Research On Intelligent Fault Diagnosis Of Motor Bearings Based On Stack Autoencoder

As an important part of modern industry,electrical equipment plays a key role in the national pillar industry.The long-term operation of the motor systems is easy to cause faults,which may lead to major safety accidents and huge economic losses.According to IEEE-IAS,the rate of bearings fault is the highest.The state of bearings directly affects the operating efficiency and output capacity of the motors.Therefore,it is of great theoretical significance and engineering application value to carry out relevant research on bearing fault diagnosis and discover potential safety hazards timely and accurately,so as to ensure the safe operation of electrical equipment,avoid major accidents,reasonably arrange maintenance time and reduce unplanned downtime.In the model-based fault diagnosis method,it is necessary to establish an accurate fault mechanism model.However,the motor is a complex system involving mechanical-electrical-magnetic.Accurate modeling has extremely high requirements for operators,and it is difficult to obtain the detailed structural size information required in the modeling process.With the development of modern signal processing technology and artificial intelligence technology,intelligent motor fault diagnosis methods have emerged to avoid complex fault mechanism modeling process,improve diagnosis efficiency,and promote the intelligent development of diagnosis work.To improve the diagnosis accuracy,meet the complex diagnosis requirements and expand the application,this paper focuses on the research of intelligent fault diagnosis methods for motor bearings.The research results of this paper provide a theoretical and technical basis for intelligent fault diagnosis of motor bearing,and are of great significance to improve the intelligent level of motor system.The research content of this paper is as follow:(1)To solve the problem that traditional intelligent fault diagnosis relies on manual experience and has low accuracy in complex diagnosis tasks,this paper carried out the research on the intelligent fault diagnosis method of motor bearing based on stacked autoencoder(SAE).The vibration signal has a lot of redu ndant information,which is easy to cause the algorithm to overfit.Therefore,a sparse denoising SAE algorithm(SDSAE)is proposed to improve the robustness and generalization ability of the algorithm.The improved algorithm leads to the increase of network super parameters,so quantum particle swarm optimization algorithm is used to realize the adaptive optimization design of the network.Experiments verify the feasibility of the proposed adaptive SDSAE in dealing with complex diagnosis tasks,which can realize high-accuracy diagnosis of fault type and fault degree at the same time.(2)To solve the restriction of unbalanced training data on diagnosis accuracy,a data expansion method based on Generative adversarial network(GAN)is proposed in this paper.This paper studies the feasibility of GAN in dealing with the problem of data imbalance.Aiming at the limitations of training difficulty and unknown category,a conditional Wasserstein GAN is proposed(CWGAN).Under the constraint of data labels,the training loss function is modified to optimize the training process,avoid the algorithm falling into gradient disappearance and improve the quality of generated data.The feasibility and effectiveness of bearing fault diagnosis based on CWGAN-SDSAE in dealing with data imbalance are verified by experiments.(3)To solve the problem of limited diagnosis accuracy under the difference of data distribution,an intelligent fault diagnosis method based on deep transfer learning is proposed.This paper deeply analyzes the deep transfer diagnosis framework and network training mode to solve the diagnosis application problem of motor under variable working conditions.To further improve the adaptability of differences in data distribution,an improved diagnosis method based on class separation and domain fusion is proposed,which can extend the application to cross motor variable condition scenarios.The feasibility and effectiveness of bearing fault diagnosis based on SDSAE-CSDF in dealing with expanding its application range are verified by three experimental platforms.(4)To solve the problem of high cost and limited installation of vibration sensors,the bearing fault diagnosis method based on current signal is studied.Considering the weak representation of bearing fault in current,a diagnosis method based on noise elimination and decision fusion is proposed to improve the accuracy of bearing fault diagnosis using current signal.The noise elimination method based on encoder can effectively improve the current signal-to-noise ratio.The decision fusion is carried out in a simple way to make full use of the fault information in the three-phase stator current and improve the diagnosis accuracy.The results of two experimental platforms show that this method can be used not only for bearing fault diagnosis,but also for eccentric fault diagnosis.