Research On Motor Fault Diagnosis Method Based On Deep Learning Theory

As an indispensable equipment for industrial and agricultural production,motor plays a very important role.If a fault occurs during the operation of the motor,it will not only cause economic losses,but even threaten people's lives.At present,most traditional fault diagnosis methods use signal processing to extract feature vectors.The extraction process requires a large amount of signal processing methods and diagnostic experience,which makes the diagnosis process relatively complicated and leads to an increased possibility of misjudgment.Therefore,it is of great significance to study the advanced fault diagnosis methods of motors by converting the regular maintenance of motors into predictive maintenance and reducing maintenance costs.In recent years,with the rapid development of artificial intelligence,Deep learning demonstrates the ability to handle complex tasks in multiple areas.This dissertation focuses on two models of deep learning and its application in the field of fault diagnosis.The main research work and contents are as follows:(1)The basic theory and common methods of deep learning are systematically studied.because the shallow machine learning method needs a lot of prior knowledge and signal processing theory,and the generalization ability is weak,two models of Long Short-Term Memory neural networks and Stack Sparse Auto-Encoder in deep learning are emphatically studied.The basic theory and algorithm of the model is discussed,and the algorithms applied by these two models all effectively improve the limitations of the traditional methods.(2)The Long Short-Term Memory neural networks are studied: Because traditional neural networks are used in the diagnosis of motor faults,the correlation between different data is neglected,long-term dependencies cannot be learned,and gradients disappear in the process of feedback information.a fault diagnosis model is constructed by combining the Long Short-Term Memory neural networks with the Softmax multi-classifier.According to the good characteristics of the network in extracting time series features,the characteristics of fault data are extracted effectively,and the Softmax multi-classifier with strong generalization ability and robustness is classified,so as to identify three kinds of common motor faults.Through the simulation test of TensorFlow,the results show that the diagnosis method of the model is better than the traditional fault diagnosis method of motor.(3)The Stack Sparse Auto-Encoder is studied: Aiming at the problem that the traditional network is easy to fall into the local optimal,the fault characteristics are extracted according to the label data,and the amount of experimental sample data is small,while in the background of the big data of motor monitoring,the shallow network is constrained to the complex classification problem.Therefore,a feature learning method of Stack Sparse Auto-Encoder is applied,which constructs a deep neural network from Sparse Auto-Encoder combination,extracts the fault signal characteristics,and transforms the collected data into the input of the Stack Sparse Auto-Encoder after frequency domain transformation,which is combined with the Softmax multi-classifier.Finally,the effectiveness of the applied method is verified by simulation,and the accuracy of the diagnosis is improved.