International Journal of Reliability, Quality and Safety Engineering

In this note, the early research in reliability is reviewed. The focus is on the theoretical development of aging and reliability bounds. Some conceptual issues with the interpretation of subjective reliability are also discussed.

Throughout the developed world the problem of how best to manage aging assets is being faced. It is particularly critical in distribution systems such as mains water reticulation, sewage networks and electricity supply lines. In many towns, some underground water and sewer pipelines are more than 50 years old. It is already prohibitive on cost grounds to replace all these deteriorating assets, so a cost-effective replacement/repair strategy must be used. Reliability modeling-based solutions to this problem are discussed, together with issues related to the quality of the available data. Some of the outcomes for industry are improved monitoring and record keeping, objective evaluation of alternative management options, and reliability prediction to enable sound planning decisions to be made. The applicability of these methods to other distribution systems is also discussed.

This study discusses an opportunity-based age replacement policy for a system which has a warranty period (0, S]. When the system fails at its age x≤S, a minimal repair is performed. If an opportunity occurs to the system at its age x for S<x<T, we take the opportunity with probability p to preventively replace the system, while we conduct a corrective replacement when it fails on (S, T). Finally if its age reaches T, we execute a preventive replacement. Under this replacement policy, the design variable is T. For the case where opportunities occur according to a Poisson process, a long-run average cost of this policy is formulated under a general failure time distribution. It is, then, shown that one of the sufficient conditions where a unique finite optimal T* exists is that the failure time distribution is IFR (Increasing Failure Rate). Numerical examples are also presented for the Weibull failure time distribution.

Suppose that shocks arrive and act on a system according to a Poisson distribution with mean rate of arrival equal to λ shock(s) per unit time. A δ-shock failure model is proposed in this paper, which assumes that when a system is acted on by a shock, it will recover fully in time δ(>0), and after that it will function as if no shock had occurred before. If the time lag between two successive shocks is less than δ, the second shock will cause failure of the system. Theoretical expressions related to the distribution of the failure time of the system are derived. These results can be used to optimize the design of a system from a costing point of view.

A structural model for the estimation of software failure rates is proposed which is based on a partition of the code into components. Systematic failure is assumed to be induced by the interaction of these components. A Bayesian inference scheme is used to perform failure rate estimation on the basis of N failure free test runs. The approach splits the problem of constructing a system prior up into the smaller, conceivably simpler problems of constructing subtask priors. Thereby a wider range of prior information is used in the process of assessing system safety. The work in this paper constitutes a first step towards a formal statistical understanding of the relationship between system complexity and system testing for reliability. Long-term implications are the achievement of more informative reliability estimates and guidelines on cost-effective testing.

We present two alternative perspectives to the current way of planning for constant-stress accelerated life tests (CSALTs) and step-stress ALT (SSALT). In 3-stress CSALT, we consider test plans that not only optimize the stress levels but also optimize the sample allocation. The resulting allocations also limit the chances of inconsistency when data are plotted on a probability plot. For SSALT, we consider test plans that not only optimize both stress levels and holding times, but also achieve a target acceleration factor that meets the test time constraint with the desirable fraction of failure. The results for both problems suggest that the statistically optimal way to increase acceleration factor in an ALT is to increase lower stress levels and; in the case of CSALT, to decrease their initial sample allocations; in the case of SSALT, to reduce their initial hold times. Both problems are formulated as constrained nonlinear programs.

Rapidly evolving accelerated degradation testing (ADT) has provided an efficient channel for reliability assessment without the labor of conventional life testing that observations are complete or incomplete data. In step-stress loading, a specimen is subjected to successively higher level stress until an appropriate termination time. All specimens go through the same specified pattern of stress levels and testing time. A step-stress degradation testing (SSDT) is corresponding to collect the degradation measurements over time from a step-stress life testing before any specimen fail. The advantages of SSDT may shorten the testing time and reduce the test units compared to ADT. Conventional statistical approach for modeling the step-stress degradation process may be complicated. In this paper, we address a nonparametric approach by applying the methodology of neural networks to calibrate the step-stress degradation measurements and conducting the reliability prediction at use condition. A light emitting diode (LED) example is given to illustrate the implementation procedure.

Manufacturing processes often consist of a number of sequential stages. Of interest is to control the variation in one or more quality characteristics of a production unit at the final stage. By understanding how variation is transmitted and added across the stages, remedial actions in reducing variation at the final stage can be properly planned. With one quality characteristic measured at each stage, a set of naive estimators is previously proposed and shown to perform indistinguishably well with maximum likelihood estimators. Thus naive estimators are more convenient than maximum likelihood estimators as the former exist in closed form while the latter do not. This article considers situations when more than one quality characteristic is measured throughout the stages. Methods of analyzing variation transmission are briefly reviewed and the finite sample properties of naive and maximum likelihood estimators for multivariate measurements are further examined. A broad conclusion is that for moderate number of production units, naive estimators have smaller bias and variability. Furthermore, "proper" naive estimates provide more accurate interval estimates at a given confidence level. Finally, a set of piston-machining data is used for illustration.