Problem of Model Selection for Estimation of Equipment Remaining Life

This work discusses an approach to solve a problem of estimation of equipment remaining life based on historical failure records. An approach typically adopted to solve this problem is based on parametric models of equipment lifetime, fitted to failure data. Since the lifetime data is most often censored, the model fit as well as goodness of fit procedures should handle censored observations. While fitting a parametric model to censored data is readily available in statistical packages, an objective way to select the best fit model is lacking in most data analysis tools, even those dedicated to reliability analysis. Goodness of fit in the censored data case is verified based on probability plots, which makes the task of best fit model selection rather subjective. This work presents a method to make this procedure more objective. The proposed approaches are illustrated by a numerical example based on real life data.