Numerical approximation of the classical solutions for plates

Matheus S. Carvalho1, Jorge C. Costa1

1
Dept. of Civil Engineering, Federal University of Sergipe
Rod Marechal Rondom, SN, 49.100-000, São Cristóvão, Sergipe, Brazil
matheuss020@gmail.com, jorgecostase@gmail.com

Abstract. This article proposes a solution, obtained by line adjustment, to the problem of bending of thin plates
with small deflections. The aim of this solution is to make the programming and manual calculation of bending
moments and deflections easier than classical method equations and calculation tables. The values to make the fit
was obtained by programing the Levy’s classical method in MATLAB® and the results were close to the values
given by Timoshenko e Goodier (1959). The line adjustment was made in MATLAB® and the shape function was
a polynomial equation of third degree with four variables. Furthermore, the values obtained numerically was
compared with the coefficients in Chust and Figueiredo’s (2014) calculations table. It was verified that the
tabulated values were not the maximums because they were always obtained in the middle of the plate. Therefore,
a new calculation table for each kind of plate with the maximum values is proposed along with closed form
approximations. The analyses were carried out for rectangular plates with clamped or simply supported edges.

Keywords: Classical theory of plates, thin rectangular plates, calculation tables, bending moments, deflections.

1 Introduction
Thin plates are bidimensional structures with one of the dimensions, the thickness, significantly smaller than
the two others. Usually, the ratio between the two smallest dimensions of a plate is no larger than 1/10. As such,
bidimensional plate theories are sufficient to analyze small deflections caused by loading, as long as the deflections
remain small.
The classical methods were the first developed to analyze plate bending, especially with the advent of
reinforced concrete slabs. They are based on algebra and differential calculus to describe the mechanics and
provide satisfactory analytical solutions for most usual plate shapes and loadings.
Nevertheless, those methods provide solutions in the form of trigonometrical series and, for that reason, both
manual calculations and computer programming are not straightforward or easily accomplished. This resulted in
many calculation tables being created to summarize the values obtained by these methods in order to simplify
hand-made analysis. These calculation tables became the standard practice [1] for concrete slab design, at least
before the dissemination of computer-based solutions such as matrix solution of grids or finite elements.
Early efforts to make these calculations automatic would look towards an algorithm based on searching these
tables. Besides being inefficient and unpractical, it would imply a rework since they already come from a
simplification and numerical solution.
In this context, there are presently two options for slab analysis: manual calculations with the help of tables,
or numerical methods such as finite elements. Hence, this work proposes an intermediary solution based on
classical methods that can be used on hand calculations and easily programmable. This way, this solution can be
inserted in automated analysis routines of slabs with no need to resort to tables or numerical methods.
In this work, italic letters (𝑎, 𝑏, 𝛼, 𝛽, … ) represent scalar. Especially, 𝑤 will always stand for plate deflection,
𝐷 for plate stiffness in many forms, 𝑞 for the load, 𝜈 for Poisson’s ratio, 𝑎, 𝑏 for the plate larger dimensions, ℎ for
thickness and 𝐸 for the Young modulus of concrete. A Cartesian reference system will be used with the plate’s
midsurface in the 𝑥 × 𝑦 plane and thickness in the 𝑧 direction.

CILAMCE 2020
Proceedings of the XLI Ibero-Latin-American Congress on Computational Methods in Engineering, ABMEC
Foz do Iguaçu/PR, Brazil, November 16-19, 2020
Numerical approximation of the classical solutions for plates

2 Classical solutions
Efforts to mathematically model the behavior of plates remount to the works of Leonhard Euler (1707-1783),
Jacques Bernoulli (1759-1789), Ernst Chladni (1756–1827), Sophie Germain (1776-1831), Joseph-Louis Lagrange
(1736-1813), Siméon Poisson (1781-1840) and others [2, 3, 4].
The first complete model for plate bending can be attributed to Navier in 1820. This solution consisted of
using a double Fourier trigonometric series to approximate the load imposed on the plate and consequently its
transversal displacement. Latter, Maurice Levy (1839-1910) introduced as a solution a single trigonometric Fourier
series for plates with two opposing sides simply supported and the remaining sides either free or also simply
supported [2].
The nomenclature for the plates will follow Reddy [5] and is represented in Fig.1. Also, the 𝑥 axis will always
be parallel to the smaller side, with dimension 𝑎, and the 𝑦 axis will be parallel to the larger dimension 𝑏. We
emphasize that the ratio between the plate sizes will be represented by 𝜆 = 𝑏/𝑎 and that the plates will be analyzed
under uniform distributed load 𝑞𝑜 , the most common situation for the structural analysis of civil engineering
structures.

Figure 1. Plate nomenclature.

2.1 Navier’s Method

According to Timoshenko and Goodier [6], Navier, analyzing the plate equilibrium represented by the
differential equation
∂4 𝑤 ∂4 𝑤 ∂4 𝑤
𝐷( + 2 + ) = 𝑞, (1)
∂𝑥 4 ∂ ∂𝑥 2 ∂𝑦 2 ∂𝑦 4
noticed that this equation admits a relatively trivial solution for trigonometric loads such as
𝑚𝜋𝑥 𝑛𝜋𝑦 𝑚𝜋𝑥 𝑛𝜋𝑦
𝑞𝑚𝑛 = 𝑎𝑚𝑛 𝑠𝑒𝑛 𝑠𝑒𝑛 → 𝑤𝑚𝑛 = 𝑏𝑚𝑛 𝑠𝑒𝑛 𝑠𝑒𝑛 . (2)
𝑎 𝑏 𝑎 𝑏
Consequently, if a loading can be approximated by a series of 𝑞𝑚𝑛 , the displacements would result in another
series.
∞ ∞ ∞ ∞
𝑚𝜋𝑥 𝑛𝜋𝑦 𝑚𝜋𝑥 𝑛𝜋𝑦
𝑞 = ∑ ∑ 𝑎𝑚𝑛 𝑠𝑒𝑛 𝑠𝑒𝑛 → 𝑤 = ∑ ∑ 𝑏𝑚𝑛 𝑠𝑒𝑛 𝑠𝑒𝑛 (3)
𝑎 𝑏 𝑎 𝑏
𝑚=1 𝑛=1 𝑚=1 𝑛=1

In the case of SSSS plates such as depicted in Fig.2, the boundary conditions are
𝜕2𝑤
𝑤 = 0; = 0, for 𝑥 = 0 and 𝑥 = 𝑎,
𝜕𝑥 2
(4)
𝜕 2𝑤
𝑤 = 0; = 0, for 𝑦 = 0 and 𝑦 = 𝑏.
𝜕𝑦 2

CILAMCE 2020
Proceedings of the XLI Ibero-Latin-American Congress on Computational Methods in Engineering, ABMEC
Foz do Iguaçu/PR, Brazil, November 16-19, 2020
 M. S. Carvalho, J. C. Costa

Figure 2. Navier’s SSSS Plate.

A uniform load 𝑞𝑜 can be approximated by

∞ ∞
16𝑞𝑜 𝑚𝜋𝑥 𝑛𝜋𝑦
𝑞= ∑ ∑ 2 𝑠𝑒𝑛 𝑠𝑒𝑛 , (5)
𝜋 𝑚𝑛 𝑎 𝑏
𝑚=1 𝑛=1

so that the deflection is obtained as

∞ ∞ 𝑚𝜋𝑥 𝑛𝜋𝑦
16𝑞0 𝑠𝑒𝑛 𝑠𝑒𝑛
𝑎 𝑏
𝑤= 6 ∑ ∑ . (6)
𝜋 𝐷 𝑚 2 𝑛 2
2
𝑚=1 𝑛=1 𝑚𝑛 ( 2 + 2 )
𝑎 𝑏

Other quantities such as moments and shear stresses can be obtained from the deflection.

2.2 Lévy’s Method

According to Timoshenko and Goodier [6], Levy proposed a solution in the form of
𝑤 = 𝑤1 (𝑥) + 𝑤2 (𝑥, 𝑦). (7)
The 𝑤1 part is inspired by the deflection of a beam in the 𝑥 direction in such a way as to comply with the
boundary conditions on the larger sides
𝜕 2 𝑤1
𝑤1 = 0 ; = 0 for 𝑥 = 0 and 𝑥 = 𝑎 (8)
𝜕𝑥 2
and the non-homogeneous part of the equilibrium equation
∂4 𝑤1 ∂4 𝑤1 ∂4 𝑤1 (9)
𝐷( + 2 + ) = 𝑞.
∂𝑥 4 ∂𝑥 2 ∂𝑦 2 ∂𝑦 4
Hence, for a uniform load 𝑞𝑜 and simply supported longer sides,
𝑞𝑜
𝑤1 = (𝑥 4 − 2𝑎𝑥 3 + 𝑎3 𝑥). (10)
24𝐷
Function 𝑤2 must obey a homogeneous equilibrium eq. (13) and also the boundary conditions on the longer
sides. Moreover, the sum (7) must obey all the boundary conditions
𝜕 2 𝑤2
𝑤2 = 0 ; = 0, for 𝑥 = 0 and 𝑥 = 𝑎 (11)
𝜕𝑥 2

𝜕 2 𝑤 𝜕 2 𝑤2 𝜕 2 𝑤1
𝑤 = 𝑤1 + 𝑤2 = 0; = + = 0, for 𝑦 = 0 and 𝑦 = 𝑏 (12)
𝜕𝑦 2 𝜕𝑦 2 𝜕𝑦 2

CILAMCE 2020
Proceedings of the XLI Ibero-Latin-American Congress on Computational Methods in Engineering, ABMEC
Foz do Iguaçu/PR, Brazil, November 16-19, 2020
Numerical approximation of the classical solutions for plates

∂4 𝑤2 ∂4 𝑤2 ∂4 𝑤2
𝐷( + 2 + )=0 (13)
∂𝑥 4 ∂𝑥 2 ∂𝑦 2 ∂𝑦 4
In this manner, 𝑤2 is proposed as a series (14) where 𝑌𝑚 is only dependent on 𝑦
∞
𝑚𝜋𝑥
𝑤2 = ∑ 𝑌𝑚 (𝑦) 𝑠𝑖𝑛 . (14)
𝑎
𝑚=1

A. Nadai proposes the 𝑌𝑚 part of the solution as a hyperbolic trigonometric function

𝑞𝑎4 𝑚𝜋𝑦 𝑚𝜋𝑦 𝑚𝜋𝑦 𝑚𝜋𝑦 𝑚𝜋𝑦 𝑚𝜋𝑦
𝑌𝑚 = (𝐴𝑚 𝑐𝑜𝑠ℎ + 𝐵𝑚 𝑠𝑖𝑛ℎ + 𝐶𝑚 𝑠𝑖𝑛ℎ + 𝐷𝑚 𝑐𝑜𝑠ℎ ), (15)
𝐷 𝑎 𝑎 𝑎 𝑎 𝑎 𝑎
where the constants 𝐴𝑚 , 𝐵𝑚 , 𝐶𝑚 , 𝐷𝑚 are chosen as to obey the support conditions. For an SSSS plate 𝐶𝑚 , 𝐷𝑚 are
zeroes as symmetry along the 𝑥 axis imposes that only even functions are kept. The imposition of the other
boundary conditions yields
2( 𝛼𝑚 𝑡𝑎𝑛ℎ 𝛼𝑚 + 2) 2
𝐴𝑚 = − ; 𝐵𝑚 = . (16)
𝜋 5 𝑚5 𝑐𝑜𝑠ℎ 𝛼𝑚 𝜋 5 𝑚5 𝑐𝑜𝑠ℎ 𝛼𝑚
Lastly, we arrive at eq. (17), that represents the deflection of an SSSS plate with uniformly distributed load.
∞
4𝑞𝑜 𝑎4 1 ( 𝛼𝑚 𝑡𝑎𝑛ℎ 𝛼𝑚 + 2) 2 𝛼𝑚 𝑦
𝑤 = 5 ∑ 5 (1 − 𝑐𝑜𝑠ℎ
𝜋 𝐷 𝑚 2 𝑐𝑜𝑠ℎ 𝛼𝑚 𝑏
𝑚=1 (17)
𝛼𝑚 2𝑦 2 𝛼𝑚 𝑦 𝑚𝜋𝑥
+ 𝑠𝑒𝑛ℎ ) 𝑠𝑒𝑛
2 𝑐𝑜𝑠ℎ 𝛼𝑚 𝑏 𝑏 𝑎

2.3 Calculation Tables

The main calculation tables used for the analysis of concrete slabs are those by Bares (1970) with several
values for Poisson’s ratio; by Kalmanok (1961) with 𝜈 = 0; and, in Brazil, those by Chust e Figueiredo Filho
(2014) with 𝜈 = 0,2 [7].
The tables present coefficients 𝛼,µ𝑥 , µ′𝑥 , µ𝑦 and µ′𝑦 used to find the maximum bending moments and
deflections through the expressions. The subscript 𝑥, 𝑦 indicates the direction of the fibers being stressed by the
moment and 𝑀′ is the moment at a clamped boundary.
100 𝑀𝑥
µ𝑥 or µ′𝑥 = , (18)
𝑞0 𝑎2

100 𝑀𝑦
µ𝑦 or µ′𝑦 = , (19)
𝑞0 𝑎2

100 𝑤𝐸ℎ3
𝛼= . (20)
𝑞0 𝑎4
For an SSSS plate with uniform load, Lévy’s solution in eq. (17) renders a maximum transversal displacement
of
∞
4𝑞𝑜 𝑎4 (−1)(𝑚−1)⁄2 ( 𝛼𝑚 𝑡𝑎𝑛ℎ 𝛼𝑚 + 2)
𝑤= 5 ∑ 5 (1 − ). (21)
𝜋 𝐷 𝑚 2 𝑐𝑜𝑠ℎ 𝛼𝑚
𝑚=1

Using the definition Eq.(20) of 𝛼, 𝜆 = 𝑏⁄𝑎 and 𝛼𝑚 = 𝑚𝜋𝜆⁄2, we arrive at

∞
48(1 − 𝜈 2 ) (−1)(𝑚−1)⁄2 ( 𝛼𝑚 𝑡𝑎𝑛ℎ 𝛼𝑚 + 2)
𝛼(𝜆) = 5 ∑ 5 (1 − ). (22)
𝜋 𝑚 2 𝑐𝑜𝑠ℎ 𝛼𝑚
𝑚=1

CILAMCE 2020
Proceedings of the XLI Ibero-Latin-American Congress on Computational Methods in Engineering, ABMEC
Foz do Iguaçu/PR, Brazil, November 16-19, 2020
 M. S. Carvalho, J. C. Costa

3 Comparison of the Methods

One approach to comparing the presented methods is through a convergence analysis. Fig.4 depicts the error
in displacement for both Navier and Lévy solutions with 1, 2 or 3 terms in the series in respect to the exact
(converged) solution for an SSSS plate.
It can be seem that Lévy’s solution converge faster than Navier’s, as results with 3 terms are already
indistinguishable from the reference solution for all the 𝜆 analyzed.

Figure 3. Displacement error for Lévy’s and Navier’s approximation with 1, 2 and 3 terms

Fig.5 depicts the first term of the series used to approximate the uniform load. It can be noted that Lévy’s
solution present an excellent approximation for the load even with only one term, whereas Navier’s solution do
not approximate the load in a satisfactory manner.

Figure 4. Load representation for the first term of Lévy’s and Navier’s solutions.

As such, we can verify what is stated by Timoshenko e Goodier [6], that Levy’s method is more convenient for
computer implementation.

4 Curve Fitting
The equations were approximated using the Curve Fitting Toolbox available in MATLAB®. The methods to
evaluate the quality of a fitting are of two kinds. The first are the graphic methods as residue analysis that help on
visual interpretation. The second are the numerical methods as quality statistic data and the confidence interval
that provide numerical evidence to help the statistical reasoning. The referred toolbox [8] provide data such as the
sum of squared errors (SSE), R-square, adjusted R-square and the root mean squared error as depicted in Fig.6.

CILAMCE 2020
Proceedings of the XLI Ibero-Latin-American Congress on Computational Methods in Engineering, ABMEC
Foz do Iguaçu/PR, Brazil, November 16-19, 2020
Numerical approximation of the classical solutions for plates

Figure 5. Goodness-of-Fit Statistics

5 Results and discussion

5.1 Linear Fitting

Linear fittings were developed as to serve as an alternative to the calculation tables and as to make easier the
programming of the values. The objective is to provide a way to obtain the 𝛼 and 𝜇 parameters with no need to
calculate Levy’s series or resort to calculation tables. The fittings were carried such as to achieve a good quality
in the results with the least number of parameters.
μ = 𝑎𝜆3 + 𝑏𝜆2 + 𝑐𝜆 + 𝑑 (23)
A visual inspection of the curves for coefficients α, µ𝑥 , µ′𝑥 , µ𝑦 and µ′𝑦 motivated the use of 4-parameter 3rd -
degree polynomial function approximation as in Eq.(22). In this equation, 𝜇 stand for any of the desired parameters.
Other approximating functions were considered, such as rational and trigonometrical approximations,
nevertheless, the polynomial provided better results. The fitting was carried with 500 points for each parameter in
the interval 1 < 𝜆 < 2 and a confidence interval of 95%. Poisson’s ratio was taken as 𝜈 = 0,2.
As for the goodness of fit, the values of R-square and Adjusted R-square remain close or equal to 1, what
points to a good fit. Moreover, the values of SSE and RMSE remain close to 0, also an indication of a good fit.
The observed residues were also close to null, with a random distribution characteristic of a suitable approximation
function. The resulting parameters for eq.(23) are reported in Tab.1.

6 Conclusions
It is remarked that the equations obtained through the fitting are a good alternative to calculate maximum
deflections and bending moments on concrete slabs. These equations simplify the classical methods and make
easier for both manual and automated calculations. Moreover, these equations are easier to be programmed than
calculation tables and, for manual calculations, they preclude the use of interpolations and the large number of
tables necessary to cover every kind of plate.
This work can be further developed through comparative analyses with other numerical and algebraic
methods available in the literature. It is also subject to a better analysis of the fitting procedures, through the use
of other approximating functions or mathematical analysis of the series solution of the differential equation.

Acknowledgements. The authors acknowledge the Federal University of Sergipe for the funding of this research.

Authorship statement. The authors hereby confirm that they are the sole liable persons responsible for the
authorship of this work, and that all material that has been herein included as part of the present paper is either the
property (and authorship) of the authors, or has the permission of the owners to be included here.

CILAMCE 2020
Proceedings of the XLI Ibero-Latin-American Congress on Computational Methods in Engineering, ABMEC
Foz do Iguaçu/PR, Brazil, November 16-19, 2020
 M. S. Carvalho, J. C. Costa

Table 1. Fitted coefficients

1<𝜆<2
Plate supports µ∗ = 𝑎𝜆3 + 𝑏𝜆2 + 𝑐𝜆 + 𝑑 ∞
a B c d
α -0,3048 -1,513 13,66 -7,169 15,00
SSSS µx -0,0236 -2,4 12,94 -6,1 12,50
µy 1,968 -9,346 13,68 -1,872 3,83
α -2,47 10,46 -6,574 0,7825 15,00
µx -1,872 7,449 -2,946 -0,4958 12,50
CCSS
µy 1,217 -7,695 15,34 -5,704 3,35
µy' -1,22 9,674 -25,41 9,986 -12,50
α 0,7025 -4,127 8,263 -2,626 3,00
µx 0,9655 -5,624 11,14 -3,306 4,18
SSCC
µx' -1,743 9,867 -18,84 3,719 -8,33
µy -0,932 4,86 -8,459 6,672 1,75
α -1,492 4,698 3,835 -3,759 15,00
µx -0,9674 2,411 5,653 -3,927 12,50
CSSS
µy 1,873 -9,944 16,75 -4,777 3,64
µy' -2,127 13,34 -28,91 9,306 -12,50
α 0,8021 -5,609 13,75 -5,657 6,20
µx 0,9215 -6,322 15,22 -5,926 7,06
SSSC
µx' -1,921 12,1 -26,59 8,02 -12,50
µy 0,05056 0,6529 -3,038 5,536 1,41
α 0,5094 -3,676 8,928 -4,311 3,00
µx 0,8005 -5,537 12,98 -6,136 4,18
CCCC µx' -1,779 11,44 -25,04 10,25 -8,33
µy 1,423 -6,025 7,491 -0,7616 1,49
µy' -1,759 9,006 -15,26 2,858 -5,70
α 0,3966 -3,934 12,14 -6,041 6,20
µx 0,5398 -4,792 13,94 -6,606 7,02
CSCS µx' -1,405 10,46 -26,59 10,77 -12,50
µy 1,117 -4,931 6,507 0,4269 2,55
µy' -2,173 11,38 -19,79 3,793 -7,65
*Where µ may be α, µx, µx', µy or µy'

References
[1] B. Mosley, J. Bungey and R. Hulse. Reinforced concrete design to Eurocode 2. New York: Palgrave Macmillan, 2007.
[2] R. Szilard. Theory and Analysis of Plates: classical and numerical methods. New Jersey: Prentice-Hall Inc. New Jersey:
John Wiley & Sons, 2004.
[3] S. P. Timoshenko. History of Strength of Materials. Mcgraw-Hill Book Company, New York, 1953.
[4] M. C. Santos. Uma Revisão dos Métodos Clássicos de Análise de Placas. Graduation dissertation [in Portuguese]. Federal
University of Sergipe, São Cristóvão, 2020.
[5] J. N. Reddy. Theory and Analysis of Elastic Plates and Shells. 2.ed. New York, 2007.
[6] S. P. Timoshenko and J. N. Goodier. Theory of Plate and Shells, Mcgraw-Hill Book Company. 2 ed. Singapure, 1959.
[7] J. M. Araújo. Curso de concreto armado – volume 2. 2. ed. [in Portuguese]. Rio Grande do Sul: Dunas, 2010.
[8] MathWorks®. Evaluating Goodness of Fit. How to Evaluate Goodness of Fit. Disponível em:
<https://www.mathworks.com/help/curvefit/evaluating-goodness-of-fit.html>. Accessed on 05 February 2020.

CILAMCE 2020
Proceedings of the XLI Ibero-Latin-American Congress on Computational Methods in Engineering, ABMEC
Foz do Iguaçu/PR, Brazil, November 16-19, 2020