Early Warning from Car Warranty Data using a Fuzzy Logic Technique

Abstract

Car manufacturers are interested to detect evolving problems in a car fleet as early as possible so they can take preventive actions and deal with the problems before they become widespread. The vast amount of warranty claims recorded by the car dealers makes the manual process of analyzing this data hardly feasible. This chapter describes a fuzzy-based methodology for automated detection of evolving maintenance problems in massive streams of car warranty data. The empirical distributions of time-to-failure and mileage-to-failure are monitored over time using the advanced, fuzzy approach to comparison of frequency distributions. The authors’ fuzzy-based early warning tool builds upon an automated interpretation of the differences between consecutive histogram plots using a cognitive model of human perception rather than “crisp” statistical models. They demonstrate the effectiveness and the efficiency of the proposed tool on warranty data that is very similar to the actual data gathered from a database within General Motors.