r n s h o w n in r e d

Table 7.5-1 shows the initial connectivity array for the eight elements that is easily deduced from the above figure.
Also shown in the table is the initial MT Matrix that is constructed from the element connectivity array as described in
Step 1 of the algorithm. For Step 2 it is observed from the last column in the MT matrix that the overall maximum
bandwidth is eight (MAXALL = 8).

In Step 3, which is the heart of the algorithm, the MT matrix shows that both node #3 and node #11 are candidates for
the primary swapping node because both have a maximum bandwidth equal to 8. For node #3, the candidate secondary
node numbers are # 4 to #10, which is the node range that can reduce the bandwidth at node #3. Similarly for node #11,
the candidate secondary node numbers are #4 to #10, which is the node range that reduces the bandwidth at node #11.

7-8
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

Of all the possible combinations of primary and secondary node pairs, the winning pair, passing the final test in Step 3,
is shown at the bottom of Table 1 as primary node = #3 and secondary node = #7.

Table 7.5-1. NOD and MT Matrices for initial node numbering (Node swaps = 0)
NOD – Matrix MT-Matrix
(Element Connectivity) (Bandwidth Analysis)
Element Node a Node b Node c Node d Pivot NDMAX NDMIN MAXB
Node
1 6 1 0 0 1 6 1 5
2 8 6 0 0 2 7 2 5
3 7 2 0 0 3 11 3 8
4 9 7 0 0 4 11 3 7
5 10 8 5 4 5 11 4 6
6 4 5 9 11 6 8 1 5
7 10 4 3 0 7 9 2 5
8 4 11 3 0 8 10 4 4
9 11 4 5
10 10 3 7
11 11 3 8
Results of Algorithm: Nodes to be swapped
Node-1 = 3
Node-2 = 7

Step 4 updates the element connectivity matrix NOD with the node pair #3 and #7. After the update is complete the
process starts again at Step 1 to build a new MT matrix.

Table 7.5-2, like the previous table, shows the current NOD and MT matrices after the first node swap.
The MT matrix shows that the bandwidth at node #3 was reduced from 8 to 6. However the largest bandwidth among
all nodes is now reduced to seven (MAXALL = 7), occurring at nodes #4 and #11.

Table 7.5-2. NOD and MT Matrices after 1st node swap (Node swaps = 1)
NOD – Matrix MT-Matrix
(Element Connectivity) (Bandwidth Analysis)
Element Node a Node b Node c Node d Pivot NDMAX NDMIN MAXB
Node
1 6 1 0 0 1 6 1 5
2 8 6 0 0 2 3 2 1
3 3 2 0 0 3 9 2 6
4 9 3 0 0 4 11 4 7
5 10 8 5 4 5 11 4 6
6 4 5 9 11 6 8 1 5
7 10 4 7 0 7 11 4 4
8 4 11 7 0 8 10 4 4
9 11 3 6
10 10 4 6
11 11 4 7
Results of Algorithm: Nodes to be swapped
Node-1 = 4
Node-2 = 8

In like fashion Tables 7.5-3 to 7.5-7 show the algorithm results for five more successive node swaps.

7-9
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

Table 7.5-3. NOD and MT Matrices after 2nd node swap (Node swaps = 2)
NOD – Matrix MT-Matrix
(Element Connectivity) (Bandwidth Analysis)
Element Node a Node b Node c Node d Pivot NDMAX NDMIN MAXB
Node
1 6 1 0 0 1 6 1 5
2 4 6 0 0 2 3 2 1
3 3 2 0 0 3 9 2 6
4 9 3 0 0 4 10 4 6
5 10 4 5 8 5 11 4 6
6 8 5 9 11 6 6 1 5
7 10 8 7 0 7 11 7 4
8 8 11 7 0 8 11 4 4
9 11 3 6
10 10 4 6
11 11 5 6
Results of Algorithm: Nodes to be swapped
Node-1 = 10
Node-2 = 7

Table 7.5-4. NOD and MT Matrices after 3rd node swap (Node swaps = 3)
NOD – Matrix MT-Matrix
(Element Connectivity) (Bandwidth Analysis)
Element Node a Node b Node c Node d Pivot NDMAX NDMIN MAXB
Node
1 6 1 0 0 1 6 1 5
2 4 6 0 0 2 3 2 1
3 3 2 0 0 3 9 2 6
4 9 3 0 0 4 8 4 4
5 7 4 5 8 5 11 4 6
6 8 5 9 11 6 6 1 5
7 7 8 10 0 7 10 4 3
8 8 11 10 0 8 11 4 4
9 11 3 6
10 11 7 3
11 11 5 6
Results of Algorithm: Nodes to be swapped
Node-1 = 3
Node-2 = 6

7-10
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

Table 7.5-5. NOD and MT Matrices after 4th node swap (Node swaps = 4)
NOD – Matrix MT-Matrix
(Element Connectivity) (Bandwidth Analysis)
Element Node a Node b Node c Node d Pivot NDMAX NDMIN MAXB
Node
1 3 1 0 0 1 3 1 2
2 4 3 0 0 2 6 2 4
3 6 2 0 0 3 4 1 2
4 9 6 0 0 4 8 3 4
5 7 4 5 8 5 11 4 6
6 8 5 9 11 6 9 2 4
7 7 8 10 0 7 10 4 3
8 8 11 10 0 8 11 4 4
9 11 5 4
10 11 7 3
11 11 5 6
Results of Algorithm: Nodes to be swapped
Node-1 = 5
Node-2 = 7

Table 7.5-6. NOD and MT Matrices after 5th node swap (Node swaps = 5)
NOD – Matrix MT-Matrix
(Element Connectivity) (Bandwidth Analysis)
Element Node a Node b Node c Node d Pivot NDMAX NDMIN MAXB
Node
1 3 1 0 0 1 3 1 2
2 4 3 0 0 2 6 2 4
3 6 2 0 0 3 4 1 2
4 9 6 0 0 4 8 3 4
5 5 4 7 8 5 10 4 5
6 8 7 9 11 6 9 2 4
7 5 8 10 0 7 11 4 4
8 8 11 10 0 8 11 4 4
9 11 6 3
10 11 5 5
11 11 7 4
Results of Algorithm: Nodes to be swapped
Node-1 = 5
Node-2 = 6

7-11
Chapter 7 – Bandwidth Minimization CANDE-2007 Solution Methods and Formulations

Table 7.5-7. NOD and MT Matrices after 6th node swap (Node swaps = 6)
NOD – Matrix MT-Matrix
(Element Connectivity) (Bandwidth Analysis)
Element Node a Node b Node c Node d Pivot NDMAX NDMIN MAXB
Node
1 3 1 0 0 1 3 1 2
2 4 3 0 0 2 5 2 3
3 5 2 0 0 3 4 1 2
4 9 5 0 0 4 8 3 4
5 6 4 7 8 5 9 2 4
6 8 7 9 11 6 10 4 4
7 6 8 10 0 7 11 4 4
8 8 11 10 0 8 11 4 4
9 11 5 4
10 11 6 4
11 11 7 4
Results of Algorithm: Nodes to be swapped
None (job done)
No further node swaps will reduce the maximum bandwidth less than 4

The algorithm exited after six node swaps, and Table 7.5-7 portrays the final element connectivity matrix and MT
matrix. The final maximum bandwidth is 4, which is a 50% reduction of the original bandwidth.

It is reiterated here that the algorithm is intended to reduce the maximum bandwidth, not to minimize the skyline
profile of all nodes, which in this example, includes eight different nodes sharing maximum the bandwidth. Indeed it is
possible to construct a node pattern for this example in which only one node has a maximum bandwidth of 4 and all
other nodes have a bandwidth of 3 or less.

Table 7.5-8 shows the final ISWAS vector for both definitions discussed in the algorithm. The second definition is used
to reference finite element output back to the original nodal numbering system so that the user need not be concerned
with the new numbering system that is only used internally in the solution routines.

Table 7.5-8. Final results of new and old node numbering

ISWAS Definition # 1 ISWAS Definition # 2
New Nodes Old Nodes Old Nodes New Nodes
1 1 1 1
2 2 2 2
3 6 3 10
4 8 4 8
5 7 5 7
6 10 6 3
7 5 7 5
8 4 8 4
9 9 9 9
10 3 10 6
11 11 11 11

As a closing comment, it is not claimed that the above method provides an absolute global minimum to the bandwidth
problem. However all the results to date have produced node-numbering patterns that could not be further improved to
reduce the maximum bandwidth. The method is robust and rapid, typically, achieving maximum bandwidth reduction
with a minimal number of node swaps.

7-12
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

8 MODELING TECHNIQUES
When applying CANDE-2007 to analyze or design any culvert installation (or more generally any soil structure
interaction problem), modeling questions always arise. Some common questions are; what extent of the insitu soil
should be included in the finite element mesh, what soil models should be used for insitu and backfill soil, should
interface elements be included between structure and soil, how many construction increments should be used, and is
large deformation theory required. There is no clear-cut answer to these and similar modeling questions because each
problem is unique and modeling assumptions that are appropriate for one case may not be appropriate for another.

The best advice is to try as many modeling approaches as can be practically undertaken for a particular problem.
Sometimes it will be found that the differences in solutions from alternative modeling assumptions are inconsequential.
At other times, a particular modeling assumption may substantially alter the structural behavior, impacting the safety of
the design. In all cases, however, useful information is obtained that guides our insight into soil-structure interaction
and permits a more intelligent assessment of the problem.

The remainder of this chapter provides suggested techniques for modeling live loads, pipe-group connections and
combinations, and construction increments and soil compaction. However the “best advice” approach described above
should always be kept in mind.

8.1 Live Loads

External loads (as opposed to element body weight loads) are defined by boundary-condition input data described in
CANDE’s user instructions. All external loads are applied at nodes as force components in the global x and y directions
per unit length in the z direction. Alternatively, if desired, the force components can be defined in rotated x-y
coordinates defined by the user.

Figure 8.1-1 illustrates Fx and Fy force components acting along a particular node strip along the z-direction. Also
shown in the figure is a uniform surface-pressure strip load with a uniform pressure P0.

Figure 8.1-1. Illustration of nodal strip load and surface-pressure strip load.

Pressure strip loadings are converted to equivalent nodal strip loads by distributing the pressure load acting on each
element to the neighboring nodes and summing the results to determine the equivalent force strips, Fx and Fy. Figure

8-1
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

8.1-2 and the accompanying table shows an example of converting a pressure-strip load to nodal-strip loads in the
negative y-direction.

Figure 8.1-2 Illustration of converting pressure strip to equivalent nodal forces

Table 8.1-1 Symbolic calculation of equivalent nodal forces

Nodal force at nodes

Contributing Node 1 Node 2 Node 3 Node 4
Element Fy1 Fy2 Fy3 Fy4
a Pala/2 Pala/2 0 0
b 0 Pblb/2 Pblb/2 0
c 0 0 Pclc/2 Pclc/2

Pala/2 Pala/2 + Pblb/2 Pblb/2 + Pclc/2 Pclc/2

Totals

In the above, Pa, Pb and Pc are the average pressures on each element surface, and la, lb and lc are the surface lengths of
each element, respectively. The total force recorded at the bottom of the table is the boundary-condition y-force value
to be entered as input data (negative).

8.1.1 Live load modeling problem

If the live load is actually in the form of a very long surface-pressure strip, then there is no modeling problem, and the
actual surface pressure is used to calculate the equivalent boundary-condition nodal forces for CANDE analysis.
Usually, however, the surface pressure is not a very long strip, but rather a finite patch like a truck-tire footprint. In this
case additional calculations are required to correct the surface-pressure for the CANDE analysis.

The live-load modeling problem is depicted in Figure 8.1.1-1, showing both an infinite surface-pressure strip and a
finite pressure patch. When the pressure-patch load is transmitted through the soil, the load spreads outward in two
planes, the x-y plane and y-z plane. In contrast, the surface pressure strip load only spreads outward in the x-y plane.
Clearly, if the same surface pressure is assigned to the patch and to the strip, then the pressure strip load will cause
greater soil pressure impinging on the culvert than will the pressure patch load.

8-2
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Figure 8.1.1-1 Illustration of surface-pressure strip and patch-pressure footprint

The live-load modeling problem is summarized as follows. If we use the actual patch pressure (say 80 psi representing
an HS-20 truck tire) in a CANDE model or any two-dimensional model, we will over predict the loads on the culvert.

Although there is no exact cure for this problem (short of performing a three-dimensional analysis), reasonable
CANDE solutions can be obtained by appropriately correcting the strip pressure magnitude. Two methods are
presented to reduce the strip pressure, an elasticity-based method and the AASHTO-based load spreading method.

8.1.2 Elasticity-based correction for live-load strip pressure

The elasticity-based correction for strip pressure uses an exact solution for peak vertical stress in a homogenous elastic
half space loaded by a pressure patch with footprint dimensions L x W and pressure Pt, as shown in above figure. As
developed in Reference 25, the solution for peak vertical stress as a function of depth is given in the following
expression.

Pp α αh 1 1
σ y-patch = 2 [arctan( )+ ( 2 + 2 )] Equation 8.1-1
π hR 3 R 3 R1 R 2

where, σ y-patch = peak vertical stress for any patch dimensions (L x W)

Pp = surface patch pressure
L = patch length (in x-y plane)

8-3
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

W = patch width (in y-z-plane)

H = depth of interest to compute peak vertical stress
α = W/L = ratio of patch width to length
h = H/(L/2) = non-dimensional soil depth in half-tire lengths
R 1 = 1+h 2
R 2 = α 2 +h 2
R 3 = 1 +α 2 +h 2

To specialize the peak vertical stress equation to the case of pressure strip loading, we let the patch width approach
infinity (W →∞) to get.

Ps 1 h
σ y-strip = 2 [arctan( ) + 2 ] Equation 8.1-2
π h R1
where, σ y-strip = vertical stress for strip pressure for any dimension L with W = ∞.
Ps = surface strip pressure

If we equate the vertical stress prediction for the patch load (Equation 8.1-1) to the vertical stress prediction from the
strip load (Equation 8.1-2), we can determine the ratio of strip surface pressure to patch surface pressure that results in
equal vertical stress at any depth of interest. Specifically, we find

α αh 1 1
arctan( )+ ( 2+ 2)
Ps hR 3 R 3 R1 R 2
r= = Equation 8.1-3
Pp 1 h
arctan( ) + 2
h R1

The “r” ratio, is the reduction factor applied to Pp to get Ps , which is the key result of this development. For any
specific pressure patch dimension (L and W) and for any specific depth H (usually culvert cover height), we can
compute the reduced strip pressure that produces the same vertical stress on the culvert crown as the patch load for any
depth H. Table 8.1.2- provides the solution to Equation 8.1-3 for most practical problems.

8-4
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Table 8.1.2-1 Reduction factor “r” applied to strip load pressure

Non-dimensional Ratio of patch dimensions width-to-length (α = W/L)

depth (W/L) (W/L) (W/L) (W/L) (W/L)
h = H/(L/2) 0.8 1.2 1.6 2.0 2.4

1.2 0.717 0.862 0.931 0.963 0.979

1.8 0.558 0.730 0.837 0.900 0.937
2.4 0.449 0.615 0.737 0.820 0.877
3.0 0.372 0.524 0.647 0.740 0.809
3.6 0.317 0.454 0.571 0.666 0.742
4.2 0.275 0.399 0.508 0.601 0.679
4.8 0.243 0.354 0.456 0.546 0.623
5.4 0.217 0.319 0.413 0.498 0.573
6.0 0.196 0.289 0.376 0.457 0.530
6.6 0.179 0.265 0.346 0.421 0.491
7.2 0.164 0.244 0.319 0.391 0.457
7.8 0.152 0.226 0.297 0.364 0.427
8.4 0.141 0.210 0.277 0.340 0.401
9.0 0.132 0.197 0.259 0.320 0.377
9.6 0.124 0.185 0.244 0.301 0.356
10.2 0.117 0.174 0.230 0.285 0.337
10.8 0.110 0.165 0.218 0.270 0.320
11.4 0.105 0.156 0.207 0.256 0.304
12.0 0.100 0.149 0.197 0.244 0.290

To demonstrate the use of the above table, suppose we wish to analyze the effect of an HS-20 truck tire located 24
inches above the culvert crown. The tire footprint is L = 10 inches, W = 20 inches, and the surface pressure of 80 psi.

To find the equivalent strip pressure, we compute W/L = 2.0 and h = 24/(10/2) = 4.8, and enter the above table to find
the reduction factor, r = 0.55. Thus the pressure assigned to the CANDE strip load is,

Ps = 0.546Pp = 44 psi Equation 8.1-4

Although the above procedure ignores the interaction of the culvert with the soil stress, the technique generally works
well.

8.1.1 AASHTO-based method to compute live-load strip pressure

The AASHTO LRFD load spreading method is a simple concept that assumes the pressure patch load spreads
uniformly with soil depth in the overall shape of a truncated pyramid. The top plane of the truncated pyramid has the
patch pressure and area, L x W. The base-plane at any depth H is assumed to expand in the x-y plane and the y-z plane
at a constant angle. The total force on any base-plane area remains constant so that the uniform pressure decreases in
proportion the increase in base-plane area.

Single wheel correction. In order to adapt the AASHTO method to correct the strip pressure problem, we recognize
that only the y-z plane needs to be corrected because the pressure distribution in the x-y plane is accurately determined
by the plane-strain solution. Figure 8.1.3-1 shows the AASHTO approximation for load spreading in the y-z plane,
which is along the length of the culvert.

8-5
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Figure 8.1.3-1 AASHTO LRFD single wheel load distribution along axis of culvert.

As illustrated in the above figure, the base-line pressure along the length of the culvert is given by,

1
PH = Pp Equation 8.1-5
H
(1 + 1.15 )
W
where, PH = reduced pressure at depth H
Pp = surface pressure of patch, LxW.
H = depth to culvert crown
W = width of pressure patch along axis of culvert
Since the CANDE pressure strip does not account for the pressure reduction along the culvert axis, it follows that
reduction factor is given as,

PH 1
r= = Equation 8.1-6
Pp H
(1 + 1.15 )
W
Again, the “r” ratio, is the reduction factor applied to Pp to get the corrected strip pressure Ps for CANDE analysis.

To demonstrate the use AASHTO method, we will repeat the previous example from the elasticity-based method.
Again, we wish to analyze the effect of an HS-20 truck tire located 24 inches above the culvert crown. The tire

8-6
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

footprint is L = 10 inches, W = 20 inches with a surface pressure of 80 psi. Therefore, using H/W = 1.2, we evaluate
Equation 8.1-6 to get the reduction factor, r = 0.42. Thus the pressure assigned to the CANDE strip load is,

Ps = 0.42Pp = 34 psi Equation 8.1-7

In comparing the above AASHTO-based correction with the elasticity-based correction (Equation 8.1-4), we find the
elasticity-based method is more conservative because it results in a higher CANDE loading pressure than the
AASHTO-based method. This trend holds true for all practical patch dimensions and cover depths, however the
percentage difference between the elasticity-based and AASHTO-based methods diminish with depth.

Two wheel correction. One useful feature of the AASHTO approach is that the interaction of two tires separated by a
distance, S, can be accommodated by a simple extension of the one-wheel method. Typically the spacing distance S
represents the truck axle length separating two tires.

The two-wheel approach should be used whenever the culvert cover depth is deeper than the so-called wheel interaction
depth given by

S-W S-W
H int = 0
= Equation 8.1-8
2tan(30 ) 1.15

where, H int = wheel interaction depth

S = spacing between wheels, center to center.
W = patch or tire width in y-z plane
When the culvert is buried deeper than Hint , the portion of the culvert between the wheels begins to experience soil
pressure from both wheels. When pressure overlap begins to occur, the AASHTO load spreading method assumes the
pressure remains uniform over the entire interaction zone. The total force from the two tires equilibrates the force from
the pressure distribution along the culvert,. Figure 8.1.3-2 illustrates and summarizes the above concepts.

Figure 8.1.3-1 AASHTO LRFD two-wheel load distribution along axis of culvert.

8-7
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

As illustrated in the above figure, the uniform pressure along the length of the culvert is given by,

1
PH = 2Pp Equation 8.1-9
S H
(1 + + 1.15 )
W W
where, PH = reduced pressure at depth H
Pp = surface pressure of tire 1and tire 2
H = depth to culvert, providing H> Hint
Following the same argument as presented for the single wheel correction, the two-wheel reduction factor is
determined as,

PH 2
r= = Equation 8.1-10
Pp S H
(1 + + 1.15 )
W W
As before, r is the reduction factor to compute strip pressure, Ps = rPp for CANDE analysis.

To demonstrate the two-wheel AASHTO method, the same parameters for the one-wheel example are used except we
add a second wheel spaced 44 inches away. Thus the parameters are L = 10 inches, W = 20 inches, Pp = 80 psi, H = 24
inches, and S = 44 inches. We first calculate the interaction depth as Hint = 21 inches and observe that H=24 > Hint=21,
therefore, the two-wheel solution is valid. Evaluating Equation 8.1-10 the reduction factor is found as r = 0.44 so that
the pressure assigned to the CANDE strip load is,

Ps = 0.44Pp = 35 psi Equation 8.1-7

It is observed that the addition of the second wheel raised the equivalent strip pressure by 1 psi compared to the single
wheel (Equation 8.1-7). The increase is small because the depth of burial is only slightly greater than the depth of in the
interaction zone.

In closing this section on live loads, it is reiterated that there is no exact technique to simulate pressure-patch loads in a
two-dimensional framework. The methods presented above provide a reasonable approximation for design and safety
evaluations of buried culverts. However for point-by-point comparison with fully instrumented experimental tests,
there are likely to be discrepancies that can be better predicted by three-dimensional analysis.

8-8
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

8.2 Pipe Group Connections and Combinations

Finite element models utilizing multiple pipe groups is a new feature in CANDE-2007 that offers many modeling
capabilities that were not available in previous versions of CANDE. Discussed below are several techniques for
combing pipe groups with each other and soil elements to achieve various desired objectives. Topics include
connection modeling (fixed, pinned, and roller), long-span stiffeners, culvert rehabilitation with liners, and other pipe
group combinations.

8.2.1 Connections among element types

A pipe group is a connected sequence of beam-column elements joined head-to-toe at common nodes and tracing an
arbitrary curvilinear path in x-y space. Each beam-column node has three degrees of freedom, x-displacement, y-
displacement and a rotation. Continuum elements (triangle or quadrilateral) only have two degrees of freedom per
node, x-displacement and y-displacement.

When two or more pipe groups are connected to a common node the connection is called fixed, meaning moment
continuity is preserved at the junction node. On the other hand, when a pipe group is attached to a continuum element
the connection is called pinned, meaning the moment at the junction node is zero. These fundamental connections,
representing the default conditions, are illustrated in the figure below.

Figure 8.2.1-1 Connections made by sharing common node at junction.

If it is desired to simulate a pinned or roller connection between two pipe groups such as pipe group #1 and pipe group
# 2 in the above figure, then an interface element must be inserted at the junction of the two the two pipe groups. An
interface element joins two independent nodes, initially assigned to the same x-y coordinate location, as a pinned
connection, friction connection or roller connection depending on the properties assigned to the interface element.
Figure 8.2.1-2 illustrates an interface element joining pipe group # 1to pipe group # 2 including the third interface node
whose degrees of freedom are the interface forces required to enforce the desired connection.

8-9
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Figure 8.2.1-2 Techniques to change connection conditions.

The assigned interface properties required to achieve various connection conditions are listed in the table below
wherein the value 106 means a very large number such that frictional slippage or tensile rupture cannot occur.

Table 8.2.1-1 Interface property values to simulate various connection conditions

Interface properties

Desired Connection Type Friction coefficient Tensile breaking force

(μ) (lbs/inch)

Pinned 106 106

Friction (no tensile separation) 0.3 (typical) 106

Friction (tensile separation) 0.3 (typical) 0.0

Roller (no tensile separation) 0.0 106

Roller (tensile separation) 0.0 0.0

Shown on the right side of Figure 8.2.1-2 is a modeling technique to simulate a fixed connection between a pipe group
and a concrete footing (continuum elements). Here a beam-column element is added to the pipe group and imbedded in
the concrete, mimicking reality.

8.2.2 Stiffeners and culvert rehabilitation with parallel pipe groups

When two pipe groups share the same set of nodes, referred to as parallel pipe groups, it is evident that both groups
experience identical deformations (displacements and rotations) because both pipe groups are attached to the same
nodes. The combined stiffness is additive, exactly like two parallel springs with individual stiffnesses k1 and k2 have a
combined stiffness equal to k1 + k2. Even though parallel pipe groups experience identical deformations, the resulting
internal forces (moment, thrust and shear) and stresses in one group may differ substantially from the other group,
depending on individual stiffness properties of each group.

Many types of culvert stiffeners can be modeled with parallel pipe groups with one pipe-group modeling the culvert
and the second pipe-group modeling the stiffener. As noted above, a stiffener that is modeled as a parallel pipe group
provides an additive stiffness to the culvert without additional increase in bending stiffness from composite action. To
achieve composite action the stiffener must be integrally attached to the culvert including complete shear bonding like
a continuously welded built-up cross sections. Culvert sections that are truly stiffened in a composite manner can be

8-10
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

modeled with a single pipe group using composite section properties. Otherwise it is best to model stiffeners with the
more conservative parallel-pipe-group method as illustrated below.

Long-span stiffener. Long-span corrugated metal culverts are often reinforced with discrete ring-beam stiffeners
periodically spaced along the top arch. Typically the ring beams are lightly bolted to the corrugated metal so that the
assumption of additive stiffness instead of composite stiffness is quite reasonable, erring slightly on the conservative
side. Figure 8.2.2-1 illustrates how to model ring-beam stiffeners with parallel pipe groups.

Figure 8.2.2-1 Modeling technique for long-span culvert with ring beam stiffeners

As shown in the table below input for the CANDE ring beam properties for pipe group #2 are smeared values based on
a ring-beam spacing of z inches on center.

Section Properties of Ring Individual ring values CANDE input values

Beam
Area A (in2) A/z (in2/inch)
Moment of Inertia I (in4) I/z (in4/inch)
Section Modulus S (in3) S/z (in3/inch)

Another advantage of the parallel-pipe-group method is due the fact that CANDE provides an individual assessment of
each group so that the structural integrity of stiffener is evaluated in addition to the culvert.

Culvert rehabilitation stiffener. One method to rehabilitee structurally distressed culverts is to insert a pipe liner
inside the old culvert to stiffen the old soil-culvert system. Typically, the liner is smooth wall plastic pipe inserted into
a host culvert such as corrugated metal or reinforced concrete pipe. If the plastic-pipe liner is inserted into the host pipe
with a snug fit, it is reasonable to assume the combined stiffness of the host pipe and liner is properly simulated by
parallel pipe groups. The rationale is that the interface between the host pipe and liner does not have a sufficient shear
bond for the two pipes to act in a composite manner. The left side of Figure 8.2.2-2 shows the recommended parallel
pipe-group method for modeling a snug fit liner wherein both pipe groups share the same nodes.

8-11
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Figure 8.2.2-2 Modeling techniques for culvert rehabilitation with liners

The right side of the above figure shows an alternative rehabilitation method in which the outside diameter of the liner
is several inches smaller than the inside diameter of the host pipe thereby creating an annular space that is filled with
grout. In this case the combined stiffness of liner, host pipe and grout is significantly greater than the snug fit condition
using parallel pipe groups. As illustrated in the figure, an accurate simulation is achieved by assigning the liner an
independent set of nodes tracing its smaller diameter and modeling the grout with continuum elements.

If the grout elements are assigned common nodes with the liner pipe and with the host pipe, the net effect is a
composite bending stiffness that may be overly optimistic if the grout is not securely bonded both pipes. Therefore, a
more conservative approach would be to assign independent node numbers to the grout elements and use frictional
interface elements to connect the grout nodes to the pipe nodes, allowing relative slip along the pipe-grout boundaries.

8.2.3 Illustrations of pipe group combinations

In addition to the pipe group connections and combinations discussed above, additional pipe-group applications are
illustrated in the Figure 8.2.2-3.

8-12
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Figure 8.2.3-1 Illustrations of multiple pipe group applications

The above illustrations are intended to demonstrate the wide variety of modeling options that are possible with
CANDE’s multiple pipe-type capability. For example, the reinforced earth application (last figure) implies that pipe
groups can be used to simulate soil reinforcement strips. If interface elements are inserted along these pipe groups, the
model is capable of determining the ultimate pullout strength of the soil reinforcement.

8-13
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

8.3 Construction Increments

Incremental construction is the physical process of placing and compacting soil layers, one lift at a time, below, along
side and above the culvert as the installation is constructed. Analytically this is achieved by adding incremental
solutions from successive finite element configurations, where each new configuration contains additional soil elements
(and/or structural elements) that mimics the real-world buildup of the soil-structure system, one load step at a time.

The terms “load step” and “construction increment” are closely related terms and are often used interchangeably,
however they have slightly different meanings. Load step is an all-encompassing term referring to any incremental
solution, which may or may not include a construction increment. For example a load step may only include boundary-
condition loads without additional elements added to the system. A construction increment is a special kind of load step
wherein additional elements are added to the global stiffness matrix usually representing an additional layer of soil. The
body weight of added elements form the load increment associated with the load step along with any boundary loads
such as soil compaction pressures, discussed subsequently.

8.3.1 Rules and insights for construction increments

With the above understanding, construction increments are subject to the following rules. Each element is assigned a
construction increment number, which corresponds to the load step number that the element stiffness matrix and load
vector is assembled into the global system. Once an element stiffness enters the system it remains active for all
subsequent load steps (there is no element death). Of course, the element body-load vector is only applied during the
load step corresponding to the element construction increment number; it is not reapplied on subsequent load steps.
Similarly, surface pressure and/or point loads (i.e., force boundary conditions) are also assigned a particular
construction increment number and are only applied during the corresponding load step. Likewise, displacement
boundary conditions are applied during the load step they are specified and remain fixed for all subsequent load steps.
In short, the rules simulate reality.

Although the rules seem simple, it is easy to be fooled. For example, novices generally think that if all construction
increments are linear elastic and no other nonlinearities are involved, then a structure built in several steps behaves no
differently then a structure built as a monolith (all in one step). A simple example demonstrating this is not true is
shown in the figure below.

Figure 8.3.1-1 Illustration of two construction increments versus a single monolith

The figure illustrates that after the first construction increment is set in place with roller boundary conditions on the
sidewalls, the body weight produces a uniform stress distribution properly aligned in the vertical direction as expected.
However, when the second construction increment is set in place (sharing common nodes with the first construction
increment), some of the second construction increment’s body weight is transferred to the first construction increment’s
stiffness. Therefore, the resulting stress and strain distribution in the two-step system is not uniform and differs from
the uniform stress distribution in the single monolith system shown on the right.

8-14
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

The specific point of this simple example is that the sequence of construction increments is important and influences
the structural behavior of the final system. More generally, when developing a finite element model for a particular
soil-structure system, the sequence of construction increments should be modeled as realistically as practical.

8.3.2 Techniques for initial construction increment

All too often the first construction increment is not modeled properly, causing incorrect deformation in the structure
due to the insitu soil model. Proper modeling techniques depend on whether the insitu soil is linear or nonlinear and the
assignment of body weight as discussed below.

Linear insitu soil. Generally insitu soil can be realistically modeled with a linear elastic soil model because the
existing soil is pre-consolidation and/or recently unloaded from excavation. In this case, the initial construction
increment may include the insitu soil, footing or bedding and the structure as illustrated in the example below.

Figure 8.3.2-1 Initial configuration for an arch-trench installation with linear insitu soil

Only the arch should be assigned a non-zero body weight for this initial configuration. A common mistake made
by novices is to also assign body weight to the insitu soil and footing, which causes these components to deform
under there own weight and introduce fictitious distortions into the arch. Physically we know the insitu soil and
footing are at rest under their own weight before the arch is attached to the footing, thus only the arch body
weight cam cause deformations in the first load step.

Nonlinear insitu soil. If it is desired to model the insitu soil with a nonlinear soil model like Duncan/Selig, then
the arch should be eliminated from the initial configuration and introduced in the second construction increment.
In this case we need to assign the actual body weight to the insitu soil because the soil stiffness is dependent on
the stress state caused by the soil body weight, requiring iteration within the load step. Figure 8.3.2-2 illustrates
the proper initial configuration when nonlinear soil model is used for insitu soil.

8-15
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Figure 8.3.2-2 Initial configuration for an arch-trench installation with nonlinear insitu soil

If the arch were included in the initial configuration, it would experience fictitious distortions due to deformation of the
insitu soil by self-weight. Now however, after the initial configuration has converged and the insitu soil and footing are
at rest in a deformed state, the arch may be correctly assigned to the second construction increment.

8.3.3 Soil compaction and construction increments

The last technique to be described is the modeling of soil compaction using the so-called squeeze layer method first
introduced in Reference 18. The purpose of soil compaction is to densify soil after it is laid down in loose layers and
then compacted to achieve a specified density usually stated in terms of AASHTO T-99 relative density.

Compaction equipment ranging from hand-held compactors to large-tracked bulldozers is used to compress each soil
layer by creating a temporary vertical pressure to compact the soil. As the soil is compressed vertically (squeezed), it
tends to expand laterally due to the Poisson effect, which creates horizontal pressure on the sides of the culvert. These
lateral pressures cause inward movement of the culvert sides and peaking at the culvert crown, that is, a reverse
deformation pattern from that caused by overburden loading. Large culverts such as long-span corrugated metal
structures often experience as much as 2 % crown peaking as a result of compacting soil layers between the footing and
the crown.

Compaction loads are beneficial not only in compacting the soil to achieve the desired soil stiffness properties but also
in introducing reverse deformations and bending moments into the culvert, which provides the culvert with additional
capacity to resist the opposite bending deformations that subsequently come from soil placed on top of the culvert.

The squeeze layer technique is a simple modeling concept that simulates the effect of soil compaction and is applicable
to both linear and nonlinear soil models. The concept is illustrated in Figure 8.3.3-1 and discussed below.

1. To start, a uniform compaction pressure representing the compaction equipment, typically on the order 5-psi
pressure, is placed on the surface of the insitu soil at the level of the footing.

8-16
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

2. Next, the 2nd construction increment (first layer of backfill soil with body weight) is added to the system
along with another uniform compaction pressure applied to its top surface. At the same time, however, the first
compaction pressure is removed by applying an equal but opposite compaction pressure to the bottom surface.
These two opposing pressures squeeze the soil layer and increase the lateral pressure on the arch via the
Poisson effect.

3. The intermediate construction increments 3rd, 4th and 5th are treated exactly like the 2nd construction increment
so that each layer is squeezed, increasing lateral pressure on the arch and inducing more peaking. Note that
after the 5th load increment (or more generally, the increment before the crown-level increment) all the
temporary compaction pressures have been removed except the compaction pressure on the surface of 5th
construction increment.

4. The squeeze layer process is terminated with the 6th construction increment (i.e., crown-level increment). As
the 6th construction increment is added to the system with its stiffness and body weight, the last remaining
compaction pressure from the 5th construction is removed.

5. All the remaining construction increments, representing soil layers above the crown, are added in the normal
manner with body weight but without any compaction pressures.

Examples of applying the squeeze layer technique and the behavior of long-span culverts can be found in References
18, 26 and 27.

8-17
Chapter 8 – Modeling Techniques CANDE-2007 Solution Methods and Formulations

Figure 8.3.3-1 Squeeze-layer method for compaction loads on long-span culvert

8-18
Chapter 9 – References CANDE-2007 Solution Methods and Formulations

9 REFERENCES
1. CANDE: A Modern Approach for the Structural Design and Analysis of Buried Culverts, Federal Highway
Administration Report No. FHWA-RD-77-5, October 1976, (Authors: M. G. Katona, J .M. Smith, R. S.
Odello, and J. R. Allgood).

2. CANDE User and System Manuals, Federal Highway Administration Report No. FHWA-RD-77-6, October
1976, (Authors: M. G. Katona and J. M. Smith).

3. CANDE-1980: Box Culverts and Soil Models, Federal Highway Administration, Report No. FHWA-RD-172,
May 1981 (Authors: M. G. Katona, P. D. Vittes, C. H. Lee, and H. T. Ho).

4. CANDE-89: Culvert Analysis and Design computer program – User Manual, Federal Highway Administration
Report No. FHWA-RD-89-169, June 1989. (Authors: S. C. Musser, M. G. Katona and E T. Selig).

5. Predicting Performance of Buried Conduits, Report No. FHWA/IN/JHRP-81/3, Joint FHWA and Indiana
State Highway Commission, June 1982. (Authors; G.A. Leonards, T.H. Wu, and C.H. Juang)

6. AASHTO (2004), AASHTO LRFD Bridge Design Specifications, Third Edition, AASHTO, Washington, D.C.

7. Burns, J. Q., and R. M. Richard, “Attenuation of Stresses for Buried Cylinders,” In Proc., Symposium on Soil-
Structure Interaction, University of Arizona Engineering Research Laboratory, Tucson, 1964, pp. 378-392.

8. Herrmann, L. R., “User’s manual for plane strain incremental construction program,” Department of Civil
Engineering, University of California, Davis, 1972.

9. Katona, M. G., and A. Y. Akl, “Analysis and Behavior of Buried Culverts with Slotted Joints,”
Transportation Research Record, No. 1008, 1985, pp. 22-32.

10. Katona, M. G., and A. Y. Akl, “Structural Design of Buried Culverts with Slotted Joints,” ASCE, Journal of
Structural Engineering, January 1987.

11. Herrmann, L. R., “Efficiency evaluation of a two-dimensional incompatible finite element,” Journal of
Computers and Structures, vol 3, 1973, pp 1377-1395.

12. Katona, M. G., Minimum Cover Heights for Corrugated Plastic Pipe Under Vehicle Loading,” Transportation
Research Record, No. 1288, 1990, pp. 127-135.

13. McGrath, T.J., I.D. Moore, E.T. Selig, M.C. Webb, and B. Taleb, Recommended Specifications for Large-
Span Culverts, NCHRP Report 473, Transportation Research Board, Washington, DC, 2002.

14. Heger, F.J., and T.J. McGrath, Shear Strength of Pipe, Box Sections, and Other One-Way Flexural Members,
ACI Journal, Proceedings Vol. 79, 1982, pp. 470-483.

15. Heger, F.J. and T.J. McGrath. Crack Width Control of Reinforced Concrete Pipe and Box Sections, ACI
Journal, Proceedings Vol. 81, 1984, pp. 149-157.

16. Gergely, P. and L.A. Lutz. Maximum Crack Width in Reinforced Concrete Flexural Members, SP-20,
American Concrete Institute, Detroit 1968, pp. 87-117

17. Katona, M. G., and P. D. Vittes, Soil-Structure Evaluation of Buried Box Culvert Designs, Transportation
Research Record, No. 878, 1982, pp 1-7.

9-1
Chapter 9 – References CANDE-2007 Solution Methods and Formulations

18. Katona, M. G., D. F. Meinhert, R. Orillac, and C. H. Lee, Structural Evaluation of New Concepts for Long-
span Culverts and Culvert Installations, Report No. FHWA-RD-79-115, December 1979,

19. Duncan, J. M., and C. Y. Chang, Nonlinear Analysis of Stress and Strain in Soils, Journal of Soil Mechanics
and Foundations, ASCE, vol. 96, No. SM5, September 1970, pp. 1629-1653.

20. Duncan, J. M., et. al., Strength, Stress-strain and Bulk modulus Parameters for Finite Element Analysis of
Stresses and Movements in Soil Masses, Report University of California, Berkeley, GT 78-02 to National
Science Foundation, April 1978.

21. Selig, E. T., Soil Parameters for Design of Buried Pipelines, Proceedings, Pipeline Infrastructure Conference,
ASCE, 1988, pp 99-116.

22. Hardin, B. O., Effects of Strain Amplitude on the Shear Modulus of Soils, Air Force Weapons Laboratory
Technical Report AFWL-TR-72-201, Kirtland AFB, NM, March 1973.

23. Katona, M. G., and T. J. McGrath, A Guideline for Interpreting ASSHTO LRFD Specifications to Design or
Evaluate Buried Structures with Comprehensive Solution Methods, Transportation Research Record, No.
2028, 2007, pp. 211-217.

24. Katona, M. G., “A Simple Contact-Friction Interface Element with Application to Buried Culverts,
International Journal for Numerical and Analytical Methods in Geomechanics, Vol. 7, 1983, pp. 371-384.

25. Poulos, H. G., and E. H. Davis, Elastic Solutions for Soil and Rock Mechanics, John Wiley, New York, 1980.

26. Katona, M.G., “On the Analysis of Long-span Culverts by the Finite Element Method,” Transportation
Research Record, No. 678, 1978, pp. 59-66.

27. Katona, M.G., “CANDE: A Versatile Soil-Structure Design and Analysis Computer Program,” Journal of
Advances in Engineering Software, Vol. 1, No. 1, 1978, pp. 3-9.

9-2
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

10 CANDE ANALYSIS SOURCE CODE

The purpose of this chapter is to provide the reader with sufficient understanding of the CANDE Analysis Engine
source-program architecture to enable the reader to modify, change and extend the program. Researchers are
encouraged to modify and/or extend CANDE-2007 to investigate new soil models, pipe-material behavior and other
new theories and concepts. By sharing new developments among the engineering community, CANDE will continue to
grow in the future, in the same manner as it has grown in the past.

10.1 Overview of CANDE Analysis Engine Architecture

The CANDE-2007 source program is written in modern Fortran and developed with the following:

ƒ Microsoft Visual Studio 2005, Version 8.0.50727.762 (SP.050727-7600)

ƒ Intel Fortran Compiler version 9.1.3427.2005.
ƒ Microsoft .NET Framework version 2.0.50727

The MVS framework permits grouping subroutines together to form projects and then grouping projects together to
form a whole solution. CANDE’s whole solution is composed of six projects described in the table below.

Table 10.1-1 CANDE-2007 projects contained in MVS solution

Number of
Project Description programs or
subroutines
1. Cande_dll The primary executive routine controlling the overall solution strategy and 1
calls to subroutines in other projects including input/output files and error
messages. (dynamic link)
2. Engine Contains all the engineering subroutines including all finite element 56
formulations, equation solver, elasticity solution, soil models and pipe types
with design logic and structural evaluations.
3. Inc Contains all named common statements and parameter declarations whose 25
variables are shared between subroutines by means of “Include statements”.
4. Intel Utility subroutines for controlling screen monitor. 3

5. IO Utility subroutines for input/output message writing to print files and screen 27
monitor for all output including error messaging.
6. Main Dynamic link interface between Cande_dll and monitor screen to identify 1
input file.

In the following discussion, the CANDE architecture is presented from a top down viewpoint, starting with the primary
executive subroutine, Project Cande_dll, and then narrowing to specific subroutines in Project Engine.

10.2 Executive Routine (Cande_dll)

Overall, the executive routine is divided into three parts, input phase, solution phase and output phase. The executive
routine defines the Fortran variable ICOME to communicate with other subroutines as to which phase is currently
being addressed as follows.

• ICOME = 1 means input phase

• ICOME = 2 means solution phase
• ICOME = 3 means output phase

10-1
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

As might be expected, the solution phase is the most complicated wherein the executive routine must decide if the
solution for the current load step is valid (converged) and how to proceed next. For this purpose another Fortran
variable IEXIT is defined with one of four possible values as follows.

• IEXIT = 1 means entire problem is complete, normal exit from CANDE.

• IEXIT = 0 means current load step has converged, advance to next load step
• IEXIT = -1 means current load step has not converged, repeat load step.
• IEXIT = -2 means design-mode solution did not converge, repeat all load steps.

With the above understanding, Figure 10.2-1 shows a flow chart of the executive routine wherein the “ICOME
divisions” are clearly evident on the left. Within each ICOME division, there are two colored boxes, one representing
calls to pipe-type subroutines and one representing calls to solution-level subroutines. Both the pipe-type subroutines
and the solution-level subroutines are contained in Project Engine and will be discussed in turn.

10.3 Pipe-type Subroutines

Currently there are five pipe types in CANDE’s pipe library. Each pipe-type is associated with one or more subroutines
to accomplish the three basic functions, input, evaluating trial solution, and updating results. Table 10.3-1 identifies the
subroutines associated with each pipe type.

Table 10.3-1 Subroutines associated with each pipe-type

Pipe-type and material Main pipe-type Auxiliary subroutines called by main pipe-type
subroutine name subroutine

Corrugated aluminum alum emod: calculates nonlinear section properties.

emod: calculates nonlinear section properties.

Corrugated steel steel jmod: special model for slipping joints.
slpjnt: process input/output for slotted joints.
conmat: calculates nonlinear section properties.
Reinforced concrete concre conend: computes shear capacity, radial tension, and
crack width.
plasbro: calculates section properties of profile wall.
Plastic pipe plasti plsnbk: computes strains and local buckling losses.

Basic basic none

Each of the main pipe-type subroutines listed above performs the same three basic functions as dictated by the
executive routine via the variable ICOME. Table 10.3-2 lists the basic functions for a generic main pipe-type
subroutine.

Table 10.3-2 Tasks performed by main pipe-type subroutines for each ICOME value.

ICOME Tasks performed in main pipe-type subroutine

value
1 • Read input properties for pipe group
(input phase) • Compute initial linear-elastic section properties.
2 • Revise section properties based on current solution
(solution phase) • Determine if nonlinear pipe group has converged
• Send IEXIT value back to executive routine (-2,-1, or 0)
3 • Save and update pipe-group structural responses
(output phase) • Print safety evaluation for pipe group

10-2
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

Figure 10.2-1 Executive routine flow chart: input phase, solution phase, and output phase

10-3
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

10.4 Solution-level Subroutines

For any given problem, only one of the three solution levels is operative, that is, Level 1, Level 2 or Level3. The
closed-form Burns and Richard elasticity solution, Level 1, only employs one subroutine, named Burns. In contrast, the
finite element solutions, Level 2 and Level 3, employ numerous subroutines. Level 2 and Level 3 share the same
subroutines for the solution phase (ICOME = 2) and the output phase (ICOME = 3). For the input phase ( ICOME = 1),
Level 2 includes a separate set of subroutines to create an input file that is identical to a Level 3 input file. Hence, the
subroutines employed for the Level 3 input phase (ICOME = 1) are also used to read the internally created Level 2
input file. The above comments are clarified in the following tables and charts.

10.4.1 Elasticity solution Level 1

As stated above, the Level 1 solution employs one subroutine called Burns that is not shared by any other solution
level. The tasks performed by the Burns subroutine are shown in the table below in accordance with the calls from
executive routine via the parameter ICOME.

Table 10.4-1 Burns subroutine tasks as called

ICOME Tasks performed by Burns subroutine

value
1 Read system input variables; pipe diameter, soil density, number of load steps, soil
(input phase) stiffness and fill heights, interface condition.

2 Compute incremental trial solution for moments, thrusts and shear around the pipe and
(solution phase) form trial total solutions for processing by pipe-type subroutine.

3 No action. (Soil responses are not computed in Burns subroutine, only pipe responses are
(output phase) computed which are processed in the pipe-type subroutine)

10.4.2 Finite element solutions Level 2 and Level 3

Solution levels 2 and 3 share most subroutines in common as illustrated in the following finite element flow charts. For
clarity, the flow charts are broken down into the input phase (ICOME = 1), solution phase (ICOME = 2) and the output
phase (ICOME = 3)

Input phase (ICOME = 1) When the executive routine calls for the finite element input phase, the sequence of
subroutines shown in Figure 10.4.2-1 come into play. Note that the flow path shows a diversion path for Level 2
wherein one of the three canned-mesh subroutines are entered to create an input file identical to Level 3.

Solution phase (ICOME = 2) The sequence of subroutines that come into play for the solution phase is shown in
Figure 10.4.2-2. Note there is no distinction between Level 2 and Level 3, the flow path is the same for both.

Output phase (ICOME = 3). When the executive routine determines that the load step has converged, ICOME is set
equal to 3 to signal the printout of the solution. The sequence of subroutines for the output phase is shown in Figure
10.4.2-3. Again, there is no distinction between Level 2 and Level 3.

10-4
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

Figure 10.4.2-1 Subroutine flow chart for input phase of finite element solution Level 2 and Level 3.

Asterisk (*) means subroutine serves in more than one phase

10-5
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

Figure 10.4.2-2 Subroutine flow chart for solution phase using finite element method

Asterisk (*) means subroutine serves in more than one phase ex

10-6
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

Figure 10.4.2-3 Subroutine flow chart for output phase using finite element method

10.5 Extensions to CANDE-2007

As stated at the beginning of this appendix, modifications and extensions to CANDE’s capabilities are welcomed and
encouraged. Based on CANDE’s history, the three areas that are most likely to be extended are: (1)new pipe-type
model, (2) new soil model and (3) new canned mesh model. Brief guidance for undertaking each of these endeavors is
provided below.

10.5.1 New pipe-type model

The basic programming for installing a new pipe model is already included in CANDE-2007. The executive routine
contains the Fortran name NEWPIP in the list of recognized pipe-type names that may be specified in the input data file
on line A-2. In addition, the executive routine is programmed with calls to a subroutine called Newpip for the three
executive phases; input, solution and output (i.e., ICOME =1, 2, and 3).

The shell of subroutine Newpip along with its arguments is installed in Project Engine, therefore the researcher needs
only to provide the programming within the Newpip subroutine. The programming logic should follow the structure
outlined in Table 10.3-1, and the existing Basic subroutine, which is the simplest of the pipe types, is useful as a
programming guide.

10.5.2 New soil models

To install a new soil model, the subroutines Readm and Heroic located in Project Engine need to be modified.
Subroutine Readm is associated with the input phase wherein input parameters for all soil models are read and
processed. For the first input line D-1, common to all soil models, the new soil model should be identified as ITYP = 7
because six other models already exist. The remaining input lines, D-2, D-3, etcetera, are defined as needed to read and
process the input parameters associated with the new model. It is recommended to store the Fortran variables for the
soil model in a new Include statement (named common statement) inserted in subroutines Readm and your new
subroutine, say Newmud, called from subroutine Heroic as discussed next.

10-7
Chapter 10 – CANDE Analysis Source Code CANDE-2007 Solution Methods and Formulations

In the solution phase of subroutine Heroic (ICOME = 2), the appropriate soil-model subroutine is called to compute the
updated modulus values for the element’s constitutive matrix (see Chapter 3). Comment cards in subroutine Heroic
identify exactly where to insert the call to subroutine Newmud (ITYP = 7). Using the element’s current stress /strain
values as input (ST array), subroutine Newmud’s job is to update moduli values and assign values to the incremental
plane-strain constitutive matrix CP(3x3) stored in the include statement, materl.fi. See subroutine Duncan as an
illustration for programming and for signaling convergence using the simple logic, ICONDS = 1 or 0. Note the array
STHARD may be used as desired in subroutine Newmud.

10.5.3 New canned mesh for Level 2

Installing a new canned mesh requires a trivial modification to the executive routine, inserting a call to the new
subroutine, say Newcan, in subroutine Prep, and programming the Newcan subroutine. Input line A-2 contains the
input variable NPCAN, which is currently input as 1, 2, or 3, corresponding to the Level 2 choice of pipe mesh, box
mesh, or arch mesh. To expand NPCAN to 4 choices requires modifying line 443 in the executive routine to avoid the
error check.

As indicated in Figure 10.4.2-1, subroutine Prep directs the calling to the canned mesh subroutine based on the value of
NPCAN. Accordingly, a new call statement to subroutine Newcan needs to inserted in subroutine Prep immediately
following line 62

Programming subroutine Newcan may seem daunting, particularly after inspecting other canned mesh subroutines.
However, the fundamental goal is simple. Using physical input data describing the configuration shape being
addressed, Newcan generates the entire finite element input stream and writes it to a data file (unit 11) in the same
sequence and format as Level 3 input instructions. This means line C-1 (PREP), line C-2 (control), multiple lines C-3
(nodes), multiple lines C-4 (elements) and multiple lines C-5 (boundary conditions) are all written to unit 11. The
mesh/generation package (Figure 10.4.2-1) reads the data file on unit 11 just the same as if it were a batch input file on
unit 5.

10-8
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

11 CANDE GUI Source Code

The CANDE Graphical User Interface (GUI) source code is the topic of this chapter. While the source code itself is the
ultimate documentation of the program, this chapter provides a simple guideline for using the code.

11.1 Overview of CANDE GUI Architecture

The CANDE-2007 source program is written in modern Fortran and developed with the following:

ƒ Microsoft Visual Studio 2005, Version 8.0.50727.762 (SP.050727-7600)

ƒ GUI written in .NET C# programming language
ƒ Microsoft .NET Framework version 2.0.50727

The MVS framework permits grouping subroutines together to form projects and then grouping projects together to
form a whole solution. CANDE’s whole solution is composed of six projects described in the table below.

Table 11.1-1 CANDE-2007 projects contained in MVS solution

Files required to initiate the
Project Description project

cande1 This is the primary project and can be opened by

opening the cande1.sln project file. The primary
functions of this project are as follow:
ƒ Main CANDE GUI project that initiates
the CANDE Multiple document
interface (MDI)
ƒ All other projects are instantiated from
the main module Form1
ƒ Project also contains the input interface
and all of the input screen forms
ƒ Contains the help about screens and
auxiliary forms
CANDEBeamResults Contains the form that plots the beam results aaa_MeshGeom.xml
using a linear graphing tool. This project uses a aaa_BeamResults.xml
third party graphing tool, Graphics server. This
tool is distributable royalty free, but a license
must be purchased in order to develop with it.
CANDEGraphics This project drives all of the mesh plotting for aaa_MeshGeom.xml(required)
CANDE. The viewer permits viewing of the (aaa_MeshResults.xml)
mesh geometry without the results file. Last file is optional but must
be present to view analysis
results.

11-1
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

Files required to initiate the

Project Description project

CANDEImport This project provides support for importing Requires an external import
external files into CANDE. The files currently file of the type defined.
supported are:
ƒ MeshGeom.XML – the format for this
file is defined in the CANDE user
manual.
ƒ NASTRAN – The NASTRAN
commands supported by the CANDE
import are described in an appendix to
the CANDE user manual.
ƒ CANDE-89 – The CANDE-89 plot file
import is supported.

CANDEInputViewer This project defines the form needed for the Requires a CANDE input file
CANDE text input viewer. A Feature of this tool with an extension .CID.
is the column-sensitive help.
CANDENewInput This project controls the CANDE input wizard Generates a new CANDE
screens. This form is generally made to be run in input file.
modal mode.
CANDEOutputGenerator The output generator project initiates the aaa_MeshGeom.xml
windows to dynamically generate CANDE output aaa_MeshResults.xml
based on the data stored in the CANDE. The aaa_BeamResults.xml
program provides the user with choices to
customize the data stored in the three CANDE All three CANDE xml output
xml output files. The format of these files is files are required to run this
described in the CANDE user manual. module.
CANDEOutputViewer This project uses the CANDE table of contents CANDE output file
file (ctc) and CANDE output file to create an CANDE table of contents file
interactive browser for CANDE output files. (ctc extension).
If the CANDE table of contents file is not
present, the form will still work without a
browser input tree.
This form performs double duty as the viewer for
the standard CANDE output file and the file
produced by the CANDE output generator.
Utiltiy Classes
CANDEOutputClass Classes used to build output files for the CANDE
output generator.
AbanGeometry, General geometry classes
absngnrl
CANDE_InputData_ClassLibrary Group of classes that reads/write the CANDE
input file using the format described explicitly in
the CANDE user Manual (Chapter 5). The
classes are named according to the input names
defined in the user manual.
CANDE_XML_ClassLibrary Library of classes for reading the various XML
files used by CANDE.
RTBExCs Expanded rich text box control that supports
printing.
CANDE Setup Project that builds the CANDE installation.

11-2
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

11.2 CANDE Files

The CANDE-2007 GUI relies on a series of files to function properly; some are generated by the analysis engine, some
are soft-coded and some are produced by the GUI. This section provides a guideline for these files.

11.2.1 CANDE Input Definition

The CANDE input definition XML file provides information used by the GUI to check ranges, determine context-
sensitive help tags, short help descriptions, long help descriptions, and input formatting. The file is used primarily by
the GUI input system and is read in when the input system is initiated. Each input item defined in Chapter 5 of the
CANDE user manual has a <Data> entry in the CANDE input definition file. A sample of the XML file is shown
below.

<?xml version="1.0" encoding="UTF-8"?>

<CANDEInputDefinition>
.
.
.
<Data>
<className>CLRFDLoadFactors_E1</className>
<ShortInputName>E-1</ShortInputName>
<CANDEVariableName>FACTOR</CANDEVariableName>
<CANDEInputLine>E-1</CANDEInputLine>
<shortDescription>LRFD load factor applied to the load increments INCRS
through INCRL.</shortDescription>
<longDescription>FACTOR is the net load factor applied to the load steps
INCRS to INCRL (inclusive). It is the user’s responsibility to determine the appropriate
value of FACTOR that correlates to each load step. Table E-1 provides information on
load factors based on the</longDescription>
<inputColumnStart>11</inputColumnStart>
<inputLength>10</inputLength>
<inputFormat>F10.0</inputFormat>
<ouputFormat>F10.0</ouputFormat>
<unitsUS>.none.</unitsUS>
<unitsSI>.none.</unitsSI>
<defaultUS>1</defaultUS>
<defaultSI>1</defaultSI>
<lowerlimitUS_W>0</lowerlimitUS_W>
<upperLimitUS_W>10</upperLimitUS_W>
<lowerlimitSI_W>.none.</lowerlimitSI_W>
<upperLimitSI_W>.none.</upperLimitSI_W>
<lowerlimitUS_E>.none.</lowerlimitUS_E>
<upperLimitUS_E>.none.</upperLimitUS_E>
<lowerlimitSI_E>.none.</lowerlimitSI_E>
<upperLimitSI_E>.none.</upperLimitSI_E>
<HelpTag>5_7_1_E_1_LRFD_Net_Load_Factor_per_Load_step</HelpTag>
</Data>
.
.
.
<Data>
<className>CPTypeDef_AlumB2_A</className>
<ShortInputName>B-2.Alum.A</ShortInputName>
<CANDEVariableName>PA</CANDEVariableName>
<CANDEInputLine>B-2</CANDEInputLine>
<shortDescription>Area of pipe wall section per unit
length</shortDescription>
<longDescription>The cross-sectional area of one corrugation period divided
by the period length.</longDescription>
<inputColumnStart>1</inputColumnStart>

11-3
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

<inputLength>10</inputLength>
<inputFormat>F10.0</inputFormat>
<ouputFormat>F10.0</ouputFormat>
<unitsUS>in^2/in</unitsUS>
<unitsSI>mm^2/mm</unitsSI>
<defaultUS>.none.</defaultUS>
<defaultSI>.none.</defaultSI>
<lowerlimitUS_W>.001</lowerlimitUS_W>
<upperLimitUS_W>10</upperLimitUS_W>
<lowerlimitSI_W>.none.</lowerlimitSI_W>
<upperLimitSI_W>.none.</upperLimitSI_W>
<lowerlimitUS_E>.none.</lowerlimitUS_E>
<upperLimitUS_E>.none.</upperLimitUS_E>
<lowerlimitSI_E>.none.</lowerlimitSI_E>
<upperLimitSI_E>.none.</upperLimitSI_E>
<HelpTag>5_4_1_2_B_2_Aluminum_Analysis_Section_Properties</HelpTag>
</Data>
.
.
.
<CANDEInputDefinition>

CANDE Input Definition Tag Descriptions.

Tag Type Description
className string Class name as defined in CANDE.
shortInputName string Short input name description. This name is generally the
command name followed by the variable description.
CANDEVariableName string The CANDE variable name as defined in the CANDE user
manual.
CANDEInputLine string CANDE input line command (e.g. E-1) as defined in Chapter 5
of the CANDE user manual.
shortDescription string Short description of the input item.
longDescription string More detailed description of the input item.
inputColumnStart integer The column that the input item starts in the input file.
inputLength integer The total length of an input item as it is written to the input file.
inputFormat string Input format defined in Chapter 5 of the user manual for each
input item. The format in general follows FORTRAN guidelines
for formatting. Currently used for the persistent help shown at
the bottom of the input screen.
outputFormat (not currently used)
unitsUS string This value is used in labeling the US ‘units’ field on all CANDE
input menus
unitsSI string This value is used in labeling the SI ‘units’ field on all CANDE
(set by not currently input menus
used)
defaultUS string The value set for a CANDE input file when initially generated.
defaultSI (not currently used)
lowerlimitUS_W integer, double lower limit for CANDE input for this particular input item.
CANDE allows input with range violations but provides a
warning message.
upperlimitUS_W integer, double upper limit for CANDE input for this particular input item.
CANDE allows input with range violations but provides a
warning message.
lowerlimitSI_W (not currently used)
upperlimitSI_W (not currently used)
lowerlimitUS_E (not currently used)
upperlimitUS_E (not currently used)

11-4
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

Tag Type Description

lowerlimitSI_E (not currently used)
upperlimitSI_E (not currently used)
HelpTag string This tag is used by the GUI to set the help tag associated with the
CANDE-2007.chm help file. When CANDE opens and input
menu it uses this tag to search for context sensitive help.

11-5
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

11.2.2 CANDE analysis XML output files

The CANDE XML output files are described in detail in the CANDE User Manual and Guideline. Which modules use
the files is described in a table in section 11.1 of this manual.

11-6
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

11.2.3 CANDE table of contents files

The CANDE analysis engine and the CANDE output generate produce a table of contents file that is used in tandem
with the corresponding CANDE output files to create a ‘browsable’ output file. The format of the CANDE table of
contents file is simple and could be applied to other programs that produce ASCII text based output. The format is as
follows:

Format:

IIII|aaaa…
IIII|aaaa…
IIII|aaaa…

Where
IIII – the heading level to be used in the tree explorer
Aaaa… - represents the key text to be searched for when the tree item is clicked.

11-7
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

Perhaps the best way to demonstrate the use of this file is by example. The following is a snippet from a CANDE ctc
file and demonstrates how it gets translated to a tree browser.

1|MASTER CONTROL AND PIPE-TYPE DATA FOR PROBLEM # 1

1|REVIEW SYSTEM INPUT DATA
2|LEVEL-2 DATA FOR PIPE-SHAPED CANNED MESH
2|FINITE ELEMENT INPUT CONTROLS (PREP)
2|NODAL INPUT DATA TO GENERATE COORDINATES
2|NODE COORDINATES AS GENERATED FROM INPUT DATA
2|ALL ELEMENT DATA AS INPUT "I" AND GENERATED
2|FINAL LIST OF ALL NODE COORDINATES
2|BOUNDARY CONDITIONS AS GENERATED FROM INPUT
2|EXTENDED LEVEL-2: CHANGES TO CANNED MESH
2|BEAM-NODE SEQUENCE NUMBERS FOR EACH GROUP
2|MATERIAL DESCRIPTION FOR SOILS AND INTERFACE
1|SOLUTION OUTPUT RESULTS
2|FINITE ELEMENT OUTPUT FOR LOAD STEP 1
3|ALL NODAL DISPLACEMENTS FOR LOAD-STEP 1
3|ALL BEAM ELEMENT OUTPUT FOR LOAD-STEP 1
3|INTERFACE ELEMENT OUTPUT FOR LOAD-STEP 1
3|CONTINUUM ELEMENT OUTPUT FOR LOAD-STEP 1
3|PLASTIC, SERVICE-EVALUATION FOR GROUP 1, LOAD-STEP 1
4|STRUCTURAL RESPONSES OF PLASTIC-GROUP 1, LOAD STEP 1
4|STRAIN DIAGONOSTICS OF PLASTIC-GROUP 1, LOAD-STEP 1
4|LOCAL BUCKLING DIAGONOSTICS OF PLASTIC-GROUP 1, LOAD-STEP 1
4|ASSESSMENT SUMMARY PLASTIC-GROUP 1, LOAD-STEP 1

Text file
converted to
hierarchal tree
viewer

11-8
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

11.2.4 CANDE help files

CANDE comes with a predefined CANDE help file CANDE-2007.chm. This file was generated using Adobe
RoboHelp HTML 6, Build 99. The file was produced by taking the CANDE User Manual and Guideline Microsoft
Word document and importing the document into Robohelp for HTML 6. Once the document is imported a small
amount of cleanup is required to finalize the document.

To create the CANDE -2007 help file following the following steps:

ƒ Start up RoboHelp HTML

ƒ Create a new project through ‘File -> New Project’
ƒ Select the ‘Import’ tab
ƒ Select ‘Word Document’ as the import type
ƒ Browse to the CANDEUserManualAndGuideline.doc file
ƒ Import the document
ƒ Follow the import wizard using the following settings

11-9
Chapter 11 – CANDE GUI Source Code CANDE-2007 Solution Methods and Formulations

Once the project has been generated, select ‘New Window…’ by right-clicking on the ‘Windows’ icon in the RoboHelp
HTML explorer (see below).

This will create a new instance of the RoboHelp project that can then be opened. Once opened select the option to
‘Remember Window Size and Position’ (see below).

Close the window and select the ‘Generate’ option from the ‘File’ menu.

11-10