Distribution‐free inferences on system reliability

Bayesian reliability analysis (BRA) technique has been actively used in reliability assessment for engineered systems. However, there are two key controversies surrounding the BRA: the reasonableness of the prior and the consistency among all data sets. These issues have been debated in Bayesian analysis for many years. As we observed, they have not been resolved satisfactorily. These controversies have seriously hindered the applications of BRA as a useful reliability analysis tool to support engineering design. In this paper, a Bayesian reliability analysis methodology with a prior and data validation and adjustment scheme (PDVAS) is developed to address these issues. As the part of the PDVAS development, a consistency measure is first defined that judges the level of consistency among all data sets including the prior. The consistency measure is then used to adjust either the prior or the data or both to the extent that the prior and the data are statistically consistent. This prior and data validation and adjustment scheme is developed for Binomial sampling with Beta prior, called Beta-Binomial Bayesian model. The properties of the scheme are presented and discussed that provides some insights of PDVAS. Various forms of the adjustment formulas are shown, and a selection framework of a specific formula, based on engineering design and analysis knowledge, is established. Several illustrative examples are presented, which show the reasonableness, effectiveness, and usefulness of PDVAS. General discussion of the scheme is offered to enhance the Bayesian reliability analysis in engineering design for reliability assessment.

Bayesian approaches have been demonstrated as effective methods for reliability analysis of complex systems with small-amount data, which integrate prior information and sample data using Bayes’ theorem. However, there is an assumption that precise prior probability distributions are available for unknown parameters, yet these prior distributions are sometimes unavailable in practical engineering. A possible way to avoiding this assumption is to generalize Bayesian reliability analysis approach by using imprecise probability theory. In this paper, we adopt a set of imprecise Dirichlet distributions as priors to quantify uncertainty of unknown parameters and extend traditional Bayesian reliability analysis approach by introducing an imprecise Dirichlet model (IDM). When the prior information is rare, the result of imprecise Bayesian analysis method is too rough to support engineering decision-making, so we proposed an optimization model to reduce the imprecision of the new method. Spindles are crucial for machine tools and reliability data related to spindles of new-developed machine tools are often rare. We can then use the imprecise Bayesian reliability analysis method to assess its reliability. In this paper, we mainly investigate the reliability assessment of a motorized spindle to illustrate the effectiveness of the proposed method.

Bayesian system reliability assessment method with maximum entropy as prior distribution is developed in the present analysis. The system reliability assessment model is established, and the method and process of system reliability assessment are put forward. The method is based on the theory of information entropy. The probability density function and distribution function of system reliability parameters are given by using maximum entropy method. Prior distribution of Bayesian analysis method is the distribution function. Likelihood function is given by a small amount of field test data. According to Bayesian methods, posterior distribution is determined by using prior distribution and likelihood function. The point estimation and interval estimation of system reliability parameters are calculated in order to evaluate the system reliability. The method solves better the problem of system reliability assessment for small sample cases, and is an optimum method for reliability assessment of large and complex systems.

Prior information and elicitation are the prerequisite in Bayesian reliability inference. Multiple sources for priors such as probability fitting based on historical data and expert judgment are often available when estimating the reliability of complex systems. This paper investigates the integration of multiple priors in Bayesian reliability analysis. Specifically, methods for multiple priors' integration based on Bayesian Melding are investigated. The performance of the studied methods with different prior information integration algorithms such as the arithmetic and geometric averaging is investigated. The impacts of the prior misspecification and the pooling parameter selection for prior integration algorithms are also studied. In numerical examples, simulation methods are applied for posterior reliability inference under the proposed prior integration methods and the performance of the two methods are compared.

An approach for simulating models of highly dependable systems with general failure and repair time distribution is described. The approach combines importance sampling with event rescheduling in order to obtain variance reductions in such rare event simulations. The approach is general in nature and allows a variety of features commonly arising in dependability modeling to be simulated effectively. It is shown how the technique can be applied to systems with redundant components and/or periodic maintenance. For different failure time distributions, the effect of the maintenance period on the steady-state availability is explored. The amount of component redundancy needed to achieve a certain reliability level is determined. >

Failure Models. Qualitative System Analysis. Systems of Independent Components. Component Importance. Markov Models. Counting Processes. Dependent Failures. Life Data Analysis. Accelerated Life Testing. Bayesian Reliability Analysis. Reliability Data Sources. Appendices. References. Indexes.

A two-component parallel or series system is studied where the failure time of each component follows the Burr distribution. The testing environment of the system is assumed different from the operating one, causing the parameters (i=1, 2) to increase or decrease by a positive, random amount η. Two priors are considered for the behaviour of η, a gamma and a uniform. The Bayesian reliability is derived in closed form. Three hypothetical examples show how this approach could be used by reliability engineers.

Reliability analysis is the key to evaluate software’s quality. Since the early 1970s, the Power Law Process, among others, has been used to assess the rate of change of software reliability as time-varying function by using its intensity function. The Bayesian analysis applicability to the Power Law Process is justified using real software failure times. The choice of a loss function is an important entity of the Bayesian settings. The analytical estimate of likelihood-based Bayesian reliability estimates of the Power Law Process under the squared error and Higgins-Tsokos loss functions were obtained for different prior knowledge of its key parameter. As a result of a simulation analysis and using real data, the Bayesian reliability estimate under the Higgins-Tsokos loss function not only is robust as the Bayesian reliability estimate under the squared error loss function but also performed better, where both are superior to the maximum likelihood reliability estimate. A sensitivity analysis resulted in the Bayesian estimate of the reliability function being sensitive to the prior, whether parametric or non-parametric, and to the loss function. An interactive user interface application was additionally developed using Wolfram language to compute and visualize the Bayesian and maximum likelihood estimates of the intensity and reliability functions of the Power Law Process for a given data.

The paper begins with an arbitrary single-output, discrete-signal, irredundant system consisting of interconnected computing units each of which has the same exponential distribution for failure time. Then it presents several corresponding redundant designs that compute the same function, derives their distributions for failure time, and illustrates them with quantitative examples. None of the straight replicative designs very much improves the original distribution for failure time; nor do the releasing designs do better. The standby designs do best, even after the standby management-system failure distribution is factored into the system failure distribution. Replicate releasing designs, as a group, are better than straight replicate designs; and standby designs are best. >

Bayesian methods have become increasingly popular in recent years because of its central application in the new technologies related to artificial intelligence and machine learning. Here, a brief introduction to Bayesian thinking and Bayesian methods in reliability analysis is given with focus on the basic concepts and methods. These are illustrated by models including a single parameter. A number of practical examples are used to illustrate the concepts and methods. After having read this chapter, we hope the reader has got a sufficient background to probe further into specialized texts on Bayesian reliability analysis.

Components within a system may be related to one another in two primary ways: in either a serial or a parallel configuration. The other way associated system components is developed on the basis of series and parallel systems. Although some complex systems cannot generate from series and parallel system, the result of this paper is helpful to study more complex gradual system reliability problems. In this paper, a hypothesis is formed that the reliability of the components of the system is variable with time. Then, the reliability of the system is gradual, on condition that the components are combined by some related way. If the system reliability analysis is based on the static analysis, it may be different from the actual value of the reliability because the components of the system would afford stochastic load or generate degradation of performance. The method considering the random resistant degradation may make the result of reliability accurate. In this paper, gamma process is used to simulate the use of the components of the system, as deterioration is generally uncertain and non-decreasing, it can best be regarded as a gamma process. Combining the Edgeworth series method and forth moment technique, the method that calculate the reliability of series and parallel system is deduced. Then, the reliability of different systems and the failure rate of the systems are illustrated with time. The results show that the reliability of the systems will decline in different ways with time.

Extensions of the conjugate prior through the Kullback–Leibler separators

The failure and repair of modules in an N-modular redundant (NMR) system are governed by a failure time distribution and a repair time distribution, respectively. It is generally reasonable to assume that a module's failure time distribution is a simple exponential distribution. However, it is not reasonable to assume that the repair time distribution is also exponential. Reliability models with non-exponential repair have a higher computational complexity than a model of the same system with an exponential repair time distribution. This paper presents the results of a systematic study to determine whether non-exponential repair distributions produce significant differences in calculated NMR system unreliability, relative to an exponential repair distribution with the same mean time to repair (MTTR). Our approach is to embed Erlang repair distributions in generalized stochastic Petri net (GSPN) models of NMR systems and evaluate the unreliability. Our results show that, for a wide range of system parameters, the choice of a repair time distribution has minimal impact on the calculated unreliability. Rather, it is the MTTR that is the dominant parameter affecting unreliability.

Fatigue and rutting are two primary failure mechanisms in asphalt pavements. The evaluations of fatigue and rutting performances are significantly uncertain due to large uncertainties involved with the traffic and pavement life parameters. Therefore, deterministically it is inadequate to predict when an in-service pavement would fail. Thus, the deterministic failure time which is known as design life (yr) of pavement becomes random in nature. Reliability analysis of such time (t) dependent random variable is the survival analysis of the structure. This paper presents the survival analysis of fatigue and rutting failures in asphalt pavement structures. It is observed that the survival of pavements with time can be obtained using the bathtub concept that contains a constant failure rate period and an increasing failure rate period. The survival function (S(t)), probability density function (pdf), and probability distribution function (PDF) of failure time parameter are derived using bathtub analysis. It is seen that the distribution of failure time follows three parametric Weibull distributions. This paper also works out to find the most reliable life (YrR) of pavement sections corresponding to any reliability level of survivability.

Many models have been proposed that relate failure times and stochastic time-varying covariates. In some of these models, failure occurs when a particular observable marker crosses a threshold level. We are interested in the more difficult, and often more realistic, situation where failure is not related deterministically to an observable marker. In this case, joint models for marker evolution and failure tend to lead to complicated calculations for characteristics such as the marginal distribution of failure time or the joint distribution of failure time and marker value at failure. This paper presents a model based on a bivariate Wiener process in which one component represents the marker and the second, which is latent (unobservable), determines the failure time. In particular, failure occurs when the latent component crosses a threshold level. The model yields reasonably simple expressions for the characteristics mentioned above and is easy to fit to commonly occurring data that involve the marker value at the censoring time for surviving cases and the marker value and failure time for failing cases. Parametric and predictive inference are discussed, as well as model checking. An extension of the model permits the construction of a composite marker from several candidate markers that may be available. The methodology is demonstrated by a simulated example and a case application.