| Asynchronous motor is one of the important power sources in production and life.Motor failure is easily caused by frequent start and stop,overload operation and component wear failure,which is easy to cause serious consequences such as safety accidents,and economic losses.Therefore,it is of great significance to study motor fault diagnosis technology deeply.There are problems that poor diagnosis effect under variable working conditions and unreliable diagnosis result with single signal in the existing motor intelligent diagnosis algorithm.To solve it,this thesis introduces signal analysis,deep learning,transfer learning,and information fusion methods to study the fault diagnosis algorithm of asynchronous motor.The main research contents are as follows:(1)The fault diagnosis methods based on traditional machine learning rely on professional experience to select statistical features,and the diagnosis results are greatly affected by human factors.Aiming at this problem,the motor fault diagnosis method based on the deep residual network is being studied.The three-phase current input strategy of different modes is designed,and the motor fault diagnosis model of deep learning based on feature adaptive extraction is established.So the deep residual network can more effectively extract the fault depth characteristics of motor current signals.Experimental results show that compared with traditional machine learning,the depth feature can be automatically mined by deep learning.When applied to motor fault diagnosis,the diagnosis accuracy of the deep learning algorithm is higher than traditional machine learning.(2)The actual operating conditions of the motor are complex and changeable.The spatial distribution of signal fault characteristics is easy to change with the change of operating conditions,resulting in a decrease in the diagnostic accuracy of the fault diagnosis model.To solve this problem,the deep migration learning method is studied.And two kinds of motor fault diagnosis models under variable working conditions based on parameter transfer and feature transfer are designed.The adaptability of the diagnostic model under different working conditions was improved from the perspective of model parameter adjustment and feature space distribution adaptation.Experimental analysis shows that the two migration algorithms can improve the accuracy of fault diagnosis under variable working conditions.And the diagnosis effect of the feature migration model is better than that of the parameter migration model.(3)Motor faults include electrical faults and mechanical faults.It is easy to misidentify mechanical faults only by using current signals.To solve this problem,vibration signals and current signals are simultaneously used for motor fault diagnosis.And the motor fault diagnosis model based on multi-source information fusion is established.The deep residual network is used to extract and diagnose current signals and vibration signals,respectively.On this basis,the decision-making layer fusion method of multi-source information is introduced,and the improved D-S evidence theory is used to fuse the recognition results and output the final classification results.Experimental analysis shows that the deep learning motor fault diagnosis method based on multi-source information fusion can improve the reliability of diagnosis results. |
