基于机电信号融合的电励磁双凸极电机绕组匝间短路故障诊断

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

定子绕组匝间短路是电励磁双凸极电机常见的故障,当匝间短路故障匝数较少时,对电机磁场的影响较小,使得难以通过单一故障特征精确地识别故障。该文通过分析该电机的本体结构和故障特征的提取机理,提出了一种基于振动和相电流信号的多源机电信号融合的电励磁双凸极电机短路故障综合诊断方法——基于多分类支持向量机及改进卷积神经网络的信号融合故障诊断。该文首先根据电励磁双凸极的结构特点研究了短路故障发生时特征信号的变化情况及提取机理;其次根据理论分析结果提出了一种自寻优卷积神经网络结合支持向量机的多源信号融合的故障诊断方法,该方法首先分别利用两种算法计算两个单信号特征下的故障诊断概率,再利用DS证据理论,得出融合后的诊断结果;最后通过实验表明,该文所提出的诊断方法可以更有效地识别电机不同的短路故障,降低了误判率,解决了单一信号源诊断方法精度较低的问题,具有良好的可靠性和互补性。

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 multisource 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.