Model based automated diagnosis of bearing knock faults in internal combustion engines

Abstract

Intelligent diagnosis of bearing knock faults in Internal Combustion Engines (IC engines) was studied in this paper. Because of previous successful application of Artificial Neural Networks (ANNs) to the condition monitoring of rotating machinery, an ANN based automated diagnosis system was proposed for the diagnosis of big-end bearing knock faults in IC engines. It consists of three separate ANNs: a fault detection network, a fault localization network, and a fault severity identification network. In order to solve the problem that ANNs need a lot of data for training, a simulation model was built to simulate various degrees of bearing knock faults. The impact forces of the bearing with different clearance were simulated first, and then the accelerations at the measurement point on the engine block were calculated. A series of experiments were also carried out, and the results were used to evaluate and update the simulation model. It was also found that the squared envelope signals, rather the raw acceleration signals, have more useful diagnostic information. The extracted/selected amplitude features were used for fault detection and severity identification, and the extracted/selected phase features were used for fault localization. It is worth pointing out that because a saturating linear function was selected as the transfer function of the ANN for the fault severity identification stage, the networks can linearly classify the fault levels and the output agrees better with the real situation. All the networks were trained using simulated data and tested using experimental data. The final results have verified that the system could efficiently diagnose bearing knock faults, especially the accurate identification of the fault levels.