The use of discriminant function techniques in reliability assessment and data classification

Abstract

In some cases of reliability assessment and data classification, the influence of loading and environmental factors are not difficult to see. In many instances, however, this is not the case, and particularly with mechanical equipment the relevance of such factors can be almost impossible to define. In this situation the significance of each can often be quantified through the use of statistical discrimination techniques. Discrimination function techniques are described as a means of classifying reliability information in a meaningful form. The techniques and software described offer a method of quantifying the significance of environmental parameters as a measure of their influence to some function, which in most reliability problems will be to one or more modes of failure or success. The process is ideally suited to multi-variable dependent problems and results in a series of weighting parameters which quantify the relevance of each variable to the overall outcome. The techniques are demonstrated on several examples of mechanical seal data and shown to produce discrimination, and variable ranking information useful to both data classification and failure prediction.