Condition-based maintenance with multi-target classification models

Condition-based maintenance (CBM) recommends maintenance actions based on the information collected through condition monitoring. In many modern cars, the condition of each subsystem can be monitored by onboard vehicle telematics systems. Prognostics is an important aspect in a CBM program as it deals with prediction of future faults. In this paper, we present a data mining approach to prognosis of vehicle failures. A multitarget probability estimation algorithm (M-IFN) is applied to an integrated database of sensor measurements and warranty claims with the purpose of predicting the probability and the timing of a failure in a given subsystem. The results of the multi-target algorithm are shown to be superior to a singletarget probability estimation algorithm (IFN) and reliability modeling based on Weibull analysis.