A generalized similarity measure for similarity-based residual life prediction

Abstract

Equipment residual life (RL) prediction is one of the essential elements of condition-based maintenance. Recently, some efforts are targeted towards similarity-based RL prediction. Similarity-based RL prediction approaches can ease the burden of degradation modelling and perform long-term prediction, by predicting a system's RL using weighted RL of similar systems. In this article, a generalized similarity measure is proposed for a similarity-based RL prediction approach. By the proposed similarity measure, more weight is assigned to a system's most recent performance than to its former performance when measuring its similarity with other systems. The proposed similarity measure includes the conventional one as a special case, hence the name generalized similarity measure. The implementation results from (a) a numerical experiment based on a typical degradation model and (b) a case study of RL prediction of ball grid array solder joints in vibration environments provides evidence of the superiority of the generalized similarity measure, in terms of statistically more accurate prediction (i.e. smaller mean prediction error) close to system failure.