Sunteți pe pagina 1din 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260524748

Reliability Analysis for Multi-Component Systems Subject to Multiple


Dependent Competing Failure Processes

Article  in  IEEE Transactions on Reliability · March 2014


DOI: 10.1109/TR.2014.2299693

CITATIONS READS

95 908

4 authors, including:

Sanling Song David W. Coit


Rutgers, The State University of New Jersey Rutgers, The State University of New Jersey
13 PUBLICATIONS   226 CITATIONS    213 PUBLICATIONS   7,804 CITATIONS   

SEE PROFILE SEE PROFILE

Qianmei Feng
University of Houston
66 PUBLICATIONS   1,205 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

data mining View project

Complex System Reliability Optimization View project

All content following this page was uploaded by Sanling Song on 08 June 2016.

The user has requested enhancement of the downloaded file.


IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014 331

Reliability Analysis for Multi-Component Systems


Subject to Multiple Dependent Competing
Failure Processes
Sanling Song, David W. Coit, Qianmei Feng, and Hao Peng

Abstract—For complex multi-component systems with each NOTATION


component experiencing multiple failure processes due to simul-
taneous exposure to degradation and shock loads, we developed
a new multi-component system reliability model, and applied two number of shock loads that have arrived by time
different preventive maintenance policies. This new model extends
previous research, and is different from related previous research number of components in a series or parallel
by considering an assembled system of degrading components
with -dependent failure times resulting from shared shock expo- system
sure. Previous research primarily pertained to a single component arrival rate of random shocks
or simple system, or systems with -independent failure processes
and failure times. In our new system model, the individual failure threshold for catastrophic/hard failure of th
processes for each component and the component failure processes component
are all -dependent. These models can be applied directly, or cus-
tomized for many complex systems with multiple components size or magnitude of the th shock load on the
that experience -dependent competing failure processes. In this th component
model, each component can fail due to a soft failure process, or a cumulative distribution function (cdf) of
hard failure process. These two component failure processes are
mutually competing and -dependent. If one component fails rela- critical wear degradation failure threshold of the
tively frequently, it is likely that the number of shocks is relatively th component
large, and these shocks impact all components potentially causing
them to fail more often as well. Therefore, failure processes of all wear volume of the th component due to
components are also -dependent. An age replacement policy and continuous degradation at
an inspection-based maintenance policy are applied for a system
with multiple components. The optimal replacement interval total wear volume of the th component at due
or inspection times are determined by minimizing a cost rate to both continual wear and instantaneous damage
function. The model is demonstrated on several examples.
damage size contributing to soft failure of the th
component caused by the th shock load
Index Terms—Multiple dependent competing failure processes cumulative shock damage size of the th
(MDCFP), multi-component systems, degradation, random component at
shocks, preventive maintenance.
cumulative distribution function (cdf) of
cdf of
ACRONYMS AND ABBREVIATIONS
pdf of the sum of independent and identically
MDCFP multiple dependent competing failure processes
distributed (i.i.d.) variables
MEMS micro-electromechanical systems
cdf of the failure time,
pdf probability density function
pdf of the failure time,
periodic replacement interval
cumulative maintenance cost by time
Manuscript received September 04, 2011; revised July 05, 2012; accepted average long-run maintenance cost rate of the
July 26, 2013. Date of publication January 24, 2014; date of current version first policy
February 27, 2014. This work was supported by USA National Science Foun-
dation (NSF) under Grants CMMI-0970140 and CMMI-0969423. Associate expected value of the renewal cycle length, of
Editor: J. Shortle.
the first policy
S. Song and D. W. Coit are with the Industrial and Systems Engineering De-
partment, Rutgers University, Piscataway, NJ 08854 USA. expected value of the number of failures, in a
Q. Feng is with the Industrial Engineering Department, University of
Houston, Houston, TX 77004USA.
renewal cycle
H. Peng is with the Department of Industrial Engineering and Innovation Sci- expected value of the total maintenance cost of
ences, Eindhoven University of Technology, Eindhoven, The Netherlands.
Digital Object Identifier 10.1109/TR.2014.2299693
the renewal cycle,

0018-9529 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
332 IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014

replacement cost per unit component failure processes and failure times. If one compo-
nent fails relatively frequently, it is more likely that the number
cost of replacement caused by failure
of shocks is relatively large, and these shocks impact all com-
periodic inspection interval ponents, potentially causing them to fail more often as well.
average long-run maintenance cost rate of the When components of a system fail, they do not necessarily
second policy fail independently of each other. The failure could be syn-
chronized, for example, by an extreme environment, such as
expected value of the renewal cycle length, of
the failure of an external piece of hardware, or human error.
the second policy
In such cases, the synchronized failures can be attributed to
expected value of the number of inspections a common cause [4]. This problem shares similarities to the
expected value of system downtime (i.e., time research problem addressed in this paper because each shock
from a system failure to the next inspection when increases the risk exposure to all components simultaneously.
the failure is detected) Different maintenance strategies for degrading systems
with a single unit or multiple units have been extensively
cost associated with each inspection;
examined in the literature [5]. Based on the reliability analysis
penalty cost rate during downtime; of a system with multiple components, each exposed to two
competing -dependent failure processes, we consider two dif-
ferent maintenance policies: 1) an age replacement policy with
I. INTRODUCTION fixed replacement interval, and 2) a fixed inspection interval
maintenance policy. We demonstrate the developed reliability

T HERE are many complex system design and optimization model and maintenance policies for multi-component systems
problems featuring combined exposure to degradation subject to MDCFP using several representative examples.
and shocks impacting each component, and resulting in -cor- The paper is organized as follows. Section 2 introduces re-
related or -dependent component failure processes. Many lated previous research and presents basic information about
traditional approaches to reliability are inadequate or inappro- competing failure processes. Section 3 analyzes the reliability
priate for more complex system design problems when there are of the complex system experiencing MDCFP with degradation
two or more -dependent failure processes and multiple com- and shocks. Section 4 explains that the failure processes of all
ponents subject to failure. The -dependency among the failure components are -dependent. Section 5 describes the age re-
processes and among the component failure times presents placement maintenance modeling and optimization based on the
challenging issues in reliability modeling. In this paper, a new replacement interval variable, and an inspection-based mainte-
system reliability model is presented offering distinct advan- nance modeling and optimization based on the inspection in-
tages compared to other models for multi-component systems terval variable. Multi-component system examples are shown
experiencing both degradation and shocks. in Section 6 to illustrate the developed reliability and mainte-
For many engineering design problems, traditional failure- nance models.
based reliability methods present difficulties due to the lack of
available failure data. When there are multiple failure processes, II. LITERATURE REVIEW, AND BACKGROUND
and particularly when those failure processes are in the same
time scale, the failure processes for an individual component There has been significant, meaningful prior research that
are competing and -dependent. This competition creates a chal- addresses degradation modeling with shocks, multi-component
lenging problem, which is to analyze and predict the system re- degradation modeling, and maintenance policies used for degra-
liability performance. As an example, the reliability analysis of dation models. In our new system model, we extend and com-
micro-electromechanical systems (MEMS) is an evolving, chal- bine previously developed models and research to develop a
lenging analysis task. MEMSs are complex design systems ex- new result.
periencing -dependent component failure processes and mul-
A. Literature on Reliability and Maintenance
tiple dependent competing failure processes (MDCFP) for each
component. Previous research and studies on degradation have focused on
In this paper, we develop a system reliability model and two establishing degradation models and estimating time-to-failure
associated maintenance policies in which each component expe- distributions [6]–[12], using experimental design to improve
riences soft and hard failure processes [1], [2]. These two failure reliability [13]–[16], and developing maintenance policies for
processes are mutually competing and -dependent. The funda- continuously monitored degrading systems [17]–[19]. Lu &
mental difference between this paper and [2] is that this paper Meeker [7], and Meeker et al. [20] developed general statis-
considers systems with multiple components whereas [2] con- tical models to estimate the time-to-failure distribution from
siders systems with only one component. This is a significant degradation measures. System reliability models for complex
extension because, when we update our model from one compo- multi-component systems subject to MDCFP were developed
nent to multi-components, failure processes of components can by Song et al. [21], but limited to series systems with age re-
be correlated or -dependent due to shared exposure to shocks. placement. Li et al. [22] developed a reliability model of a series
The same shock processes contribute to both failure processes system with -correlated or -dependent component degrada-
for all components [3], and this condition results in -dependent tion processes based on an additive degradation process. Jiang
SONG et al.: RELIABILITY ANALYSIS FOR MULTI-COMPONENT SYSTEMS SUBJECT TO MULTIPLE DEPENDENT COMPETING FAILURE PROCESSES 333

et al. [23] presented reliability analyses for -dependent failure


processes and a -dependent failure threshold.
Kharoufeh & Cox [24] presented a degradation-based pro-
cedure for the estimation of full and residual lifetime distribu-
tions for single-unit systems using real sensor data. Li and Pham
[25] developed a reliability model pertaining to three -indepen-
dent failure processes, two degradation processes, and a random
shock process. They also describe specific applications of sys-
tems exposed to multiple failure processes. Applications can
be found in the space shuttle computer complex due to critical
mission phases such as boost, reentry, and landing; and in the
electric generator power systems due to the loss of commercial
power systems.
A simultaneous quality and reliability optimization model
for systems composed of degrading components has also been
studied [26]. For a complex system involving both degradation
and shocks, -independent multiple catastrophic and degrada-
tion failure processes have been investigated [27]. However, rel-
atively little research has been devoted to the reliability analysis
of multi-component systems with MDCFP.
For systems experiencing degradation processes, both an age Fig. 1. Two -dependent competing failure processes for component (a) soft
replacement policy and an inspection-based maintenance are failure process, and (b) hard failure process [2].
often used as policies to minimize costs associated with system
failure and maintenance. Numerous different maintenance ap-
proaches for degrading systems have been described in the liter- B. Introduction of Two Failure Processes
ature [5], [28]. Jiang & Jardine [29] showed the effectiveness of In this paper, system reliability is considered when there are
a jointly applied burn-in and preventive replacement policy for multiple components arranged according to some defined archi-
situations where the failure time follows a mixture distribution. tecture and every component within the system can fail due to
Lu [3] developed a condition-based maintenance model for a two competing failure processes: a hard failure process, and a
continuous degradation process by considering imperfect main- soft failure process. Fig. 1 parts (a) and (b) depict the two failure
tenance. A maintenance cost model for determining the optimal processes for a single component. In practice, one of the two
inspection schedule and replacement threshold for a single unit failure processes may be dominant for some of the components,
degrading system has been developed by Grall et al. [31]. but the models presented in this paper are most useful and ap-
There have been other interesting extensions and models de- plicable when both failure processes are non-trivial for the ma-
veloped for specific problems. Wang and Pham [32] developed a jority of the components within the system. This section follows
model for a degraded system subject to cumulative effects from a parallel development, and succinctly summarizes the develop-
random shocks and degradation with two kinds of path func- ment in [2]. The main difference is that we now consider mul-
tions, including additive and multiplicative. Wang and Pham tiple component system which then requires an additional sub-
[33] further developed a -dependent competing risk model for script to differentiate components. Assumptions used to model
systems subject to multiple degradation processes and random soft and hard failure processes are as follows.
shocks using time-varying copulas. Xiang et al. [34] developed 1. When the overall degradation of a component is greater
a comprehensive model when the population is not homoge- than , soft failure occurs. The total degradation of each
neous, but is instead composed of sub-populations. Rafiee, component is accumulated by both continuous degradation
et al. [35] studied an interesting problem when the component over time and abrupt damage due to random shocks.
degradation rate changes in response to the occurrence of dif- 2. When the shock load itself exceeds the maximum strength
ferent patterns of shocks. of the components , that component experiences a hard
The model in this paper aggregates continual degradation failure.
and shock behavior together to be applied to a system design 3. Random shocks arrive according to a Poisson process.
problem with -dependent failure processes. Alternatively, Peng The assumptions in this model or similar ones have been
et al. [36] defined a joint probability density function (pdf) to made by many researchers (e.g., [2], [7], [18], [25], [32], [33],
model the component degradation rates, conditional on some [38]), and have been demonstrated to be applicable for specific
defined time . This is a fundamentally different model, with rel- applications (e.g., [1], [26]). However, these assumptions will
ative advantages and disadvantages depending on the design or not necessarily be valid for all applications. Each different en-
planning problem. The joint conditional pdf has fundamentally gineering analysis and problem requires detailed consideration
sound and attractive theoretical properties and behavior. How- of the required assumptions.
ever, that model does not accommodate shocks, and the most Component hard failures occur when the shock load exceeds
important issue is that it requires determination of the functional the maximum strength of the components [1], [2]. Fig. 1 part
form and estimation of associated parameters of the joint pdf. (b) shows that a component may fail when a load or stress
334 IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014

magnitude exceeds . The probability that the th component For component , the probability that the total degradation at
survives the applied stress from the th shock is shown in (1). time is less than some specific , can be derived
As an example, if are assumed to be i.i.d. normally dis- as (4).
tributed random variables, , then the prob- Furthermore, if we consider to be the cdf of
ability that each component survives a shock becomes [2] at to be the pdf of the sum of i.i.d. variables,
then the cdf of in (4) can be derived [2] as (5).
(2) Equation (5) is a generalized function for the soft failure
process, which can be applied for various problems with
where is the cdf of a standard normal random variable. different distributional assumptions. Here we apply it to two
Soft failures of the th component occur when the overall special cases.
degradation is beyond a threshold level [1], [2], as shown First, if the soft failure shock damage for the th component
in Fig. 1 part (a). The total degradation is the sum of a con- by the th shock is normally distributed, ,
tinual degradation process and the cumulative abrupt degra- the degradation path is linear with a constant initial value
dation shifts due to the shocks, and it is given as and a normally-distributed degradation rate with
. In our model, a linear degradation path is as- , and shocks follow a Poisson process
sumed: . However, our model is general such with rate , then a more specific model can be determined
that other kinds of degradation paths can be used, including an based on (5) [2]: see (6).
exponential degradation path. The cumulative damage size due Second, if the soft failure shock damage for the th compo-
to random shocks until time , is given as nent, , follows an exponential distribution with rate , then
the summation of for shocks follows a gamma distribution
with shape parameter and scale parameter . With everything
(3) else as before, now becomes (7).
The probability that component does not experience soft
failure before time is expressed as
When the system experiences a shock, all components are im-
pacted, and thus, the component failure times are -dependent
due to the shared shock exposure. (8)

(1)

(4)

(5)

(6)

(7)
SONG et al.: RELIABILITY ANALYSIS FOR MULTI-COMPONENT SYSTEMS SUBJECT TO MULTIPLE DEPENDENT COMPETING FAILURE PROCESSES 335

III. SYSTEM RELIABILITY ANALYSIS FOR MDCFP


WITH DEGRADATION AND SHOCKS
Systems can be designed with multiple components in many
different configurations or architectures. In our models, we con- Fig. 2. Series system example.
sider systems where each individual component may fail due to
two competing -dependent failure modes [1], [2]. The compo-
nent fails when either of the two competing failure modes oc- nent survives each of the shock loads, and the total degra-
curs. The system failure definition is entirely dependent on the dation of each component is less than the threshold level. Thus,
particular design configuration. we have (9).
Specific assumptions used for the system reliability and main- Each shock to the system impacts both the hard and soft
tenance modeling in this paper are as follows. failure processes of all components. The component failure pro-
1. The model is for systems that are packaged and sealed cesses and failure times are -dependent only because of the
together, making it impossible or impractical to repair or shared number of shocks. In this model, are in-
replace individual components within the system, e.g., dependent, and are independent. This is a strong
MEMS. assumption that is appropriate for some example systems, but is
2. For the age replacement policy, the system is preventively not satisfied for other problems. For systems where the compo-
replaced at a fixed age. However, if the system fails before nents are physically or otherwise isolated, this assumption may
the specified age, it is replaced correctively immediately. very well be appropriate and valid. Furthermore, if data analyses
Replacements are assumed to be instantaneous and perfect, reveal that any correlations are negligible, then this new model
which is reasonable for highly reliable systems. can be appropriate to use. Song et al. [6] presented an approx-
3. For the second maintenance policy, the system is inspected imation model when and are -dependent for a single
at periodic intervals. If the system fails before the specified component.
inspection interval, it is not replaced until the next inspec- The component survival events (in the hard brackets) are
tion. There is a penalty cost associated with failures of the not independent because they are all impacted by a common
system during downtime, e.g., cost associated with loss of random variable . However, if we condition on the event
production or opportunity costs. that , and apply the law of total probability, then
There are different system design configurations that can be the equation can be more readily expressed. Using conditional
considered in our model. Example system configurations that probability and summing the terms for all possible numbers of
are described in this section are series, parallel, and series-par- shocks, we have (10).
allel. In all cases for the model formulation presented here, the In this model, shocks arriving at random time intervals are
component failure processes are s-dependent due to the shared modeled as a Poisson process with rate . Component failure
exposure to shocks, as described in Section 4.0. times become conditionally independent for a fixed number of
shocks: see (11).
A. Series Systems Based on (2) and (5), the generalized reliability equation for
Fig. 2 shows a series system with components. The relia- a multi-component system suffering MDCFP can be expressed
bility of this system at time is the probability that each compo- as (12).

(9)

(10)
336 IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014

The reliability function for the specific case with a linear


degradation path, and normally distributed , and , can
be expressed as (13).
As another example simply to demonstrate that this is a
general model, the reliability function with exponentially
distributed (with parameter ), exponentially distributed
(with parameter ), and normally distributed , can be
expressed as (14).
We can also express (14) in the following way [37]. First, we Fig. 3. Parallel system example.
introduce a conditional system reliability term using classical
results from reliability computation, which can be defined as
(15). level . The system fails when all components
Then, reliability can be succinctly expressed as experience either soft failure or catastrophic failure.
System reliability for a parallel system is given by the fol-
(16) lowing equations for the two specific cases introduced in the
previous section, see (17)–(18a).
A similar expression could be developed and used for any Similar to the discussion for series systems, a conditional
system structure. term could be defined as (18b).

B. Parallel Systems C. Series-Parallel Systems


Fig. 3 shows a parallel system composed of components. Fig. 4 depicts a series-parallel system made up of subsys-
The reliability of the parallel system at time is the probability tems. is the set of components in subsystem with no compo-
that at least one component of this system survives each of the nent being used in more than one system ( , for all
shock loads ( for ), and the total ), and each subsystem has components with . For
degradation of that same component is less than the threshold the example depicted in the figure, , and

(11)

(12)

(13)

(14)

(15)
SONG et al.: RELIABILITY ANALYSIS FOR MULTI-COMPONENT SYSTEMS SUBJECT TO MULTIPLE DEPENDENT COMPETING FAILURE PROCESSES 337

, and , and ; For a multi-component system, the number of shocks


and , and . has an effect on each component. When is sufficiently
The reliability of a series-parallel system at time is the large, the sum of the shock damage size contributing to soft
probability that at least one component within each sub-system failure for each component is large, and there are also greater
survives each of the shock loads ( for opportunities for hard failure; thus, a failure is more likely for
), and the total degradation is less than the all components. Alternatively, when there are relatively few
threshold level for that same component. The shocks, times to failure are relatively longer for all components.
system fails when all components for at least one parallel Thus, the component failure processes are -dependent. We
subsystem experience either soft failure or catastrophic failure. prove the -dependency of soft failure processes for all compo-
System reliability for a series-parallel system is given by the nents in this section. The dependence in this paper is captured
following equations for the two specific cases previously intro- due to a common number of shocks, . Alternatively,
duced as (19) and (20). dependence due to the size and impact of each shock on all
components is not considered in this paper.
If the covariance of two events is greater than zero, then the
IV. DEPENDENT FAILURE PROCESSES AMONG COMPONENTS
occurrences of these two events are correlated, and they are
In this new model, the components within the system have -dependent (but not necessarily physically dependent). With
-dependent soft and hard failure processes as in [2], but now this idea, we can show that soft failure processes for all compo-
there are multiple components, and the failure processes are nents are -dependent. We have the covariance of total degra-
-dependent among components. This represents a funda- dation of component and component as (21).
mental extension and advancement because previous research For component , and this equation
and models were either limited to an individual component or applies for all components, see (22).
it was assumed that the failure processes were independent. Because the value of the first three items are zero, we can
We demonstrate this extension by deriving the covariance of obtain (23).
the degradation for any two components, and proving that it is As an example, consider that are i.i.d random variables
greater than zero given the presence of shocks. Furthermore, that follow a normal distribution. Then we can observe that, see
we define events for component survival from hard failure, and (24).
demonstrate that the covariance of these events for any two Similar covariance derivations were done for hard failure
components is positive. events. If we define event no hard failure of component ,

(17)

(18a)

(18b)

(19)

(20)

(21)
338 IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014

size for the th component’s soft failure process from shocks,


, impact the relative -dependency among failure processes
of all components.

V. MAINTENANCE MODELING AND OPTIMIZATION


There are different applicable maintenance strategies for de-
grading systems that have been studied, and are selectively ef-
fective for particular planning problems. In this paper, we con-
sider two different maintenance policies. First, we consider an
Fig. 4. Series-parallel system example.
age replacement policy with a fixed replacement interval for a
system with multiple components each exposed to two com-
we can prove that the covariance of events and is greater peting -dependent failure processes. Then, a fixed inspection
than zero for two components and . interval maintenance policy is considered for the same system.
If we define , and ,
the covariance can be shown to be (25) (derivation included in A. Maintenance Policy With Replacement Interval Variable
the Appendix). To evaluate the performance of the age replacement policy
Because both the soft failure process and hard failure process with a replacement interval, we consider an average long-run
for all components are -dependent, it can be concluded that maintenance cost rate model in which the periodic replacement
component survival events are also -dependent. Sensitivity interval or the fixed age is the decision variable. In the model
analysis is later performed for one example in Section 6 and it at age , the entire assembled system is replaced preventively
indicates that the arrival rate of shocks, , and mean damage with a new one. However, if the system fails before age , it is

(22)

(23)

(24)

(25)
SONG et al.: RELIABILITY ANALYSIS FOR MULTI-COMPONENT SYSTEMS SUBJECT TO MULTIPLE DEPENDENT COMPETING FAILURE PROCESSES 339

replaced correctively. The replacement interval restarts if failure rate model is also used, in which the periodic inspection interval
occurs. The average long-run total maintenance cost per unit is the decision variable. At intervals of time , the system is
time can be evaluated by (26), where is the total mainte- inspected. If the system fails before time , it is not replaced
nance cost of a renewal cycle, and is the length of time until until the next inspection. If the system is still operating satisfac-
replacement [38]. Then the expected total maintenance cost is torily with no failed components, nothing is done. The average
given as (27). Defining as failure density function, the ex- long-run total maintenance cost per unit time can be evaluated
pected time between two replacements, or the expected cycle by (30), where is the total maintenance cost of a renewal
length, is cycle, and is the length of a cycle that takes a value of a mul-
tiple of . Then the expected total maintenance cost is given as
(28)
(31)
Based on (25) through (27), the average long-run mainte-
nance cost rate as a function of is given as The expected value of the number of inspections is

(29)
(32)

For optimization problems with a single decision variable, a


golden section search method can be used to search for the min- The expected value of system downtime or the expected time
imum of the function on a given interval. It is a technique for from a system failure to the next inspection when the failure is
finding the extremum (minimum or maximum) of a unimodal detected is (33).
function by successively narrowing the range of values inside The expected time between two replacements or expected
which the extremum is known to exist. Given the convergence cycle length is (34).
conditions, we can obtain the approximate optimal solutions ef- Based on (30) through (34), the average long-run mainte-
ficiently. nance cost rate as a function of is given as (35).
Although the cost rate is complex, it is not difficult to find an
B. Maintenance Policy With Inspection Interval Variable optimal inspection interval because there is only one decision
The second model is a maintenance policy with an inspection variable, and it can be determined by a golden section search,
interval. For this model, an average long-run maintenance cost or a nonlinear programming search.

(26)

(27)

(30)

(33)

(34)
340 IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014

VI. NUMERICAL EXAMPLE


Three examples have been analyzed to study the reliability
of different system configurations and to investigate how com-
ponents in a system interact. The first two examples are series
systems with four and six components respectively, and the third
example system is a parallel system.

A. Reliability Analysis for Series System With Four


Components
In the first example, we consider a series system with four
components. For this example, and follow normal dis-
tributions. Without loss of generality, we assume that compo-
nents’ 1 and 2 parameters are the same, and components’ 3 and
4 parameters are the same.
The example is a hypothetical system for demonstration pur-
poses, but , and parameters are esti-
mated based on test data for MEMS [3]. The reliability function
is plotted in Fig. 5. Also, in Fig. 5, we present system relia-
bility for an analogous system with identical component failure
processes but independent component failure processes. In this Fig. 5. Plot of reliability function of a series system with four compo-
nents.
way, it is possible to observe one of the meaningful contribu-
tions from this new model, which is to explicitly consider com-
ponents with -dependent failure processes. As indicated in the
figure, an incorrect component independence assumption would
provide invalid results.
Figs. 6 through 8 presents sensitivity analyses of or or
both on . From Figs. 6 through 8, we can see that decreased
or decreased or both increases the reliability of both the
-dependent component system and the independent component
system as expected. For the comparisons, the decreased shock
arrival rate is 10 times less frequent, and the mean shock damage
is 30% of the previous amount for all components. When the
damage to soft failure by shocks and shock arrival rate are
decreased, the reliability for the -dependent component system
get closer to the reliability for the similar independent compo-
nent system. This is also as expected because the covariance is
smaller.
Sensitivity analysis was also performed on the ratio of
system reliability with -dependent and independent com-
ponent failure processes, and the results are presented in
Fig. 9. System reliability was computed at ,
and compared with the analogous system with independent Fig. 6. Sensitivity analysis of on .
components. The parameters we are most interested in are
the arrival rate of shocks and mean damage size for the th
component soft failure process from shocks, . The ratio linearly as increase. That means, if the shock arrival rate is
increases approximately higher, and its effect on system reliability is more pronounced,

(35)
SONG et al.: RELIABILITY ANALYSIS FOR MULTI-COMPONENT SYSTEMS SUBJECT TO MULTIPLE DEPENDENT COMPETING FAILURE PROCESSES 341

The second maintenance policy is to inspect at intervals of


and to replace the system when it is observed to have failed.
Choosing , and , we can
find the minimum average long-run maintenance cost rate of
$0.242/cycle, which is obtained at for the
-dependent component system, as shown in Fig. 11.

B. Reliability and Maintenance for Series System With Six


Components
For the second example, we study the reliability of a series
system with six components, and compare the results with the
previous results for the four component system. For this ex-
ample, we choose the age replacement policy.
Three components of the system use the data from the
first column in Table I, and the other three components use
the second column of data. The reliability function is
presented in Fig. 12, and compared to an analogous system
with independent components. The maintenance policy is to
replace the system at a fixed interval . Choosing
and , we find the minimum average long-run main-
Fig. 7. Sensitivity analysis of on . tenance cost rate of /cycle, which is obtained at
. Fig. 13 illustrates as a function of
. Compared with the system of four components, the optimal
replacement interval is approximately the same, but the cost
rate is larger, because there are more components.

C. Reliability and Maintenance for Parallel System


For the third example, we study the reliability of a parallel
system with three components. For this example, we choose the
age replacement policy for the system.
All three components of the system use the first column data
in Table I. The reliability function is presented in Fig. 14,
and compared to an analogous system with independent compo-
nents. It can be observed that, for parallel systems, the analogous
independent component system has a higher reliability. This is
the expected result because, if several components have already
failed, it is more likely that the remaining components are also
closer to the failure threshold when they are -dependent, so the
benefits of redundancy are diminished.
The maintenance policy is to replace the system at a fixed
interval . Choosing and , we can find
Fig. 8. Sensitivity analysis of and on . the minimum average long-run maintenance cost rate of
/cycle, which is obtained at . Fig. 15
illustrates as a function of . The cost rate for this case
the difference between and is is much lower than the cost rate of the series systems.
larger, and the -dependency is more notable.
Also, increases as in- VII. CONCLUSION
creases, but not linearly. When the shock damage size contribu- We develop the system reliability, and two different main-
tion to soft failure processes increases, system reliability is af- tenance models for multi-component systems that experience
fected more intensely by each shock. By (23), we already know MDCFP. The reliability model is developed based on general
that the covariance of the total degradation of different compo- degradation path and random shock models for individual com-
nents is larger, which means the relative -dependency is larger. ponents. This work represents a new system model extending
The first maintenance policy is to replace the system at a fixed component-level degradation modeling concepts to the system-
interval . Choosing and , we can level. We demonstrated that the component failure processes
find the minimum average long-run maintenance cost rate of are -dependent, which represents a fundamentally new system
/cycle, which is obtained at , model. The average long-run maintenance cost rate is evalu-
the optimal replacement interval for -dependent component ated and optimized with two different maintenance policies. For
system. Fig. 10 illustrates as a function of . the age replacement policy, the periodic replacement interval
342 IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014

Fig. 9. Ratio of with -dependent and independent failure processes versus and .

Fig. 10. Cost rate versus replacement interval for -dependent component
system. Fig. 12. Plot of reliability function of series system with six components.

Fig. 11. Cost rate versus inspection interval for -dependent component
system.
Fig. 13. Cost rate versus replacement interval for -dependent component
system.
is the decision variable, while for the preventive maintenance
policy, the periodic inspection interval is the decision variable.
We present three numerical examples to demonstrate and ob- inspection maintenance policy for three different system con-
serve the reliability, the age replacement policy, and periodic figurations.
SONG et al.: RELIABILITY ANALYSIS FOR MULTI-COMPONENT SYSTEMS SUBJECT TO MULTIPLE DEPENDENT COMPETING FAILURE PROCESSES 343

TABLE I
PARAMETER VALUES FOR MULTI-COMPONENT SYSTEM RELIABILITY ANALYSIS

APPENDIX

DEPENDENT HARD FAILURE PROCESSES FOR ALL COMPONENTS


Fig. 14. Plot of reliability function of parallel system with three
For some time , define event no hard failure of com- components.
ponent event no hard failure of component , and
, Furthermore, we define (36) at the bottom
of the page. Applying the law of total probability, we can obtain Consider shocks follow Poisson process with rate , we ob-
tain (38). We can prove easily that for
(37). We want to prove ,
and , and we can find that
that is , or that .

(36)

(37)

(38)
344 IEEE TRANSACTIONS ON RELIABILITY, VOL. 63, NO. 1, MARCH 2014

[15] H. F. Yu and C. H. Chiao, “An optimal designed degradation experi-


ment for reliability improvement,” IEEE Trans. Rel., vol. 51, no. 4, pp.
427–433, Dec. 2002.
[16] V. R. Joseph and I. T. Yu, “Reliability improvement experiments with
degradation data,” IEEE Trans. Rel., vol. 55, no. 1, pp. 149–157, Mar.
2006.
[17] S. Lu, Y. C. Tu, and H. Lu, “Predictive condition-based maintenance
for continuously deteriorating systems,” Quality Rel. Eng. Int.erna-
tional, vol. 23, no. 1, pp. 71–81, Jan. 2007.
[18] H. Liao, E. A. Elsayed, and L. Y. Chan, “Maintenance of continuously
deteriorating systems,” Eur. J. Operational Res., vol. 175, no. 2, pp.
821–835, May 2005.
[19] H. Lu, W. J. Kolarik, and S. S. Lu, “Real-time performance reliability
prediction,” IEEE Trans. Rel., vol. 50, no. 4, pp. 353–357, Dec. 2001.
[20] W. Q. Meeker, L. A. Escobar, and C. J. Lu, “Accelerated degradation
tests: Modeling and analysis,” Technometrics, vol. 40, no. 2, pp. 89–99,
1998.
[21] S. Song, D. W. Coit, Q. Feng, and H. Peng, “Reliability estimation &
preventive maintenance for complex multicomponent systems subject
to multiple dependent competing failure processes,” in Proc. 7th Int.
Conference on Math. Methods in Re. (MMR), June 2011.
Fig. 15. Cost rate versus replacement interval for -dependent component [22] J. Li, D. W. Coit, and E. A. Elsayed, “Reliability modeling of a se-
system. ries system with correlated or dependent component degradation pro-
cesses,” in Proc. Int. Conf. Quality, Rel., Risk, Maintenance Safety Eng.
(ICQR2MSE), Jun. 2011.
[23] L. Jiang, Q. Feng, and D. W. Coit, “Reliability and maintenance mod-
eling for dependent competing failure processes with shifting failure
can similarly be demonstrated to be positive, con- thresholds,” IEEE Trans. Rel., vol. 61, no. 4, pp. 932–948, Dec. 2012.
sidering that , and [24] J. P. Kharoufeh and S. M. Cox, “Stochastic models for degradation
. based reliability,” IIE Trans., vol. 37, no. 6, pp. 533–542, Feb. 2007.
[25] W. J. Li and H. Pham, “Reliability modeling of multi-state degraded
systems with multi-competing failures and random shocks,” IEEE
REFERENCES Trans. Rel., vol. 54, no. 2, pp. 297–303, June 2005.
[26] H. Peng, Q. Feng, and D. W. Coit, “Simultaneous quality and reliability
[1] Q. Feng and D. W. Coit, “Reliability analysis for multiple dependent optimization for microengines subject to degradation,” IEEE Trans.
failure processes: An MEMS application,” Int. J. Perform. Eng., vol. Rel., vol. 58, no. 1, pp. 98–105, Mar. 2009.
6, no. 1, pp. 100–102, Jan. 2010. [27] W. Huang and R. G. Askin, “Reliability analysis of electronic devices
[2] H. Peng, Q. Feng, and D. W. Coit, “Reliability and maintenance mod- with multiple competing failure modes involving performance aging
eling for systems subject to multiple dependent competing failure pro- degradation,” Quality Rel. Eng. Int.ernational, vol. 19, no. 3, pp.
cesses,” IIE Trans., vol. 43, no. 1, pp. 12–22, Apr. 2010. 241–254, Feb. 2003.
[3] D. M. Tanner and M. T. Dugger, “Wear mechanisms in a reliability [28] J. M. Van Noortwijk, “A survey of the application of gamma processes
methodology,” in Proc. SPIE 4980, Rel, Testing, Characterization of in maintenance,” Rel. Eng. Syst. Safety, vol. 94, no. 1, pp. 2–21, Jan.
MEMS/MOEMS II, 2003, pp. 22–40. 2009.
[4] C. L. Atwood, “The binomial failure rate common cause model,” Tech- [29] R. Jiang and A. K. S. Jardine, “An optimal burn-in preventive-replace-
nometrics, vol. 28, no. 2, pp. 139–148, 1986. ment model associated with a mixture distribution,” Quality Rel. Eng.
[5] H. Wang, “A survey of maintenance policies of deteriorating systems,” Int., vol. 23, no. 1, pp. 83–93, Feb. 2007.
Eur. opean J. Operational Res., vol. 139, no. 3, pp. 469–489, Jun. 2002. [30] S. Lu, Y. Chen, and H. Lu, “Predictive condition-based maintenance
[6] S. Song, D. W. Coit, Q. Feng, and H. Peng, “Reliability and preven- for continuously deteriorating systems,” Quality Rel. Eng., vol. 23, no.
tive maintenance for a component subject to competing failure pro- 1, pp. 71–81, Feb. 2007.
cesses with dependent shock damage sizes,” in Proc. Ind. Syst. Eng. [31] A. Grall, L. Dieulle, C. Berenguer, and M. Roussignol, “Continuous
Res. Conf., May 2012. time predictive-maintenance scheduling for a deteriorating system,”
[7] C. J. Lu and W. Q. Meeker, “Using degradation measures to estimate IEEE Trans. Rel., vol. 51, no. 2, pp. 141–150, Jun. 2002.
a time-to-failure distribution,” Technometrics, vol. 35, no. 2, pp. [32] Y. Wang and H. Pham, “Imperfect preventive maintenance policies
161–174, 1993. for two-process cumulative damage model of degradation and random
[8] W. Q. Meeker, L. A. Escobar, and C. J. Lu, “Accelerated degradation shocks,” Int. J. Syst. Assurance Eng. Manag., vol. 2, no. 1, pp. 66–77,
tests: Modeling and analysis,” Technometrics, vol. 40, no. 2, pp. 89–99, Mar. 2011.
1998. [33] Y. Wang and H. Pham, “Modeling the dependent competing risks with
[9] C. J. Lu, J. Park, and Q. Yang, “Statistical inference of a time-to-failure multiple degradation processes and random shock using timevarying
distribution derived from linear degradation data,” Technometrics, vol. copulas,” IEEE Trans. Rel., vol. 61, no. 1, pp. 13–22, Mar. 2012.
39, no. 4, pp. 391–400, 1997. [34] Y. Xiang, D. W. Coit, and Q. Feng, “ subpopulations experiencing
[10] S. Song, D. W. Coit, and Q. Feng, Determination of On-Condition stochastic degradation: Reliability modeling, burn-in and preventive
Thresholds for a System of Degrading Components With Competing replacement optimization,” IIE Trans., vol. 45, no. 4, pp. 391–408, Apr.
Dependent Failure Processes Rutgers Univ., Piscataway, NJ, USA, ISE 2013.
Technical Report DWC2013-1, 2013. [35] K. Rafiee, Q. Feng, and D. W. Coit, “Reliability modeling for depen-
[11] J. P. Kharoufeh, “Explicit results for wear processes in a Markovian dent competing failure processes with changing degradation rate,” IIE
environment,” Operations Res. Lett., vol. 31, no. 3, pp. 237–244, May Trans. 2014, (in print).
2003. [36] H. Peng, D. W. Coit, and Q. Feng, “Component reliability criticality or
[12] J. P. Kharoufeh and S. M. Cox, “Stochastic models for degradation- importance measures for systems with degrading components,” IEEE
based reliability,” IIE Trans., vol. 37, no. 6, pp. 533–542, 2005. Trans. Rel., vol. 61, no. 1, pp. 391–408, Mar. 2012.
[13] M. Boulanger and L. A. Escobar, “Experimental-design for a class [37] M. Rausand and A. Hoyland, System Reliability Theory-Models, Statis-
of accelerated degradation tests,” Technometrics, vol. 36, no. 3, pp. tical Methods and Applications, 2nd ed. Hoboken, NJ, USA: Wiley,
260–272, Aug. 1994. 2004.
[14] S. T. Tseng, M. Hamada, and C. H. Chiao, “Using degradation data to [38] W. Li and H. Pham, “An inspection-maintenance model for systems
improve fluorescent lamp reliability,” J. Quality Technol., vol. 27, no. with multiple competing processes,” IEEE Trans. Rel., vol. 54, no. 2,
4, pp. 363–369, Oct. 1995. pp. 318–327, Jun. 2005.
SONG et al.: RELIABILITY ANALYSIS FOR MULTI-COMPONENT SYSTEMS SUBJECT TO MULTIPLE DEPENDENT COMPETING FAILURE PROCESSES 345

Sanling Song is a Ph.D. student in industrial and systems engineering at Rut- Qianmei Feng is an Associate Professor, and the Brij and Sunita Agrawal
gers University. She received her Bachelor, and Master degrees in engineering Faculty Fellow in the Department of Industrial Engineering at the University
mechanics from Dalian University of Technology, Dalian, China in 2007, and of Houston. She received the Ph.D. degree in industrial engineering from the
2010 respectively. She has worked for Phillips Electronics North America. She University of Washington, Seattle, WA in 2005. Her research has been sup-
is a member of IIE and INFORMS. ported by NSF, Department of Homeland Security (DHS), Texas Department
of Transportation (TxDOT), and Texas Higher Education Coordinating Board
(THECB). She is also a member of INFORMS, ASQ, and Alpha Pi Mu.

David W. Coit is a Professor in the Department of Industrial & Systems En-


gineering at Rutgers University. He received a B.S. degree in mechanical engi-
neering from Cornell University, an MBA from Rensselaer Polytechnic Insti- Hao Peng is an Assistant Professor in the Department of Industrial Engineering
tute, and M.S. and Ph.D. degrees in industrial engineering from the University and Innovation Sciences at the Eindhoven University of Technology, Eind-
of Pittsburgh. He also has over ten years of experience working for IIT Research hoven, Netherlands. She received the Ph.D. degree in industrial engineering
Institute (IITRI), Rome, NY (now called Alion Science and Technology). His from the University of Houston, Houston, TX in 2010. She received her
research has been funded by NSF, U.S. Navy, U.S. Army, power utilities, and Bachelor degree in industrial engineering from Tsinghua University, Beijing,
industry. He is a member of IIE and INFORMS. China (2006). Her research interests are in optimization for condition-based
maintenance, and quality and reliability engineering for evolving technologies.
She was awarded the Marie Curie career integration grant from European
Commission in 2012. She is a member of INFORMS and IIE.

View publication stats

S-ar putea să vă placă și