Analysis of contaminated field failure data for repairable systems

Abstract

SUMMARYIn this paper we present a simple model for electronic systems with repair, and a method for analysing recorded field failure data for such systems. The work performed has resulted in analytical results that may be used for assessing the product reliability. The method was originally developed for use under quite ideal circumstances, but in this paper the method is adapted for use also with "contaminated" data, i.e., data where the failure times are observed imbedded by noise. We propose a simple model for the noise that enables an analytical solution for the mean cumulative number of failures. The expression is compared with the expression for non-contaminated data, for the purpose of assessing the effect of the contamination, and for the purpose of correcting estimates in the parametric model. The method is illustrated by a real world example of industrial failure data recorded on field basis, and the effect caused by contamination in this data is investigated under a "Worst Case" assumption. The example indicates that the model is quite robust to contamination. Only one of the three key parameters is affected significantly by assuming the data to be contaminated.