Applications of cluster analysis in diagnostics-related problems

System diagnostics facilitates reliability enhancement and condition-based maintenance of technical systems. Diagnostic problems can be formulated and solved only on the basis of mathematical models reflecting the complex stochastic nature of the failure development process. This process is affected by numerous continuous factors, but its outcome constitutes a random event. These specifics limit the applications of traditional regression models. The concept of a "continuous process with discrete-event output" is introduced, and cluster analysis is employed as a modeling approach facilitating the solution of various diagnostic problems. Important aspects of cluster analysis are the definition of the informativity criterion, selection of "informative subspaces", definition of separating rules, and finally, formulation of a cluster model are presented. Various enhancements of cluster analysis are proposed. A cluster model is utilized for the definition of so-called "probabilistic space" that, in conjunction with Bayes' technique, facilitates the solution of failure prediction problems. Application of cluster models for failure analysis and prediction is illustrated by numerical examples.