zyxwvutsrqpo
zyx
zyxw
zyxwvutsrqpon
zyxwvutsrqp
zyxwvutsrqp
The Relationship Between System Functions, Reliability and Dependent failures
Jayant Trewn
Industrial & Manufacturing Engineering
Wayne State University
Detroit, MI, 48202
Kai Yang
Industrial & Manufacturing Engineering
Wayne State University
Detroit, MI, 48202
hierarchical model proposed in ref. [5]. Section 4 will link
reliability into the model of system design and discuss the cost
of failure of functions of the system. Section 5 will discuss
dependent failure and its effect on system reliability. Section 6
is the conclusion of the paper.
ABSTRACT
This paper deals with the relationship between system design,
reliability and dependent failure. A hierarchical multi-level
system model is developed to characterize system design
mapping from the functional domain to the physical domain.
Based on this modeling framework, the reliability and cost of
reliability are linked to system functions, design structures and
dependent failures. The evaluation method for analyzing
product system designs from a reliability perspective is
proposed.
2. Engineering System Design
Engineering design involves a continuous interplay between
what we want to achieve (design objective) and how we want
to achieve it (physical solution). Many attempts have been
made to develop scientific principles to guide and evaluate the
design process [3][5]. Two notable schools of thought are Num
Suh’s axiomatic design principles and Hubka’s model for
technical systems.
1. Introduction
Onc of the fundamcntal inadcquacies o i traditional reliability
enginccring is its lack of reliability deployment and evaluation
strategy in the early concept design stage. Traditional reliability
focuses on assessment of reliability of systemdproducts based
on tcst data. Reliability assurance is often fcatured by the
cycles of build-test-evaluate and fix. This approach is usually
rime consuming and costly. Also, inherent in its delinition,
testing is after-the-fact, much in opposition to Dermng’s
philosophy of bringing quality improvements upstream.
2.1. Axiomatic Design
Dr. Nam Suh proposes the use of axioms as the pursued
scientific foundations of design [3]. In this approach, the
design is defined as the creation of synthesized solutions in the
form of products, processes or systems that satisfy perceived
needs through the mapping between the functional
requirements (FRs) in the functional domain and the design
parameters (DPs) in the physical domain. This is illustrated in
Figure 1.
Classical reliability evaluation of systems is also based on the
assumption that the components comprising the system work
independently and their failures have no mutual
interdependence [ 13. In reality, classifying failures as mutually
independent events does not hold true in most system designs.
A recent study in the automotive industry showed that, in some
situations, single component failures cause only 15% of system
failure 121. The remaining 85% of causes of system failure are
due to sub-system interactions, poor sub-system interface,
improper part installation and so on. These subsystem
interdependencies also affect the severity of the damage and the
cost to repair the damage.
Fig 1: Axiomatic Design Process
This paper is an attempt to develop a design evaluation tool
that will guide design engineers to evaluate reliability of
products in the conceptual design stage. The conceptual
design, or product system design is featured by the mapping of
the functional domain (product functions) into the physical
domain (actual physical entities) [3][4][5]. In this paper, the
multi level hierarchical (MLH) model proposed in ref. [5] is
used to map the functional requirements onto design
parameters, which is further mapped onto sub-system and
components specifications. Next reliability concepts are
introduced into this model to evaluate the soundness of system
design from the reliability perspective. In modeling reliability,
the failures of components can also be assumed to be mutually
dependent. By integrating failure dependency and multi-level
system models, the effects of dependent failure to system
reliability can also be evaluated. The cost of failures and their
effects to system performances will also be analyzed.
I
Functional Space ==> Physical Space
FR - Desired output
zy
DP - System which delivers desired output
Suh proposes the following 2 axioms as the universal principles
which any ‘good’ system design should satisfy.
Axiom 1:
requirements
Maintain the independence of finctional
It means that in a good design, the independence of functional
requirements is maintained.,
This paper will be subdivided into 6 sections. Section 2 will
discuss general models for engineering system design. Section
3 proposes a mathematical model to describe the multi level
Axiom 2: Minimize the information content of the design
4722
zyxwvutsrqpon
EO3l
zyxw
zyxwvuts
zyxwvutsrq
zyxwvutsrq
Figure 3. Mapping of Functional Requirements to
component structure
It means that among designs that satisfy axiom 1, the best
design 'is the one that has the minimum information content.
Here the information content is a measure of design
complexity. So the second axiom indicates that the design
simplicity should be pursued given the functional requirements
can be met.
I
Funchonal Space ==> Physical Space ==> Component Space
~
Based on the axiomatic design theory, there are two major
types of design vulnerabilities in most of engineering design
solution entities: coupling (non-independence) and complexity.
Intuitively, these conceptual design vulnerabilities will cause
the system reliability to decrease.
FR - Desired output DP System which delivers
~
2.2 Multi Layer Hierarchical System
I
As an independent development in the theory of engineering
design, Hubka defines a technical system as a multi layer
hierarchical (MLH) system at various levels of abstraction [ 5 ] .
The flow of customer requirements from the highest (most
abstract) level to the lowest (least abstract) level is achieved in
a conceptual design process. The layers of a MLH system are
at various levels of abstraction and are shown in figure 2.
Abstraction
Level
A
Highest
Functional Requirements Functional Structure
B
C
Comparing with Suh's system design model, Hubka's MLH
model resolves the abstraction in the design parameter level by
extending the design structure to the component structure level.
Specifically, in a design process, functional requirements are
transformed into design parameters, which are further
transformed into sub-systemskomponents or parts. The
components and parts structure is the most concrete stage of the
concept design process. This is shown in the 2-step mapping
illustrated in Figure 3.
Design Specifications - Organ
Structure
Lowest
It is author's belief that this 2-step mapping provides more
insights into the design process. There are many cases where
several DP's can be encapsulated onto a single component.
There are also cases where a single DP has to be delivered by
using several components. Therefore, the structure of
relationships among components is often different than that of
Dps. However, the subsystedcomponents structures are often
the final form of an engineering design. In reliability
perspective, it is the subsystedcomponent structure that
determines the reliability of the design. However, Hubka's two
step mapping model is only a descriptive model. In
comparison, Suh's axiomatic design model provided elaborate
analytical model and numerical design evaluation criteria.
SystedHardware model Component Structure
Hubka states that, "in the process of abstracting, the number of
specific systems covered by the model increases. Abstracting
in this sense, is a relative deterministic and analytical process,
it is a one to many mapping, and two different systems can
deliver the same model. At each transition, a designer has a
number of ways of fulfilling the requirements, and can choose
among them according to various criteria. Different designers
make different decisions in the various contexts of the sociotechnical systems (market, companies, etc.) which results in a
number of solutions offered for solving the same problem" [ 5 ] .
In this section, we will develop an analytical hierarchical
multilevel model to describe the 2 step mapping model
proposed by Hubka. This model will be used to develop a
relationship between the component structure and the
functional structure of designs. It is this relationship which
forms the basis for an evaluation criteria to evaluate the
systems ability to perform the intended functions. In effect,
functional reliability is defined as the ability of a part or
component to perform its intended functions, or on a system
level, the effect of the reliability of a part on the ability of the
system to perform its intended functions. As the system gets
more complex, the dependence of the performance of the
system on a complex structure of parts becomes more critical.
Hence, it is a measure of this complexity that determines the
ability of a concept design to reliably perform its functions.
Hubka states "It is generally accepted that technical systems of
lower degree of complexity find greater breadth of application
and versatile use." [5]
Terminology in terms of the MLH model can be stated as:
-
Function structures (as functional requirements) can be
realized by various organ structures
-
Organ structures (as design specifications) can be realized
from various component structures
desired output
3. Analytical Multi Level Hierarchical Model
Figure 2: A Multi Layer Hierarchical (MLH) system
Layer
CS -Component Structure
It is the challenge of evaluating these different and competing
system designs that satisfy the same functional requirements
that is the motivation to develop a MLH evaluation model.
Hubka's MLH model maps the design process from the
conceptualization at the customers' level in the form of
functional requirements, to the design formulization in the form
of design specifications (parameters), to the component level in
the form of component specification. This mapping takes the
design process from an abstract functional structure to a
concrete component structure. Figure 3 models this mapping
process.
3.1 The MLH mapping methodology
The multi level hierarchical mapping methodology is best
understood by the aid of a design example. Consider a system
design with 3 functions, 3 design parameters and 4
4723
zyxwvutsrqpon
zyx
3.3 Physical to component space mapping
components. The design process is split into two subprocesses. These are: the first process that relates the
functional requirements to the design process and the second
process that relates the design parameters to the component
structure.
Component structure construction is defined as the mapping
process between the design parameters in the design space to
the component structure in the component space. Figure 5
maps the relation between the design parameters and the
component structure.
3.2 Function to physical space mapping
Figure 5: Component Space
Design is defined as the mapping process between the
functional requirement in the functional space to the design
parameters in the design space to the components in the
component space. Figure 4 maps the relation between the
functional requirements to the design parameters.
Figure 4: Design Space
DP. is the jth design parameter
J
zyxwv
zyxwvutsr
zyxwvutsrq
CPk is the kth functional requirement
Let the component space be characterized by a vector {CS)
with k components.
FRi is the ith functional requirement
DP’s in the physical domain are characterized by vector {DP}
with j components.
zyx
DP. is the jth design parameter
J
Let there be i components represented by a set of independent
functional requirements.
Let [B] be the design matrix that maps the components to the
design parameters.
[FR) is the functional requirement vector.
DP’s in the functional domain are characterized by vector [ DP]
with j components.
(DP) = @ I ICPJ
Let [A] be the design matrix of rank (ij) that maps the design
parameters to the functional requirements.
bjk is the element of the matrix [B] that maps the kth
component to the jth design parameter . The element bjk is
binary, 0 = no relationship and 1 = related. It defines the
relationship between the jth design parameter and the kth
component.
Hence the relationship between design parameters and
functional requirements is:
{FRI = [AI {DPl
1 qj= (0>1>)1
(1)
[: 11 1
[B]= 0 1 0 1
In the case of figure 4, the design matrix is as follows:
Matrix
‘“’=I::I
[B] maps the relationship between
the
design
parameters to the component structure. For example:
bll = 1, denotes that component 1 maps (is related to) to design
parameter 1 .
bI2 = 0, denotes that component 2 is not related to design
parameter 2.
zyxwvutsrqp
Matrix [A] maps the relationship between the functional
requirements to the design parameters. For example:
3.4 Analytical multi level hierarchical modeling
a1 1 = 1, denotes that design parameter 1 maps (is related to) to
functional requirement 1.
(2)
In the case of figure 5, the design matrix is as follows:
is the element of the matrix [A] that maps the jth design
parameter to the ith functional requirement. The element
is
binary, where 0 = no relationship and 1 = related. It definesihe
relationship between the ith functional requirement and the jth
design parameter.
r1 0 11
{ b&=(0,1,)1
The concept design process is the exercise of choosing the right
set of components that conform to the design parameters that
satisfy the functional requirements.
AI2 = 0, denotes that design parameter 2 is not related to
functional requirement 2 .
4724
zyx
In reality, the customer relates to functional performance and
the components that provide those functions. The design
parameters are abstract to the customer/operator.
4. Reliability and Multi-Level Hierarchical Model
Since [A] and [B] matrices are relationship vectors relating
Functional, Design and Component spaces respectively through
binary elements (O,l), we hypothesize that
Reliability is usually defined as the probability that the product
perform its stated function over a stated period of time. One of
the most important tasks in reliability engineering is the
relationship between component reliability and system
reliability. For example, in a series system, system's reliability
is equal to the product of their components' reliability. In
reality, however, the failure of different components will cause
different kind of failure modes for the system, hence the degree
of damage and cost of rectifying the failure will also be
different. For example, both the failure of the battery in an
automobile and the failure of a piston rod will cause the car to
stop functioning. Clearly the damage of failure caused by
battery and the subsequent cost to repair is far less than that of
piston rod.
zyxwvutsrqp
zyxwvutsrqpo
zyxwvutsrqpon
zyx
zyxwvutsrqp
zyxwvutsrq
zyxwvutsrqponm
[: ;:] I'
The Resultant matrix [C] = [A] [B] is the relationship
between the functional requirements and the components and it
maps the relationship of each component of the system to the
functional requirements that its existence satisfies. The
operator ( 0 ) is a composite relational operator for binary
matrices.
The composite relation A
defined as:
o
B
AB = C = (cik) where ( Cik = (0,l))
In which
The hierarchical multi-level model for engineering design
outlined in section 3 provides a much more detailed description
about the relationship between system functions and
component structure. In this section, reliability model will be
established by integrating failure probabilities of components
to this hierarchical multilevel model. This model will enable us
to analyze the exact impact of failure of components on the
product and to assess the damage of the failure more precisely.
the matrix [C] where [C] is
(4)
4.1. Notations and Assumptions:
For proof refer to Appendix 1.
p k : Failure probability of component k, k=l, ...,n.
P(FR, ) : Probability that FRi can be delivered
Applying equation 4 to the example in section 3.2 and 3.3:
C=AoB
(5)
C p k : kth component, k=l, ...,n.
ccp,
: Cost of replacing component k.
c ,: Cost of losing Fri.
0 1 1.
c=o
1 l o 0 1 0 1
1 0 1 0
ECCpk: Expected
cost of failure due to the failure of kth
Component.
Re: System Reliability
ECRe: Expected cost of product reliability
It is assumed here, that all components will work
independently.
Matrix C translates to the relationship map as in figure 6.
zyxwvuts
4.2. Functional Delivery Reliability and System Reliability
Figure 6: Functional requirements to Component structure
From the definition of reliability, it is clear that the system
functions only if it delivers all its functions. Therefore:
R, =
fi P( FRi )
i=l
For each FRi, it is clear that:
P(FR, ) = fi (1 - pk)cik
(7)
k=l
Where Cik is the entry of C matrix in the ith row and kth
column. Clearly, Ck=l indicates that kth component will affect
the ith functional requirement and vice versa.
Therefore equation (7) is true since ERi can only be delivered if
all its relevant components work successfully. For example,
47 25
zyxwvutsrqponm
FRI (Figure 6) can only be delivered if CP1, CP3 and CP4 are
In reality, however, this assumption may not be true. There are
many cases where a single failure mode may cause several
components to fail simultaneously. Also, the failure of one
component may cause other component to failure.
Bhattacharya, Misra and Balaguruswamy, and Humphreys and
Jenkins have documented the dependent failure concept [l] [6]
~71.
all in working states.
Similarly:
In last example, clearly,
4
p(FRl
zyxwvutsrqpo
zyxwvutsrqponml
zyxwvu
nFR2)=n(1-pk)
5.1. Notations.
(9)
p(kl j ) : Probability of failure of component k given the
failure of component j, clearly, p(kl k ) = 1.
p k : independent failure probability for component k.
p k : : total probability of failure for component k.
k=l
Since all four components have to work properly in order to
deliver FR,and FR2.
For system reliability:
In last example, reliability of product equal to the product of
reliability of all four components.
Figure 7: Example of Dependent Failure
4.3. Cost of failure
When failure occurs, the product may lose some or all of its
functions. If the failure is due to the failure of the components,
these components may have to be repaired or replaced. In the
multi-level hierarchical model discussed earlier, each
component will affect the product function in different ways.
Some of the components may only affect some minor functions,
whereas, some of key component may affect all the functions of
the product. Also, different components have different
repairheplacement cost.
I
zyx
zyxwvutsrqpon
In this section, we will derive the expected cost of failure due
to the failures of components.
ECCpk= Expected cost of failure of component k
Dependent failure relationship
Functional relationship
Clearly, Equation (11) states that the expected cost of failure
due to component k is proportional to the component
replacement cost and the cost of losing relevant product
functions affected by the failure of component k. There are
several ways to reduce the expected cost of failure. First, it can
be reduced by reducing Pk, the failure probability of that
component. Secondly, it can be reduced by improving the
design such that the failure of the component will affect fewer
functions of the product.
By summing up the costs of failure of each component we can
get the expected cost of reliability for the product.
The function to component relationship matrix [C] is:
[Cl =
5. Dependent Failures and Engineering Design
In traditional reliability engineering, it is usually assumed that
component failures are independent events. This assumption
holds true for situations where failing components do not
induce effects (stress, load, etc.) on other components.
‘:=I
0 0 1 0
0 0 0 1
Figure 7 shows that in this design case example, the
relationship of components to the product functions are
relatively independent. Without dependent failures, the failure
of each component will only affect one or two product
functions. However, with dependent failures, the failure of
component 2 may also cause component 3 and 4 to fail, thus
affecting more functions and possibly damaging more
components.
zyxwv
47 26
zyxwvutsr
zyxwvutsrqp
zyxw
zy
[ 2 ] P. Palady, Failure Modes and Effects Analysis, PT
Publication Inc., 1995.
The cost of failure due to component k will be:
[3] N. Suh, The Principles ofDesign, Oxford University Press,
N.Y., 1990.
[4] G. Phal and W. Beitz, Engineering Design: A Systematic
Approach, Springer-Verlag, 1988.
j#k
[ 5 ] V. Hubka and W.E. Eder, Theory of Technical Systems,
Berlin, Springer-Verlag, 2ndEd., 1988.
[6] K.B. Misra and E. Balaguruswamy, “Reliability analysis of
k out of m: G systems with dependent failures,” Intl. Journal of
Systems Science, Vol. 7 , 1976, pp. 853-861.
For the example illustrated in Figure 7, if we assume that:
p(112)=0.5
p(312)zl.O
p(412)=0.8
[7] P. Humphreys and A.M. Jenkins, “Dependent failures
Developments,” Reliability Engineering and System Safety,
Vol. 34, No. 3, 1991, pp. 417-427.
Then, the expected cost of failure of component 2 is:
[SI R.E. Prather, Discrete Mathematical Structures for
Computer Science, Houghton Mifflin Company, Boston, 1976.
zyxwvutsrqpo
zyxwv
Appendix 1
Clearly, the failure of component 2 could cause the failure of
other components, thus incurring extra costs of replacing those
components and the cost of losing related product functions.
Functional Relationship Operators
In this paper, logical operations are based on Boolean algebra
and the theory of Di-graphs [SI. Notations for logical operators
are as follows:
For component 1, the cost of failure is:
o logical product
logical sum
V logical or
Clearly, component 2 is a very critical component from a
reliability perspective.
A logicaland
Operations performed on binary elements are based on Boolean
operations as follows:
The expected cost of system reliability is:
zyxwv
zyx
zyxwvutsrqpo
6. Conclusion
Define two relations:
Multi Level Hierarchical systems approach extends traditional
reliability analysis at a component level to the functional level.
This approach aids in estimating failure costs for concept
designs and hence competing concept designs can be evaluated
for functional dependence. The more complex a design, (Le.
designs with certain components being critical to providing
multiple functional requirements) the more the costs due to
non-performance of functions. In future research, the concept
of component degradation will be applied to the multi level
hierarchical approach and the analysis of dependent failure
component systems.
R from set U to V
{r,J= (0,1)}
SfromsetV t o w
{sk,=(O.l)]
Define a composite relation R o S from U to W
U
(R S)w a there exists v in V with U R v and v S w
Using Boolean algebra:
The composite relation R S has the matrix
RS = T = (t,,)
= (0,I)l
In which
7. References
n
tij = v (rik A skj)
[I] A. Bhattacharya, “Reliability Evaluation of Systems with
Dependent Failures,” Intl. Journal of Systems Science, Vol. 27,
No. 9, 1996, pp. 881-885.
k=l
Where (rjk= (0,l)) and
4727
{sk,
= (0,1)}