Fault detection of a diesel engine through vibration source separation and deep learning

Abstract Failures in a diesel engine will cause massive damage to the machine, so it is essential to monitor and detect unexpected faults in the diesel engine. Because the diesel engine vibration signal is a nonlinear superposition of multiple vibration sources, traditional vibration identification can no longer meet the growing requirements for monitoring accuracy. For this reason, a new method is proposed in this study based on the EMD - nonlinear ICA to separate the noise and other interference signal sources from the engine vibration signal and obtain the fault-related vibration sources; then, a deep learning model is built to identify the fault types. This test assessed the proposed method for diesel engine fault detection. The results signify that the signal separation method can select the fault vibration source from the engine vibration signal and correctly identify the engine faults.