Time-dependent system reliability is computed as the probability that the responses of a system do not exceed prescribed failure thresholds over a time duration of interest. In this work, an efficient time-dependent reliability analysis method is proposed for systems with bivariate responses which are general functions of random variables and stochastic processes. Analytical expressions are derived first for the single and joint upcrossing rates based on the first-order reliability method (FORM). Time-dependent system failure probability is then estimated with the computed single and joint upcrossing rates. The method can efficiently and accurately estimate different types of upcrossing rates for the systems with bivariate responses when FORM is applicable. In addition, the developed method is applicable to general problems with random variables, stationary, and nonstationary stochastic processes. As the general system reliability can be approximated with the results from reliability analyses for individual responses and bivariate responses, the proposed method can be extended to reliability analysis of general systems with more than two responses. Three examples, including a parallel system, a series system, and a hydrokinetic turbine blade application, are used to demonstrate the effectiveness of the proposed method.

Introduction

Reliability is the ability that a component or system performs its intended function in routine circumstances for a given period of time [1]. For a system, we are interested in not only the reliability of its components, but also the system reliability. For many applications, the system reliability is more critical, because it can be used to estimate the lifecycle cost, determine maintenance policies, and perform optimization [24]. The system reliability is determined by the reliability of its components and also the relationship between the system and its components. By component, in this work, we mean a failure mode. A physical component itself can also be regarded as a system, because it may have multiple failure modes.

System reliability analysis is much more difficult than component reliability analysis. The latter needs to compute the probability of a single event, while the former needs to compute the joint probability of multiple and possibly dependent events. Many progresses have been made in system reliability analysis. For example, Ditlevsen [5] approximated the system reliability using a bounding formulas. Song and Kang [6] developed a matrix-based system reliability method, which can calculate the system reliability and system parameter sensitivities by a convenient matrix-based framework. Nguyen et al. [7] later developed a reliability-based system design optimization method by using the matrix-based system reliability method. Dey and Mahadevan [8] and Ambartzumian et al. [9] proposed a system reliability method using a standard normal multivariate cumulative distribution function (CDF); by employing the Morgan's laws [10], the method expresses the system probability of failure as the intersection of a set of unions of subsystems. Youn and Wang [11] suggested a complementary intersection method (CIM) for system reliability analysis by expressing the system failure event as complementary intersection events. Based on the CIM method, a generalized CIM method is developed later in Ref. [12]. More system reliability analysis methods have also been reported in Refs. [1] and [13].

To predict the system reliability, we need at first to calculate component reliability. This can be achieved by using the common reliability methods, such as the first-order reliability method (FORM), second-order reliability method, and importance sampling [1,1419]. The next step is to calculate joint probabilities associated with the failure events of all the components. If there are only two failure modes, the system reliability can be computed using a bivariate distribution. One example was reported in Ref. [9]. When there are many components involved, it is difficult and computationally expensive to obtain the joint probabilities. For this case, the system reliability can be approximated within its lower and upper bounds by just using the component reliabilities and joint probabilities for up to two failure modes (bivariate responses) [5]; the reliability bound is normally narrow for many engineering problems.

Although reliability is defined for a period of time and is also a function of time, most of the aforementioned reliability methods are for time-invariant reliability that does not change over time. For time-invariant reliability, the associated function of a response variable, called a limit-state function, is not a function of time. The reliability is therefore time independent. In many engineering applications, however, the limit-state function changes over time, either because time appears explicitly in the function or because stochastic processes are part of the input variables, or both. Examples include function generator mechanisms [20,21], bridges under stochastic loading [22,23], hydrokinetic turbine system subjected to wave or river flow loading [24,25], and vehicles running on stochastic road surfaces [4].

Time-dependent reliability analysis is much more challenging than its time-independent counterpart. The most common time-dependent reliability method is the Rice formula [26,27] developed in 1944 and is still widely used nowadays. There are still many developments in time-dependent reliability in recent years. For instance, for component reliability problems, Yang and Shlnozuka [28,29] developed a time-dependent reliability analysis method using the Markov stochastic process. Mourelatos and coworkers [30] employed the time-series modeling and importance sampling method to approximate the time-dependent reliability. Andrieu-Renaud et al. [31] proposed a PHI2 method for the time-dependent component reliability analysis of general problems with both random variables and stochastic processes. By using the Rice's formula [26,27], Du and Hu [24] developed a time-dependent reliability model for hydrokinetic turbine blades. Singh et al. [4] proposed the concept of composite limit-state function for time-dependent reliability analysis for a special group of problems. Du and Hu [32] presented a joint upcrossing method based on the work of Madsen and Krenk [33]. Wang and Wang [34,35] developed a nested extreme value response surface method by building a surrogate model for the extreme value response. They also proposed a double-loop surrogate modeling approach for time-dependent component reliability analysis in Ref. [36]. Jiang et al. [37,38] studied the time-dependent reliability analysis of general engineering systems based on stochastic process discretization.

Studies on time-dependent system reliability have also been reported. For example, a method was developed to estimate the joint first-passage probability of failure for systems under stochastic excitation [39]. An analytical approach was proposed for linear dynamical systems in higher dimensions by calculating the conditional upcrossing rate on the surfaces of failure boundaries [40]. An approximation method was reported to approximate the conditional first passage probability of systems under modulated white noise excitation [41]. By combining Monte Carlo simulations (MCS) with the asymptotic extreme value theory, Radhika et al. [42] proposed a reliability analysis method for nonlinear vibrating systems. Gupta and Manohar [43] proposed a multivariate extreme value distribution approach to approximate the extreme value distributions of a vector of stationary Gaussian random processes. The developed multivariate extreme value distribution method was then applied to reliability analysis of randomly vibrating structures subjected to jointly stationary Gaussian excitations [44,45]. Some of the above methods have been verified to have good accuracy for systems subjected to multiple Gaussian stationary stochastic processes. They, however, cannot be directly applied to general problems where the input variables of a limit-state function contain time, random variables, and nonstationary stochastic processes. Hagen and Tvedt [46] have investigated the reliability analysis of parallel system under vector of stochastic processes. Their method, however, needs to solve multivariate normal distribution, and new random variables are introduced into the model to compute the outcrossing rate. Although MCS is capable of handling the general problems, the required computational cost is prohibitive for engineering applications.

In this work, we propose a new time-dependent system reliability analysis method for bivariate responses that are nonlinear functions of time, random variables, and stochastic processes. The stochastic process can be stationary or nonstationary. The new method is an extension of the work in Ref. [39] and is based on the FORM and the Rice's formula [26,27]. The major development is the derivations of bivariate joint upcrossing rates, which can be used for estimating time-dependent system reliability for bivariate responses. Since the bivariate joint probabilities are the basis for general system reliability analysis when the reliability bound method is used, the proposed method can also be applied to general time-dependent system reliability analysis for general systems with more than two components.

In Sec. 2, we give the background of time-dependent reliability. In Sec. 3, we first introduce the upcrossing rate method; we then derive necessary equations for the bivariate joint upcrossing rates. The numerical procedure is summarized in Sec. 4, followed by three examples in Sec. 5. Conclusions are presented in Sec. 6.

Time-Dependent System Reliability

Recall that in this work, a component corresponds to a failure mode. Suppose there are r failure modes or r components, for component i, where i=1,2,,r, let its limit-state function be Gi(t)=gi(X,Y(t),t), where X=[X1,X2,,Xn] is a vector of random variables, Y(t)=[Y1(t),Y2(t),,Ym(t)] is a vector of stochastic processes, Gi(t) is the response variable, and t stands for time. Functions, Gi(t)=gi(X,Y(t),t), i=1,2,,r, are usually computer simulation models, such as those of finite element analysis (FEA) or computational fluid dynamics. They are nonlinear functions of random variables X and stochastic processes Y(t). As a result, the response variables Gi(t), i=1,2,,r, are stochastic processes with variation and correlations over time. For a special case where Gi(t)=gi(X,t), although no stochastic processes Y(t) appear in the input, the response variable Gi(t) is still a stochastic process due to the time-varying statistical properties and correlations over time. In addition, even if X and Y(t) are mutually independent from each other, the stochastic responses Gi(t), i=1,2,,r, are still correlated multivariate stochastic processes due to the sharing random variables or stochastic processes in the limit-state functions.

The time-dependent reliability Ri(t0,ts) of component i over the time interval [t0,ts] is defined by 
Ri(t0,ts)=Pr{gi(X,Y(t),t)<ei,t[t0,ts]}
(1)
in which ei is the failure threshold, and Pr{} stands for a probability. Note that ei is usually defined as ei=0 in most of the reliability analysis problems.
The corresponding time-dependent probability of failure pf,i(t0,ts) is 
pf,i(t0,ts)=Pr{gi(X,Y(t),t)>ei,t[t0,ts]}
(2)
Let Ωs be the safe region for a system. For a series system 
Ωs={[X,Y(t)]|i=1rgi(X,Y(t),t)<ei,t[t0,ts]}
(3)
in which stands for an intersection.
For a parallel system 
Ωs={[(X,Y(t)]|i=1rgi(X,Y(t),t)<ei,t[t0,ts]}
(4)
in which stands for a union.
With above definitions, we then have the time-dependent system reliability Rs(t0,ts) and probability of failure pf,s(t0,ts) as follows: 
Rs(t0,ts)=Pr{[X,Y(t)]Ωs,t[t0,ts]}
(5)
and 
pf,s(t0,ts)=Pr{[X,Y(t)]Ωs,t[t0,ts]}
(6)
The system reliability requires not only the component reliability but also joint probabilities up to an order of r. Evaluating a joint probability with a high-order is extremely difficult. To make the system reliability easier, Ditlevsen [5] proposed abound formulas for a series system as follows: 
pf,smin(t0,ts)pf,s(t0,ts)pf,smax(t0,ts)
(7)
where 
pf,smin(t0,ts)=pf,1(t0,ts)+i=2rmax{pf,i(t0,ts)j=1i1pf,ij(t0,ts),0}
(8)
 
pf,smax(t0,ts)=i=1rpf,i(t0,ts)i=2rmaxj<ipf,ij(t0,ts)
(9)

In the above equations, pf,i(t0,ts) is the component probability of failure. Components probabilities of failure are sorted in a decreasing order, and pf,1(t0,ts) is, therefore, the maximum component probability of failure [1]. pf,ij(t0,ts) is the joint probability of failure of components i and j. It is the probability that both components fail over [t0,ts]. The above reliability bounds (i.e., Eqs. (7)(9)) are for series systems. For parallel systems, the failure probability can be expressed as a function of series system reliability using De Morgan's law [10]. For mixed system with combined series and parallel structures, the generalized CIM method presented in Ref. [12] can be employed to decompose the system failure event into multiple mutually exclusive failure paths. Based on the decomposition, the reliability bounds for series systems can be used to provide bounds for the system reliability.

Equation (9) indicates that the component probability of failure pf,i(t0,ts) and bivariate probability of failure pf,ij(t0,ts) are the bases for the system probability of failure. As reviewed previously, many time-dependent reliability methods are available for pf,i(t0,ts). In this work, we focus on developing a new method for the bivariate probability of failure pf,ij(t0,ts). A straightforward way of evaluating pf,ij(t0,ts) is to directly generate samples of X and Y(t) using Monte Carlo simulation method. After that, the samples of X and Y(t) are plugged into the limit-state functions to get samples of responses and then obtain the system failure probability. Estimating the failure probability in this way is straightforward and accurate. However, as discussed previously, the limit-state functions may be expensive simulation models, such as FEA and computational fluid dynamics models. The required computational effort for this kind of straightforward simulation method is prohibitive. We discuss how to overcome this challenge in Sec. 3. The idea is to reduce the number of times of calling the limit-state function while maintaining the accuracy of reliability analysis.

Time-Dependent System Reliability for Bivariate Responses

For limit-state functions Gi(t)=gi(X,Y(t),t) and Gj(t)=gj(X,Y(t),t), the joint time-dependent probability of failure is given by 
pf,ij(t0,ts)=Pr{gi(X,Y(χ),χ)>eigj(X,Y(τ),τ)>ej,χandτ[t0,ts]}
(10)
Equation (10) can be further transformed into [39] 
pf,ij(t0,ts)=Pr{gi(X,Y(χ),χ)>ei,χ[0,ts]}+Pr{gj(X,Y(τ),τ)>ej,τ[t0,ts]}Pr{gi(X,Y(χ),χ)>eigj(X,Y(τ),τ)>ej,χandτ[t0,ts]}
(11)
According to Eq. (2), the first two terms on the right-hand side of Eq. (11) are component reliabilities, and Eq. (11) can be rewritten as 
pf,ij(t0,ts)=pf,i(t0,ts)+pf,j(t0,ts)pf,ij(t0,ts)
(12)
where 
pf,ij(t0,ts)=Pr{gi(X,Y(χ),χ)>eigj(X,Y(τ),τ)>ej,χandτ[t0,ts]}
(13)
which is the time-dependent probability of failure for a series system with components i and j. In Secs. 3.13.3, we first discuss the method for the time-dependent component reliability. We then derive equations for time-dependent joint probability, pf,ij(t0,ts).

Time-Dependent Component Reliability Analysis.

In this work, we employ the upcrossing rate method [31] to evaluate the time-dependent component probability of failure. It is the most commonly used method for time-dependent reliability analysis.

Upcrossing Rate Method for Time-Dependent Component Reliability Analysis.

For a general limit-state function Gk(t)=gk(X,Y(t),t),k=iorj with threshold ek, the time-dependent probability of failure pf,k(t0,ts) is given by the upcrossing rate method as follows: 
pf,k(t0,ts)=1[1pf,k(t0)]exp{t0tsvk+(t)dt}
(14)
in which pf,k(t0) is the instantaneous probability of failure at t0 and is given by 
pf,k(t0)=Pr{gk(X,Y(t0),t0)>ek}
(15)
vk+(t) is the upcrossing rate of component k at time instant t. Figure 1 shows upcrossing events of component k. An upcrossing event happens when the response variable Gk passes the threshold ek at time instant t from the safe region Gk(t)<ek to the failure region Gk(t+Δt)>ek, where Δt is an infinitesimally small time interval. The upcrossing rate vk+(t) is defined by the following limit: 
vk+(t)=limΔt0Pr{[gk(X,Y(t),t)<ek][gk(X,Y(t+Δt),t+Δt)>ek]}Δt
(16)

Equation (14) is derived based on the assumption that all the upcrossings over [t0,ts] are independent. Knowing vk+(t), we can easily obtain the component probability of failure pf,k(t0,ts) using Eq. (14).

In addition to using the upcrossing rate method, we also use FORM, which linearizes the limit-state function of components i and j at the so-called most probable point (MPP). Next, we first discuss the linearization and then discuss the estimation of upcrossing rate vk+(t). The linearization later is also used in the derivation of pf,ij(t0,ts).

Transformation of Limit-State Functions.

To use FORM for a general limit-state function of component k,Gk(t)=gk(X,Y(t),t),k=iorj, we first transform random variables X and stochastic processes Y(t) into standard normal random variables U(t)=(UX,UY(t)). Then, the limit-state function becomes 
Gk(t)=gk(X,Y(t),t)=gk(T(UX),T(UY(t)),t)=gk(U(t),t),k=iorj
(17)

where T() stands for the transforming operator. The transformation can be found in Ref. [47].

Then, the MPP uk*(t)=(uX*,uY*(t)),k=iorj is found with the following optimization model: 
{minuu(t)subjecttogk(u(t),t)=ek,k=iorj
(18)
in which stands for the length of a vector.
After the limit-state function is linearized at the MPP, the failure event Gk(t)=gk(X,Y(t),t)>ek,k=iorj is equivalent to the following event [47]: 
Lk(t)=αk(t)U(t)T>βk(t),k=iorj
(19)
in which 
βk(t)=uk*(t),k=iorj
(20)
 
αk(t)=uk*(t)uk*(t),k=iorj
(21)

βk(t),k=iorj is called the Hasofer-Lind reliability index.

Therefore, the failure events given in Eq. (12) become 
pf,i(t0,ts)=Pr{Li(χ)=αi(χ)UT(χ)>βi(χ),χ[t0,ts]}
(22)
 
pf,j(t0,ts)=Pr{Lj(τ)=αj(τ)UT(τ)>βj(τ),τ[t0,ts]}
(23)
and 
pf,ij(t0,ts)=Pr{αi(χ)UT(χ)>βi(χ),χ[t0,ts]αj(τ)UT(τ)>βj(τ),τ[t0,ts]}
(24)

The equations for the time-dependent component probabilities of failure, pf,k(t0,ts),k=iorj, are already available. We reviewed these equations in Appendix  A. In Sec. 3.2, we investigate the method for the approximation of the bivariate probability pf,ij(t0,ts). Note that since the FORM method is employed to linearize the limit-state function at the MPP, the accuracy of the failure probability estimate may be affected by the accuracy of FORM. The developed method in Secs. 3.2 and 3.3 is therefore applicable to problems where FORM is accurate.

Time-Dependent Joint Probability pf,ij(t0,ts)

Outcrossing Rate Method for Time-Dependent Joint Probability Analysis.

We now derive equations for the bivariate joint probability pf,ij(t0,ts). Following the same strategy of upcrossing rate in Eq. (14), we have pf,ij(t0,ts) as follows: 
pf,ij(t0,ts)=1Rij(t0)exp{t0tsvij+(t)dt}
(25)
in which Rij(t0) is the probability that both components are safe at the initial time and is given by 
Rij(t0)=Pr{gi(X,Y(t0),t0)eigj(X,Y(t0),t0)ej}
(26)
vij+(t) is the outcrossing rate of a series system with components i and j at time instant t. An outcrossing event occurs when the system outcrosses its bounds at time instant t from the safe region to the failure region. Figure 2 shows three representative outcrossing events of the series system. For the outcrossing events, both components i and j are in the safe region at time instants tm, m = 1, 2, and 3. The system then outcrosses into the failure region as a result of the upcrossing of Gi, or upcorring of Gj, or both the upcrossings of Gi and Gj at the following time instants, tm+t, m = 1,3, and 2. Given in mathematical form, the outcrossing rate vij+(t) is given by the following limit: 
vij+(t)=limΔt0Pr{[Gi(t)<eiGj(t)<ej][Gi(t+Δt)>eiGj(t+Δt)>ej]}Δt
(27)

where Δt is an infinitesimally small time interval.

The probability in Eq. (27) can be decomposed into three components. 
Pr{[Gi(t)<eiGj(t)<ej][Gi(t+Δt)>eiGj(t+Δt)>ej]}=pij+(t)+pij+(t)+pij++(t)
(28)
where 
pij+(t)=Pr{[Gi(t)<eiGj(t)<ej][Gi(t+Δt)>eiGj(t+Δt)<ej]}
(29)
 
pij+(t)=Pr{[Gi(t)<eiGj(t)<ej][Gi(t+Δt)<eiGj(t+Δt)>ej]}
(30)
 
pij++(t)=Pr{[Gi(t)<eiGj(t)<ej][Gi(t+Δt)>eiGj(t+Δt)>ej]}
(31)

pij+(t) is the probability that Gi(t) upcrosses its barrier ei, while Gj(t) remains below its barrier ej at t, pij+(t) is the probability that Gj(t) upcrosses its barrier ej, while Gi(t) remains below its barrier ei at t, and pij++(t) is the probability that both Gi(t) and Gj(t) upcross their barriers at t.

We then have three corresponding joint upcrossing rates defined by 
vij+(t)=limΔt0pij+(t)Δt
(32)
 
vij+(t)=limΔt0pij+(t)Δt
(33)
 
vij++(t)=limΔt0pij++(t)Δt
(34)
Then 
vij+(t)=vij+(t)+vij+(t)+vij++(t)
(35)

Equations for vij+(t) are available for special limit-state functions with two stationary Gaussian vector processes [39]. In Secs. 3.2.23.2.4, we derive equations for vij+(t) and other two joint upcrossing rates for general limit-state functions. The derivations are based on the approximation discussed in Sec. 3.1.2.

vij+(t).

Substituting Eqs. (22) and (23) into Eq. (29) yields 
pij+(t)=Pr{[Li(t)<βi(t)Lj(t)<βj(t)][Li(t+Δt)>βi(t+Δt)Lj(t+Δt)<βj(t+Δt)]}
(36)
It is the probability that Li(t) upcrosses its barrier βi(t), while Lj(t) remains below its barrier βj(t) at t. With the Rice's formula [26,27], vij+(t)=limΔt0pij+(t)/Δt can be calculated by the following integral: 
vij+(t)=βj(t)β˙i(t)[l˙iβ˙i(t)]fLiLjL˙i(βi(t),lj,l˙i)dl˙idlj
(37)

where fLiLjL˙i(,,) is the joint probability density function (PDF) of Li(t), Lj(t), and L˙i(t).

As there is no closed-form available for vij+(t), we perform some transformations for Eq. (37) before we derive necessary equations for it. The transformation is given in Appendix  B.

Applying Eq. (A4), Eq. (B3) is rewritten as below: 
vij+(t)=ϕ(βi(t))βj(t)σL˙i|Li=βi(t),Lj=ljσLj|Li=βi(t)ϕ(ljμLj|Li=βi(t)σLj|Li=βi(t))Hdlj
(38)
where 
H=ϕ(β˙i(t)μL˙i|Li=βi(t),Lj=ljσL˙i|Li=βi(t),Lj=lj)β˙i(t)μL˙i|Li=βi(t),Lj=ljσL˙i|Li=βi(t),Lj=ljΦ(β˙i(t)μL˙i|Li=βi(t),Lj=ljσL˙i|Li=βi(t),Lj=lj)
(39)

The above equations indicate that μLj|Li=βi, σLj|Li=βi, μL˙i|Li=βi(t),Lj=lj, and σL˙i|Li=βi(t),Lj=lj are required to solve for vij+(t), for which we must to obtain the mean and covariance of L˙i(t) and L=[Li(t),Lj(t)].

Since Li(t)=αi(t)UT(t)=αXi(t)UXT+αYi(t)UYT(t), we have L˙i(t) as follows: 
L˙i(t)=α˙Xi(t)UXT+α˙Yi(t)UYT(t)+αYi(t)U˙YT(t)=α˙i(t)UT(t)+αYi(t)U˙YT(t)
(40)
With Eqs. (22) through (23) and (40), we write the covariance matrix of L˙ and L below: 
cL˙=[cL˙iL˙icL˙iLcLL˙icLL]=[ωi2(t)cLiL˙icLjL˙icLiL˙i1cLiLjcLjL˙icLiLj1]
(41)
in which ωi2(t) is given in Eq. (A5), and the other components of the matrix are given below: 
cLiL˙i=αi(t)α˙iT(t)
(42)
 
cLiLj=αi(t)αjT(t)
(43)
 
cLjL˙i=α˙i(t)αjT(t)
(44)
After obtaining the covariance matrix cL˙, we now derive equations for the conditional means and standard deviations. They are given by 
μL˙i|Li=βi(t),Lj=lj=cL˙iLcLL1l
(45)
 
σ2L˙i|Li=βi(t),Lj=lj=cL˙iL˙icL˙iLcLL1cLL˙i
(46)
in which l=[βi(t);lj].
Substituting Eq. (41) into Eqs. (45) and (46), we have 
μL˙i|Li=βi(t),Lj=lj=βi(t)(cLjL˙icLiLjcLiL˙i)+lj(cLiL˙icLiLjcLjL˙i)c2LiLj1
(47)
 
σ2L˙i|Li=βi(t),Lj=lj=ωi2(t)cLiL˙i(cLjL˙icLiLjcLiL˙i)+cLjL˙i(cLiL˙icLiLjcLjL˙i)c2LiLj1
(48)
Since αi(t)αiT(t)=1, we have cLiL˙i=αi(t)α˙iT(t)=0. Eqs. (47) and (48) are then simplified as 
μL˙i|Li=βi(t),Lj=lj=βi(t)cLjL˙icLiLjljcLjL˙icLiLj21
(49)
and 
σ2L˙i|Li=βi(t),Lj=lj=ωi2(t)+cLjL˙i2cLiLj21
(50)
Similarly, we have 
μLj|Li=βi(t)=cLiLjβi(t)
(51)
 
σLj|Li=βi(t)2=1cLiLj2
(52)

We have so far derived all the equations needed to calculate vij+(t), which can be solved by substituting Eqs. (47) through (52) into Eq. (38).

vij+(t).

After obtaining the first joint upcrossing rate vij+(t), we can easily get the second upcrossing rate vij+(t). We just need to change the subscripts of Eqs. (36) through (52) by switching i and j.

vij++(t).

Substituting Eqs. (22) and (23) into Eq. (31), we have 
vij++(t)=limΔt0Mij++Δt
(53)
where 
Mij++=Pr[Li(t)<βi(t)Lj(t)<βj(t)][Li(t+Δt)>βi(t+Δt)Lj(t+Δt)>βj(t+Δt)]
(54)
Defining Zi(t)=Li(t)βi(t), we have 
Mij++=0+0+00fZiZjZ˙iZ˙j(zi,zj,z˙i,z˙j)dzidzjdz˙idz˙j
(55)
t is omitted in Eq. (55) for brevity; for example, zi stands for zi(t) now. In Appendix  C, we have demonstrated that vij++(t) tends to be zero when Δt becomes infinitely small. Based on the demonstration, we, therefore, conclude that vij++(t)=0.
Having obtained all the three joint upcrossing rates, the outcrossing rate in Eq. (35) is then given by 
vij+(t)=vij+(t)+vij+(t)
(56)

Rij(t0).

Rij(t0) is another component we need for the system reliability analysis. After the MPPs of components i and j are found at t0, Rij(t0) is calculated by [8] 
Rij(t0)=Φ(βi(t0),βj(t0),ρ0)
(57)
in which Φ(,,) is the CDF of a bivariate normal random variable, and ρ0 is the coefficient of correlation between the two components. ρ0 is given by Eq. (50) as follows: 
ρ0=αi(t0)αjT(t0)
(58)

With Eqs. (25) through (58), we can estimate pf,ij(t0,ts). Then, the time-dependent probability of failure for bivariate responses, pf,ij(t0,ts), can be computed.

System Reliability Analysis.

With the availability of the outcrossing rate vij+(t), we now summarize the system reliability analysis method for bivariate responses.

For a series system with component i and j, the system probability of failure is given by 
pf,s(t0,ts)=pf,ij(t0,ts)=1Rij(t0)exp{t0tsvij+(t)dt}
(59)
For a parallel system, we have 
pf,s(t0,ts)=1Ri(t0)exp{t0tsvi+(t)dt}Rj(t0)exp{t0tsvj+(t)dt}+Rij(t0)exp{t0tsvij+(t)dt}
(60)

In above equations, Rij(t0) is computed using Eqs. (57) and (58), vij+(t) is estimated using Eq. (56), and vi+(t) is computed using Eq. (16). In addition, vij+(t) and vij+(t) are required to compute vij+(t) in Eq. (56). These two terms can be computed using the expressions given from Eq. (40) through Eq. (52). Note that this paper only focuses on reliability analysis of systems with bivariate responses. Extension of the proposed method to systems with higher number of responses will be investigated in future.

Until now we have all the equations needed for the time-dependent system reliability analysis for bivariate responses.

Numerical Procedure

We provide a flowchart for the proposed method in Fig. 3; we also summarize the main steps as below:

  • Step 1: Initialization of parameters

    Transform the non-Gaussian random variables and stochastic processes into standard Gaussian random variables and stochastic processes.

  • Step 2: FORM

    Perform the MPP search at time instant t using Eq. (18); obtain the associated reliability indices, and the derivative of reliability indices, such as αi(t),α˙i(t),βi(t),β˙i(t),αj(t),α˙j(t),βj(t),β˙j(t), by applying Eqs. (20), (21), (A3), and (A6).

  • Step 3: Initial reliability

    Calculate the initial component reliability using Eq. (25) and initial system reliability using Eqs. (57) and (58).

  • Step 4: Upcrossing rates and outcrossing rate

    Compute the component upcrossing rates using Eq. (A2) and vij+(t) and vij+(t) using Eq. (38); then obtain the joint upcrossing rate vij+(t).

  • Step 5: Integration

    Integrate the upcrossing rates over [t0,ts] using Eqs. (14) and (25).

  • Step 6: System reliability

    Obtain the system probability of failure pf,s(t0,ts) using Eq. (59) or (60).

Numerical Examples

In this section, we use three examples to demonstrate the proposed method. They are a Daniels system [48], a function generator mechanism system, and a hydrokinetic turbine system. Each example represents an important area of applications. Example 1 is a parallel system and a structural analysis problem where both random variables and stochastic processes are involved. Example 2 is a series system and a mechanism analysis problem. Even though there are no stochastic processes in the limit-state functions in Example 2, the responses of the mechanism system are still stochastic process, because the limit-state functions are functions of time and random variables. Example 3 is a hydrokinetic turbine system where the turbine blades are subjected to nonstationary stochastic river flow load.

Example 1: A Daniels System.

Figure 4 shows a structural system under stochastic loading.

The system consists of two bars. Due to different manufacturing precisions, the two bars have different standard deviations in their dimensions. As the two bars are exposed to corrosions, their widths and heights decrease at the rates of k1 and k2, respectively. Each of the two bars resists a load of P(t)/2 until both of the two bars yield. The task is to determine the time-dependent system probabilities of failure over different time intervals up to [0, 20] years.

Since the system is parallel, the time-dependent system probability of failure is given by 
pf,s(t0,ts)=Pr{g1(X,Y(χ),χ)>e1g2(X,Y(τ),τ)>e2,χandτ[t0,ts]}
(61)
where 
g1(X,Y(t),t)=P(t)2(a12k1t)(b12k1t)σb1
(62)
 
g2(X,Y(t),t)=P(t)2(a22k2t)(b22k2t)σb2
(63)

and X=[a1,b1,a2,b2,σb1,σb2], and Y(t)=[P(t)]; σb1 and σb2 are the yield strengths of bars 1 and 2, respectively. The parameters in Eqs. (62) and (63) are presented in Table 1.

The autocorrelation function of the stochastic process P(t) is given by 
ρP(t1,t2)=exp[(t2t1)2/ζ2]
(64)

where ζ=2yr is the correlation length. The longer is the time interval t2t1, the weaker is the autocorrelation.

To evaluate the accuracy of the new method, we also performed MCS using a large sample size of 107. We compared the upcrossing rates v12+(t), v12+(t), and v12+(t) obtained from the proposed method and MCS as well.

Figures 57 depict the upcrossing rates v12+(t), v12+(t), and outcrossing rate v12+(t) from both the new method and MCS.

Note that the curves of upcrossing rates and outcrossing rate from MCS are not smooth. The noise comes from the numerical discretization of stochastic process. Nevertheless, the results show the good consistency between the MCS results and those from the proposed method. This example indicates that the proposed method can produce accurate joint upcrossing rates and outcrossing rate that are needed for time-dependent system reliability analysis.

Using the outcrossing rate v12+(t), we obtained the system reliability analysis result. The joint probability pf,12(t0,ts) and time-dependent system probability of failure pf,s(t0,ts) are depicted in Figs. 8 and 9, respectively. The latter is also given in Table 2.

As shown in Fig. 9, the error of the new method becomes larger with a longer period of time or with a larger probability of failure. The error resource is mainly the assumption of independent crossings. It is the intrinsic drawback of the upcrossing and outcrossing rate method [33].

The error ε (%) is calculated by 
ε=|pf,MCSpf,new|pf,MCS×100%
(65)
in which pf,MCS and pf,new are the results obtained from MCS and the new method, respectively.

To show the computational cost of the new method, we also provide the numbers of function calls in Table 3. The results indicate that the new method is much more efficient than MCS.

Example 2: A Function Generator Mechanism System.

A function generator mechanism is a mechanism used to realize a desired motion [20,21]. Such a system is shown in Fig. 10. This system consists of two function generator mechanisms. Mechanism 1, a four-bar linkage mechanism with links B1, B2, B3, and B4, generates a sine function, while mechanism 2, the other four-bar linkage mechanism with links B1, B5, B6, and B7, generates a logarithm function.

For the sine function generator (mechanism 1), the motion input and motion output are γ and κ=κa(B,γ), respectively. The required motion output is given by 
κd(γ)=60deg+60degsin[34(γ97deg)]
(66)
For the logarithm function generator (mechanism 2), the motion input and motion output are θ and η=ηa(B,θ), respectively. The required motion output is given by 
ηd(θ)=60deglog10[(θ+15deg)/60deg]log10(2)
(67)
A motion error is the difference between the actual motion output and the required motion output. For the two mechanisms, their motion errors are 
εκ(B,γ)=κa(B,γ)κd(γ)
(68)
and 
εη(B,θ)=ηa(B,θ)ηd(θ)
(69)

where B=[B1,B2,,B7].

Links B2 and B5 are welded together, the two input angles satisfy 
γ=62deg+θ
(70)
From the mechanism analysis, the following equations can be obtained: 
κa(B,γ)=2arctan(Eκ±Eκ2+Dκ2Fκ2FκDκ)
(71)
where Dκ=2B4(B1B2cosγ), Eκ=2B2B4sinγ, and Fκ=B12+B22+B42B322B1B2cosγ. 
ηa(B,θ)=2arctan(Eη±Eη2+Dη2Fη2FηDη)
(72)
 
ηd(θ)=60deglog10[(θ+15deg)/60deg]/log10(2)
(73)

where Dη=2B7(B1B5cosθ), Eη=2B5B7sinθ, and Fη=B12+B52+B72B622B1B5cosθ.

In this problem, the time factor is the input angle θ. There are no stochastic processes in the input variables. The vector of random variables is therefore X=B=[B1,B2,,B7], and the vector of stochastic processes Y is empty. Since the time factor θ appears in both functions of the motion errors, the motion errors are still stochastic processes. The motion errors should not be large, and their allowable values are denoted by e1 and e2. All the parameters are given in Table 4.

We would like the mechanism system to perform its intended functions over an interval of [θ0,θs]=[45deg,105deg]. If either motion error is greater than its allowable value over [θ0,θs]=[45deg,105deg], a failure is considered. As a result, the system is a series system, and the system probability of failure is 
pf,s(θ0,θs)=Pr{εκ(B,χ)>e1εη(B,τ)>e2,χ,τ[θ0,θs]}
(74)

Figures 1113 show the results of joint upcrossing rates v12+(θ), v12+(θ), and outcrossing rate v12+(θ), from the proposed method and MCS. The sample size of MCS is 107.

The results show that the proposed method is able to estimate the joint upcrossing rate with good accuracy. Based on the joint upcrossing rates, we obtained the time-dependent system probability of failure as presented in Fig. 14 and Table 5.

The results show that the accuracy of the proposed method is good. Table 6 gives the number of function calls required by the new method and MCS.

Example 3: A Hydrokinetic Turbine System

Problem Statement.

A hydrokinetic turbine system is employed as our third example [24,49]. This system is used to extract energy from river water flow, and it is subjected to stochastic flow loads during its operation. A three-dimensional model of the hydrokinetic turbine blade is created as shown in Fig. 15. The blade is 1-m long and made of steel. The turbine blade is twisted and has variable chord length along the radial direction. The hydrofoil of the blade is NREL S809. The lift and drag coefficients of the hydrofoil are available in Ref. [50].

A failure occurs if either of the two blades fails when the maximum strain of the blade is larger than the allowable strain. The time-dependent failure probability of the turbine system is given by 
pf,s(t0,ts)=Pr{g1(X,Y(χ),χ)>0g2(X,Y(τ),τ)>0,χandτ[t0,ts]}
(75)
where [t0,ts]=[0,12] months, gi(X,Y(χ),χ)=Smax,i(χ)σbi, where Smax,i(χ) is the maximum strain of the blade obtained from FEA simulation, X=[ν1,σb1,ν2,σb2], νi, and σbi are the Young's modulus and maximum allowable strain of the ith blade, respectively, and Y(t)=[Vr(t)] is the stochastic river flow velocity. Table 7 gives the statistical information of the variables involved. The mean μV(t), standard deviation σV(t), and correlation function ρV(t1,t2) of the stochastic river velocity Vr(t) are given by 
μV(t)=i=14aisin(bit+ci)
(76)
where [a1,a2,a3,a4]=[3.815,2.528,1.176,0.0786], [b1,b2,b3,b4]=[0.2895,0.5887,0.7619,2.183], and [c1,c2,c3,c4]=[0.2668,0.9651,3.116,3.161]. 
σV(t)=i=140.1aiexp(((tbi)/ci)2)
(77)
where [a1,a2,a3,a4]=[0.7382,1.013,1.875,1.283], [b1,b2,b3,b4]=[6.456,4.075,9.913,1.035], and [c1,c2,c3,c4]=[0.9193,1.561,6.959,2.237] 
ρV(t1,t2)=exp[(t2t1)2/0.082]
(78)

The above information of the river flow velocity is obtained by analyzing the historical data of the Missouri river [51].

Next, we first discuss how to perform FEA analysis to obtain the maximum strain for a given river flow velocity and realization of random variables. We then present the process and result of the time-dependent reliability analysis.

Finite Element Analysis.

From the load analysis, it is found that the turbine blade is subjected to an edgewise moment generated from the edgewise force FT and a flapwise moment generated from the flapwise force FN. In order to compute FT and FN, the turbine blade is divided into 48 stations along the radial direction. After the discretization, the edgewise force FT,i and flapwise force FN,i at station i are computed using the blade element momentum theory [52] based on the geometry of the turbine blade, local pitch, and the river velocity at the station. More details about the load analysis of the turbine blade is available in Ref. [52]. After the forces at stations of the turbine blade are obtained, they are input into FEA to get the stress response of the blade. Figure 16 gives the flowchart of the stress analysis. Figure 17 plots a snapshot of the stress analysis results of the blade. From the stress response, we obtain the maximum stress and also the maximum strain.

Time-Dependent Reliability Analysis.

Figures 1820 show the results of v12+(θ), v12+(θ), and v12+(θ), from the proposed method and MCS. The sample size of MCS is 106.

The results show that the proposed method is able to estimate the joint upcrossing rate accurately even for problems with nonstationary Gaussian process. Based on the joint upcrossing rates, we obtain the time-dependent system probability of failure as presented in Fig. 21 and Table 8. The results show that the accuracy of the proposed method is good. Table 9 gives the number of function calls required by the new method and MCS.

Conclusion

Time-dependent system reliability analysis plays a vital role in the system level optimization, lifecycle cost estimation, and decision making on maintenance and warranty. With the availability of computational models, predicting how the system reliability changes with time is possible. Making such a prediction both accurate and efficient is critical.

In this work, we proposed a time-dependent reliability method for a system with two response variables that are functions of random variables and stochastic processes. The method is based on the FORM and the upcrossing rate method (the Rice's formula). The new method can be applied to general problems with random variables, stochastic processes, and time, because it can be extended to systems with more than two response variables. With the use of FORM, the proposed method is also efficient. However, the error of FORM may affect the accuracy of the proposed method. The proposed method is therefore limited to problems when FORM is applicable.

As an upcrossing rate method, the new method may produce a larger error if the probability of failure is larger. The reason is that when the probability of failure is large, the dependency between upcrossings may become strong.

The future research based on this work may be: (1) the improvement of the accuracy, (2) the extension of the method to systems with more than two response variables, and (3) the integration of the method with optimization so that the time-dependent system reliability-based design can be performed.

Appendix A: Time-Dependent Component Probability of Failure, pf,k(t0,ts)

pf,k(t0,ts),k=iorj can be computed with Eq. (14), where the initial probability of failure is 
pfk(t0)=1Φ[βk(t0)]wherek=iorj
(A1)

where Φ() is the CDF of a standard normal random variable.

The upcrossing rate vk+(t) in Eq. (14) is calculated with the Rice's formula [26,27] and is given by 
vk+(t)=ωk(t)ϕ(βk(t))Ψ(β˙k(t)/ωk(t))
(A2)
in which 
β˙k(t)=βk(t)t
(A3)
 
Ψ(x)=x(qx)ϕ(q)dq=ϕ(x)xΦ(x)
(A4)

and ϕ() is the PDF of a standard normal random variable.

ωk(t) is given by [24] 
ωk2(t)=α˙k(t)α˙kT(t)+αk(t)C¨12(t,t)αkT(t)
(A5)
where 
α˙k(t)=[αk(t+Δt)αk(t)]/Δt
(A6)
and 
C¨12(t1,t2)|t1=t2=t=[000002ρY1(t1,t2)t1t2000000002ρYm(t1,t2)t1t2]|t1=t2=t
(A7)
in which ρYl(t1,t2), where l=1,2,m, are coefficients of autocorrelation of stochastic process l.

Appendix B: Transformation of vij+(t)

Equation (37) can be rewritten as 
vij+(t)=βj(t)β˙i(t)[l˙iβ˙i(t)]fLi(βi(t))fLj|Li=βi(t)(lj)fL˙i|Li=βi(t),Lj=lj(l˙i)dl˙idlj
(B1)

where fLj|Li=βi(t)() is the conditional PDF of Lj given Li=βi(t), and fL˙i|Li=βi(t),Lj=lj() is the conditional PDF of L˙i given Li=βi(t) and Lj=lj.

Since Li(t), Lj(t), and L˙i(t) are all Gaussian processes, we have 
vij+(t)=ϕ(βi(t))βj(t)1σLj|Li=βi(t)ϕ(ljμLj|Li=βi(t)σLj|Li=βi(t)){β˙i(t)[l˙iβ˙i(t)]fL˙i|Li=βi(t),Lj=lj(l˙i)dl˙i}dlj
(B2)
Defining z˙=(l˙iμL˙i|Li=βi(t),Lj=lj)/σL˙i|Li=βi(t),Lj=lj, we transform Eq. (B2) into 
vij+(t)=ϕ(βi(t))βj(t)1σLj|Li=βi(t)ϕ(ljμLj|Li=βi(t)σLj|Li=βi(t)){σL˙i|Li=βi(t),Lj=ljz˙L(z˙+μL˙i|Li=βi(t),Lj=ljβ˙i(t)σL˙i|Li=βi(t),Lj=lj)ϕ(z˙)dz˙}dlj
(B3)

where Z˙L=(β˙i(t)μL˙i|Li=βi(t),Lj=lj)/σL˙i|Li=βi(t),Lj=lj.

Appendix C: Derivation of vij++(t)

Since Zi(t+Δt)>0 and Zj(t+Δt)>0, expanding Zi and Zj at time t, we have Zi(t+Δt)zi(t)+z˙i(t)Δt>0 and Zj(t+Δt)zj(t)+z˙j(t)Δt>0, then Eq. (55) becomes 
Mij++=0+0+z˙j(t)Δt0z˙i(t)Δt0fZiZjZ˙iZ˙j(zi,zj,z˙i,z˙j)dzidzjdz˙idz˙j
(C1)
Let Zi(t)/(Z˙i(t)Δt)=Wi(t) and Zj(t)/(Z˙j(t)Δt)=Wj(t), Eq. (C1) is further transformed into 
Mij++=0+0+1010fWiWjZ˙iZ˙j(z˙iΔtwi,z˙jΔtwj,z˙i,z˙j)z˙iz˙jΔtΔtdwidwjdz˙idz˙j
(C2)
Plugging Eq. (C2) into (53) yields 
vij++(t)=limΔt00+0+1010fWiWjZ˙iZ˙j(z˙iΔtwi,z˙jΔtwj,z˙i,z˙j)z˙iz˙jΔtΔtdwidwjdz˙idz˙jΔt=limΔt00+0+1010fWiWjZ˙iZ˙j(0,0,z˙i,z˙j)z˙iz˙jΔtΔtdwidwjdz˙idz˙jΔt=limΔt0Δtβ˙j(t)+β˙i(t)+fLiLjL˙iL˙j(βi,βj,l˙i,l˙j)(l˙iβ˙i)(l˙jβ˙j)dl˙idl˙j=0
(C3)

Acknowledgment

This material is based upon the work supported in part by the Intelligent Systems Center at the Missouri University of Science and Technology.

Funding Data

  • Division of Civil, Mechanical and Manufacturing Innovation, National Science Foundation (Grant Nos. CMMI 1234855 and CMMI 1727329).

References

References
1.
Madsen
,
H. O.
,
Krenk
,
S.
, and
Lind
,
N. C.
,
1986
,
Methods of Structural Safety
, Dover Publications, Mineola, NY.
2.
Frangopol
,
D. M.
,
2011
, “
Life-Cycle Performance, Management, and Optimisation of Structural Systems Under Uncertainty: Accomplishments and Challenges
,”
Struct. Infrastruct. Eng.
,
7
(
6
), pp.
389
413
.
3.
Liu
,
J.
,
Wang
,
Y.
, and
Teng
,
R.
,
2012
, “
Research of Maintenance Decision-Making About Existing Building Structure Based on the Life-Cycle Cost Theory
,”
Appl. Mech. Mater.
,
174–177
, pp.
1824
1827
.
4.
Singh
,
A.
,
Mourelatos
,
Z. P.
, and
Li
,
J.
,
2010
, “
Design for Lifecycle Cost Using Time-Dependent Reliability
,”
ASME J. Mech. Des.
,
132
(
9
), p.
091008
.
5.
Ditlevsen
,
O.
,
1979
, “
Narrow Reliability Bounds for Structural Systems
,”
J. Struct. Mech.
,
7
(
4
), pp.
453
472
.
6.
Song
,
J.
, and
Kang
,
W. H.
,
2009
, “
System Reliability and Sensitivity Under Statistical Dependence by Matrix-Based System Reliability Method
,”
Struct. Saf.
,
31
(
2
), pp.
148
156
.
7.
Nguyen
,
T. H.
,
Song
,
J.
, and
Paulino
,
G. H.
,
2010
, “
Single-Loop System Reliability-Based Design Optimization Using Matrix-Based System Reliability Method: Theory and Applications
,”
ASME J. Mech. Des.
,
132
(
1
), p.
011005
.
8.
Dey
,
A.
, and
Mahadevan
,
S.
,
1998
, “
Ductile Structural System Reliability Analysis Using Adaptive Importance Sampling
,”
Struct. Saf.
,
20
(
2
), pp.
137
154
.
9.
Ambartzumian
,
R.
,
Der Kiureghian
,
A.
,
Ohanian
,
V.
, and
Sukiasian
,
H.
,
1998
, “
Multinormal Probability by Sequential Conditioned Importance Sampling: Theory and Application
,”
Probab. Eng. Mech.
,
13
(
4
), pp.
299
308
.
10.
Vaurio
,
J. K.
,
2001
, “
Making Systems With Mutually Exclusive Events Analysable by Standard Fault Tree Analysis Tools
,”
Reliab. Eng. Syst. Saf.
,
74
(
1
), pp.
75
80
.
11.
Youn
,
B. D.
, and
Wang
,
P.
,
2009
, “
Complementary Intersection Method for System Reliability Analysis
,”
ASME J. Mech. Des.
,
131
(
4
), p.
041004
.
12.
Wang
,
P.
,
Hu
,
C.
, and
Youn
,
B. D.
,
2011
, “
A Generalized Complementary Intersection Method (GCIM) for System Reliability Analysis
,”
ASME J. Mech. Des.
,
133
(
7
), p.
071003
.
13.
Haldar
,
A.
, and
Mahadevan
,
S.
,
2000
,
Probability, Reliability and Statistical Methods in Engineering Design
,
Wiley
,
New York
.
14.
Du
,
X.
,
Sudjianto
,
A.
, and
Huang
,
B.
,
2005
, “
Reliability-Based Design With the Mixture of Random and Interval Variables
,”
ASME J. Mech. Des.
,
127
(
6
), pp.
1068
1076
.
15.
Huang
,
B.
, and
Du
,
X.
,
2008
, “
Probabilistic Uncertainty Analysis by Mean-Value First Order Saddlepoint Approximation
,”
Reliab. Eng. Syst. Saf.
,
93
(
2
), pp.
325
336
.
16.
Zhang
,
J.
, and
Du
,
X.
,
2010
, “
A Second-Order Reliability Method With First-Order Efficiency
,”
ASME J. Mech. Des.
,
132
(
10
), p.
101006
.
17.
Banerjee
,
B.
, and
Smith
,
B. G.
,
2011
, “
Reliability Analysis for Inserts in Sandwich Composites
,”
Adv. Mater. Res.
,
275
, pp.
234
238
.
18.
Kim
,
D. W.
,
Jung
,
S. S.
,
Sung
,
Y. H.
, and
Kim
,
D. H.
,
2011
, “
Optimization of SMES Windings Utilizing the First-Order Reliability Method
,”
Trans. Korean Inst. Electr. Eng.
,
60
(
7
), pp.
1354
1359
.
19.
Millwater
,
H.
, and
Feng
,
Y.
,
2011
, “
Probabilistic Sensitivity Analysis With Respect to Bounds of Truncated Distributions
,”
ASME J. Mech. Des.
,
133
(
6
), p.
061001
.
20.
Shi
,
Z.
,
1997
, “
Synthesis of Mechanical Error in Spatial Linkages Based on Reliability Concept
,”
Mech. Mach. Theory
,
32
(
2
), pp.
255
259
.
21.
Zhang
,
J.
, and
Du
,
X.
,
2011
, “
Time-Dependent Reliability Analysis for Function Generator Mechanisms
,”
ASME J. Mech. Des.
,
133
(
3
), p.
031005
.
22.
Caprani
,
C. C.
, and
Obrien
,
E. J.
,
2010
, “
The Use of Predictive Likelihood to Estimate the Distribution of Extreme Bridge Traffic Load Effect
,”
Struct. Saf.
,
32
(
2
), pp.
138
144
.
23.
Tao
,
Z.
,
Corotis
,
R. B.
, and
Ellis
,
J. H.
,
1994
, “
Reliability-Based Bridge Design and Life Cycle Management With Markov Decision Processes
,”
Struct. Saf.
,
16
(
1–2
), pp.
111
132
.
24.
Hu
,
Z.
, and
Du
,
X.
,
2012
, “
Reliability Analysis for Hydrokinetic Turbine Blades
,”
Renewable Energy
,
48
, pp.
251
262
.
25.
Nortier
,
B. P.
,
Voormeeren
,
S. N.
, and
Rixen
,
D. J.
,
2012
, “
Application of Residual Vectors to Superelement Modeling of an Offshore Wind Turbine Foundation
,”
Topics in Experimental Dynamics Substructuring and Wind Turbine Dynamics
(Conference Proceedings of the Society for Experimental Mechanics Series), Springer, New York, pp.
149
163
.
26.
Rice
,
S. O.
,
1944
, “
Mathematical Analysis of Random Noise
,”
Bell Syst. Tech. J.
,
23
(3), pp.
282
332
.
27.
Rice
,
S. O.
,
1945
, “
Mathematical Analysis of Random Noise
,”
Bell Syst. Tech. J.
,
24
(1), pp.
146
156
.
28.
Yang
,
J.-N.
, and
Shlnozuka
,
M.
,
1971
, “
On the First Excursion Probability in Stationary Narrow-Band Random Vibration
,”
ASME J. Appl. Mech.
,
38
(4), pp.
1017
1022
.
29.
Yang
,
J.-N.
,
1975
, “
Approximation to First Passage Probability
,”
J. Eng. Mech. Div., ASCE
,
101
(4), pp.
361
372
.http://cedb.asce.org/CEDBsearch/record.jsp?dockey=0006136
30.
Singh
,
A.
,
Mourelatos
,
Z.
, and
Nikolaidis
,
E.
,
2011
, “
Time-Dependent Reliability of Random Dynamic Systems Using Time-Series Modeling and Importance Sampling
,”
SAE Int. J. Mater. Manuf.
,
4
(
1
), pp.
929
946
.
31.
Andrieu-Renaud
,
C.
,
Sudret
,
B.
, and
Lemaire
,
M.
,
2004
, “
The PHI2 Method: A Way to Compute Time-Variant Reliability
,”
Reliab. Eng. Syst. Saf.
,
84
(
1
), pp.
75
86
.
32.
Hu
,
Z.
, and
Du
,
X.
,
2013
, “
Time-Dependent Reliability Analysis With Joint Upcrossing Rates
,”
Struct. Multidiscip. Optim.
,
48
(
5
), pp.
893
907
.
33.
Madsen
,
P. H.
, and
Krenk
,
S.
,
1984
, “
Integral Equation Method for the First-Passage Problem in Random Vibration
,”
ASME J. Appl. Mech.
,
51
(
3
), pp.
674
679
.
34.
Wang
,
Z.
, and
Wang
,
P.
,
2013
, “
A New Approach for Reliability Analysis With Time-Variant Performance Characteristics
,”
Reliab. Eng. Syst. Saf.
,
115
, pp.
70
81
.
35.
Wang
,
Z.
, and
Wang
,
P.
,
2012
, “
A Nested Extreme Response Surface Approach for Time-Dependent Reliability-Based Design Optimization
,”
ASME J. Mech. Des.
,
134
(
12
), p.
121007
.
36.
Wang
,
Z.
, and
Wang
,
P.
,
2015
, “
A Double-Loop Adaptive Sampling Approach for Sensitivity-Free Dynamic Reliability Analysis
,”
Reliab. Eng. Syst. Saf.
,
142
, pp.
346
356
.
37.
Jiang
,
C.
,
Huang
,
X.
,
Han
,
X.
, and
Zhang
,
D.
,
2014
, “
A Time-Variant Reliability Analysis Method Based on Stochastic Process Discretization
,”
ASME J. Mech. Des.
,
136
(
9
), p.
091009
.
38.
Jiang
,
C.
,
Huang
,
X.
,
Wei
,
X.
, and
Liu
,
N.
,
2017
, “
A Time-Variant Reliability Analysis Method for Structural Systems Based on Stochastic Process Discretization
,”
Int. J. Mech. Mater. Des.
,
13
(
2
), pp.
173
193
.
39.
Song
,
J.
, and
Der Kiureghian
,
A.
,
2006
, “
Joint First-Passage Probability and Reliability of Systems Under Stochastic Excitation
,”
J. Eng. Mech.
,
132
(
1
), pp.
65
77
.
40.
Taflanidis
,
A. A.
,
2010
, “
Reliability-Based Optimal Design of Linear Dynamical Systems Under Stochastic Stationary Excitation and Model Uncertainty
,”
Eng. Struct.
,
32
(
5
), pp.
1446
1458
.
41.
He
,
J.
,
2009
, “
An Approximation of the First Passage Probability of Systems Under Nonstationary Random Excitation
,”
Appl. Math. Mech. (English Ed.)
,
30
(
2
), pp.
255
262
.
42.
Radhika
,
B.
,
Panda
,
S. S.
, and
Manohar
,
C. S.
,
2008
, “
Time Variant Reliability Analysis of Nonlinear Structural Dynamical Systems Using Combined Monte Carlo Simulations and Asymptotic Extreme Value Theory
,”
CMES—Comput. Model. Eng. Sci.
,
27
(
1–2
), pp.
79
109
.
43.
Gupta
,
S.
, and
Manohar
,
C.
,
2005
, “
Multivariate Extreme Value Distributions for Random Vibration Applications
,”
J. Eng. Mech.
,
131
(
7
), pp.
712
720
.
44.
Gupta
,
S.
, and
Manohar
,
C.
,
2006
, “
Reliability Analysis of Randomly Vibrating Structures With Parameter Uncertainties
,”
J. Sound Vib.
,
297
(
3
), pp.
1000
1024
.
45.
Gupta
,
S.
,
Gelder
,
P.
, and
Pandey
,
M.
,
2006
, “
Time-Variant Reliability Analysis for Series Systems With Log-Normal Vector Response
,”
Adv. Eng. Struct., Mech. Construction
,
140
, pp.
747
759
.
46.
Hagen
,
O.
, and
Tvedt
,
L.
,
1991
, “
Vector Process Out-Crossing as Parallel System Sensitivity Measure
,”
J. Eng. Mech.
,
117
(
10
), pp.
2201
2220
.
47.
Choi
,
S. K.
,
Grandhi
,
R. V.
, and
Canfield
,
R. A.
,
2007
,
Reliability-Based Structural Design
,
Springer
, London.
48.
McDonald
,
M.
, and
Mahadevan
,
S.
,
2008
, “
Design Optimization With System-Level Reliability Constraints
,”
ASME J. Mech. Des.
,
130
(
2
), p.
021403
.
49.
Hu
,
Z.
,
Li
,
H.
,
Du
,
X.
, and
Chandrashekhara
,
K.
,
2013
, “
Simulation-Based Time-Dependent Reliability Analysis for Composite Hydrokinetic Turbine Blades
,”
Struct. Multidiscip. Optim.
,
47
(
5
), pp.
765
781
.
50.
David
,
J.
, and
Laino
,
A. C. H.
,
2002
, “
National Renewable Energy Laboratory Report, Appendix B
,” National Renewable Energy Laboratory, Golden, CO, Report No. NREL/TP-442-7817.
51.
Hu
,
Z.
, and
Du
,
X.
,
2013
, “
A Sampling Approach to Extreme Value Distribution for Time-Dependent Reliability Analysis
,”
ASME J. Mech. Des.
,
135
(
7
), p.
071003
.
52.
Hu
,
Z.
, and
Du
,
X.
,
2013
, “
Reliability Analysis for Hydrokinetic Turbine Blades Under Random River Velocity Field
,”
Seventh Annual ISC Research Symposium
(
ISC-GRS
), Rolla, Missouri, Apr. 24.http://reliadesign.net/wp-content/uploads/2015/02/RELIABILITY-ANALYSIS-FOR-HYDROKINETIC-TURBINE-BLADES-UNDER-RANDOM-RIVER-VELOCITY-FIELD.pdf