Reliability: Modeling, Prediction, and Optimization

CHAPTER 11

Model Selection and Validation

11.1 INTRODUCTION

As is apparent in the material covered in the previous chapters, modeling plays a very important role in reliability analysis. In Chapter 1, we began with the system characterization and corresponding mathematical models. Subsequently, many additional models were discussed, including models of failure mechanisms, time to failure, system failure as a function of component failures, and models representing data, e.g., those used in regression and ANOVA. In this chapter, we are concerned mainly with probabilistic models, and are concerned with selection and validation of an appropriate model. By this is meant choosing a “best” model from a set of candidate models and then validating this choice, i.e., providing evidence that the chosen model is, in fact, adequate and valid for the intended reliability objective.

In Chapter 4, we dealt with probability distributions used in reliability. These are basically models representing the probabilistic structure of time to failure and related random quantities. In Chapters 6 and 7, reliability models for components and multicomponent systems were developed and analyzed. These were based on the system structure and, where possible, on the underlying failure mechanisms of the item.

In cases where the underlying failure mechanism is well understood, the appropriate probability distribution can be deduced, and white-box models are used in the analysis. In situations where this is not the ...