Research And Implementation Of Permanent Magnet Synchronous Motor Operating State Recognition Based On Sound Signal

Compared with traditional motors,Permanent Magnet Synchronous Motor(PMSM)has the characteristics of high efficiency,good stability and high power density,and is widely used in various fields.However,the particularity of PMSM working environment and working condition makes its running state easy to produce abnormal,so in the industrial control environment,real-time diagnosis and recognition of PMSM running state has important significance and value.At present,methods to identify the running state of PMSM,such as mathematical model-based processing,signal analysis processing and general artificial intelligence processing,mainly have the following problems: the required signal data acquisition operation is complicated and the cost is high;Lack of signal characterization ability;The size of the model is large but the identification accuracy is not high.This thesis proposes an improved Alex Net neural network model to identify sound signal features.The main research work and achievements are as follows:(1)Preprocess the collected PMSM original sound data and extract sound signal features such as FBank;(2)The PMSM operating state recognition model based on improved Alex Net network is studied: Considering the limited storage resources and computing resources of intelligent terminals for PMSM running state identification,Using packet Grouped Convolution and Depthwise Separable Convolution algorithm for Alex Net basic network structure was improved in order to reduce the quantity and the amount of calculation.In order to ensure the recognition accuracy of the model,Spatial Groupwise Enhance(SGE)and Convolutional Block Attention Module algorithms are used to further optimize the network to improve its attention.(3)Design and implement a PMSM running state identification tool.The experimental results show that the size of the PMSM operating state recognition model based on the improved SGE＿DSC＿Alex Net neural network model is only 534 kb,the average recognition accuracy in the test set reaches 98.30%,and the diagnostic accuracy of some operating states reaches more than 99%.The results show that the method proposed in this paper is an efficient and accurate method for PMSM running state recognition.