Fault Diagnostics Using Network Motif Signature

Abstract


 
 
 In modern vehicles, controls are distributed over multiple Electronic Control Units (ECUs) that are connected through in-vehicle communication networks. Fault diagnostics for such a distributed control system is very challenging, which has resulted in many no-trouble-found (NTF) cases during warranty repairs. To address this problem, we propose a novel network-theoretic approach that detects, identifies, and localizes faults using both the structure of the communication network (topological information) and message flow information. The proposed method not only enables the characterization of normal operation and a-priori known faults across communication networks, which is already beyond the current practice of individual ECU centric diagnostics, but also the diagnostics of unknown or cascading failures emerging from unexpected operational environments.