ML-based vehicle downtime reduction: A case of air compressor failure detection

Abstract

This paper addresses the critical issue of reducing the downtime of vehicles by using machine learning (ML) techniques and full life cycle data. The study tests the failure prediction of one major automotive system, air compressors in long-distance trucks. To validate the learning capabilities of ML models, three different algorithms such as C5.0, C5.0 with boosting and classification and regression tree (CART) are used. The findings suggest that the C5.0 model with boosting provides a better prediction of air compressor failure than other decision trees such as CART and C5.0. The diagnostic trouble codes (DTCs) and the vehicle operational variables are important indicators of the air compressor failure whereas the vehicle configuration data are not meaningful. A hybrid model which includes both DTCs and vehicle operational data is able to generate superior prediction results.