Application of an Effective Data-Driven Approach to Real-time time Fault Diagnosis in Automotive Engines

A dominant thrust in modern automotive industry is the development of "smart service systems" for the comfort of customers. The current on-board diagnosis systems embedded in the automobiles follow conventional rule-based diagnosis procedures, and may benefit from the introduction of sophisticated artificial intelligence and pattern recognition-based procedures in terms of diagnostic accuracy. Here, we present a mode-invariant fault diagnosis procedure that is based on data -driven approach, and show its applicability to automotive engines. The proposed approach achieves high-diagnostic accuracy by detecting the faults as soon as they occur. It uses statistical hypothesis tests to detect faults, a wavelet-based preprocessing of the data, and pattern recognition techniques for classifying various faults in engines. We simulate the Toyota Camry 544N Engine SIMULINK model in a real-time simulator and controlled by a prototype ECU (Electronic Control Unit). The engine model is simulated under several operating conditions (pedal angle, engine speed, etc), and pre-and post-fault data is collected for eight engine faults with different severity levels, and a database of cases is created for applying the presented approach. The results demonstrate that appealing diagnostic accuracy and fault severity estimation are possible with pattern recognition-based techniques, and, in particular, with the support vector machines.