MVEM-Based Fault Diagnosis of Automotive Engines Using Dempster–Shafer Theory and Multiple Hypotheses Testing

Internal combustion engines exhibit fast pulsating short-time dynamics due to the reciprocating cylinder motion, around mean operating points that change comparatively slow due to inputs such as throttle and load. Comparatively, simple mean value engine models (MVEM) describe the slow changes of the averaged states for automotive control and fault diagnosis. In this paper, a bank of state estimators based on MVEMs is used for fault residual generation. Three faults: 1) throttle mass air-flow sensor fault; 2) exhaust gas recirculation valve sensor fault; and 3) exhaust leak fault are considered here. These faults are significant as they affect emission levels. Optimized thresholds for residual classification are derived for minimizing false alarm rates and missed detection rates. The diagnosis logic, based on the principles of structured residuals proposed in literature, is extended here for multiple hypotheses testing. Furthermore, the Dempster-Shafer theory is used to associate a confidence measure with the decision conclusions and this is shown to improve isolation. Performance is demonstrated with automotive engine data obtained from a four-cylinder instantaneous spark-ignition engine (gasoline) system model, developed in the simulation software AMESim.