Data-driven Fault Diagnosis Method for Transmission Sensors

Abstract:

Aiming at the limitations of model-based and rule-based fault diagnosis methods， a data-driven fault diagnosis method for transmission sensors was proposed. First， a residual sequence was obtained between the output of actual sensor and the output of sensor model established by step-wise regression. Then， the residual sequence was decomposed by wavelet packet transform（WPT）， and the Shannon entropy of each node was calculated as the feature values. Finally， a probabilistic neural network（PNN） was adopted to identify the feature values of different sensor faults. This method is verified by transmission signals from hardware-in-the-loop platform. Results indicate that the method has a diagnostic accuracy of 98.50%， and the diagnostic accuracy varies little under different sample divisions. In addition， the fault diagnoses of two speed sensors were also performed， and the diagnostic accuracy is at a relatively high value， which proves the applicability of the method.