Novelty detection framework for monitoring connected vehicle systems with imperfect data
Mohammad Badfar, Murat Yildirim, Ratna Babu Chinnam · International Journal of Production Research · 2025
Shrinking product development cycles and increasing vehicle complexities necessitate a new generation of monitoring and diagnostic algorithms that can demonstrate increased autonomy and adaptivity. Conventional approaches, which make strict assumptions about data fidelity and failure ground-truth availability, face challenges in modern connected vehicle applications. This paper proposes a novelty detection-based autonomous monitoring framework that flags anomalies under sparse and noisy data with limited or no access to ground-truth information. The framework proposes an optional mechanism for extracting age-degrading features and offers a robust approach for fusing the output of heterogeneous novelty detectors to determine the health state of target components. We validate the proposed framework using connected vehicle data for 12-volt battery systems employed by a large fleet of commercial vehicles of a global automotive manufacturer. To demonstrate versatility, we also tested the framework on bench-testing data from LFP/graphite battery cells. Results demonstrate the effectiveness of the proposed framework.