Research On Fault Diagnosis Technology Of DC Servo Motor Based On Deep Learning Theory

Featuring its superior load capacity,reliable operation and high efficiency,DC servo motor is widely used in industrial and agricultural production and daily life.As one of the main power sources of modern industry,if the motor fails,it can lead to the paralysis of mechanical systems and even threaten lives."Made in China 2025" points out that China is at a critical moment to achieve the goal of a manufacturing power.We have to make the intelligent troubleshooting of aerospace equipment,power equipment and agricultural machinery equipment come true.Therefore,it is of great significance to achieve the intelligent state monitoring and fault diagnosis of DC servo motor.Research shows that mechanical vibration signals contain abundant time-frequency characteristic information.Therefore,the diagnosis of rotating mechanical fault based on vibration signal has always been a hot topic of expert research at home and abroad.For this purpose,the four vibration signal data of normal state,rolling body fault,outer ring fault and inner ring fault are collected by using vibration signal acquisition system in this paper.In view of the diagnosis of DC servo motor fault,the main research contents of this paper are as follows:(1)Aiming at the problem of endpoint effect and modal ambiguous stacking in the processing of continuous nonlinear motor fault signals by EMD,EEMD and other methods,in this paper,the principle of VMD algorithm is analyzed and the corresponding simulation experiment is carried out.VMD decomposition is better than traditional decomposition method,but there is a problem that parameter K is difficult to pre-set.In view of the problem of K-value selection,this paper proposes a method of optimizing VMD based on instantaneous frequency average,using the instantaneous frequency average curve to determine the optimal modal number K of VMD,and comparing and analyzing with the traditional VMD algorithm,the proposed algorithm simplifies the process of determining K-values while overcoming modal ambiguous and endpoint effects.(2)In view of the problem that on account of the large amount of data the classification model has too long running time,research found that the SVD algorithm has unique advantages in data degradation,so it is introduced into the field of DC servo motor fault diagnosis.Compared with the traditional dimensional reduction method,the matrix after SVD de-dimensionality can express the time-frequency characteristics of the signal better.(3)For the slow convergence speed caused by the deepening number of network layers in the traditional convolutional neural network,this paper combines residual learning with remnant neural network,and proposes a remnant neural network(RWDCNN)model based on residual learning for fault classification.Compared with traditional converse neural networks,R-WDCNN algorithm achieves higher recognition accuracy and faster convergence speed.At the same time,the t-SNE algorithm is used to visualize the R-WDCNN model.(4)To verify the performance of the R-WDCNN model under different training data volumes,this paper selects the support vector machine learning model based on particle group algorithm,which has a fast convergence speed and realizes the selection of the optimal parameters of support vector machine.The experimental results show that the RWDCNN model achieves high recognition accuracy under different training sets of data volume.