A Fault Diagnosis Strategy for Winding Inter-Turn Short-Circuit Fault in Doubly Salient Electro-Magnetic Machine Based on Mechanical and Electrical Signal Fusion

Stator winding inter-turn short-circuit fault is a frequent failure on doubly salient electro-magnetic machine (DSEM). It is difficult to identify the fault accurately by the single fault characteristic because of the less influence by fault windings on the electromagnetic field. Failure to detect a fault condition at an early stage of fault will pose a huge threat to the entire motor system. This paper proposes a comprehensive short-circuit fault diagnosis method for DSEM based on mechanical and electrical signal fusion. The short-circuit fault can be identified effectively by analyzing the structure of the machine and the extraction mechanism of its fault features. Firstly, the mathematical model of distributed excitation DSEM was established to analyze the changes of current and vibration signals in healthy and fault states, and the extraction mechanism of current fundamental frequency amplitude and vibration signals was studied and adopted for motor fault diagnosis. Secondly, when the inter-turn short-circuit fault occurs, according to the characteristics of current and vibration signals, the current signals are input into SVM and the vibration signals are input into the improved CNN model to obtain fault probability under. Finally, the classification fault diagnosis results of multi-source signals are obtained by using the evidence combination rule of D-S evidence theory. In this integrated model, the problem of low accuracy of single-source diagnosis method is solved, and the model has good reliability, complementarity and high-accuracy. The results of the experiments conducted on the DSEM platform show that the fault diagnosis experiments using SVM alone are relatively limited in practice, making it difficult to distinguish between faults with small imbalance coefficients and normal conditions. Correspondingly, the SVM is very sensitive to diagnose faults with larger imbalance coefficients and has a high diagnostic accuracy. The vibration signal detected by CNN can make up for the lack of single-source diagnosis. CNN has a relatively impressive accuracy in diagnosing all three operational states. For example, when the fault F1 occurred, the single-source diagnosis method using SVM alone was misjudged, with a probability of 0.668 for normal and 0.33 for fault F1. The CNN method using the vibration signal as input has a probability of 0.227 for normal and 0.736 for fault F1. And the correct diagnostic result is obtained by fusing the two methods using D-S evidence theory in case of inconsistent results. When the diagnosis results of both single-source methods are correct, the fusion can output the correct label with higher probability. From the 100 test sets, the accuracy of the proposed method reaches 95%, which is more reliable, more credible and more robust than the fault diagnosis methods based on single signal. The following conclusions can be drawn from the experimental analysis:① When the inter-turn short-circuit fault occurs in the DSEM, the fundamental frequency amplitude of the four-phase current will no longer be symmetrical and the difference will be larger. The electromagnetic torque will generate torque harmonics of the 4kth harmonic, which will cause the vibration signal of the fault to change. ② The CNN network in the multi-source signal fusion diagnosis method is improved to find the optimal dropout parameters, which can ensure the success rate of neural network training and prevent the network from overfitting. ③ The proposed multi-source signal fusion fault diagnosis method can complement the advantages of two single-signal fault diagnosis methods, and greatly improve the fault diagnosis accuracy with good robustness and accuracy. © 2023 Chinese Machine Press. All rights reserved.