Research On Electromagnetic Field Prediction Of Electrical Equipment Based On Deep Learning

Since the 21 st century,the development of information technology and artificial intelligence has reached a new height.The intelligent manufacturing of electrical equipment has become the main development goal of industrial manufacturing in my country.The core part of the intelligent manufacturing of electrical equipments is to use data to establish mathematical models for real-time analysis,but for electrical equipments with large volume and complex structures,the traditional finite element analysis method involves large-scale numerical calculations,resulting in slow calculation speed,poor accuracy and cannot meet the real-time requirements.In this thesis,the deep learning theory is applied to the analysis of the electromagnetic field of the motor.The deep learning model is built and trained with the electromagnetic field distribution datas corresponding to different motor structures.By predicting the distribution of the electromagnetic field,it can replace the traditional finite element calculation,improve the research efficiency,reduce the time cost,and meet the real-time requirements of intelligent manufacturing.Firstly,this thesis takes the three-phase asynchronous motor as the main research object,inputs the structure sizes of the motor in ANSYS Maxwell,establishs the motor model,and uses the parameterized scanning function of the software,the physical structures such as the slot width,slot height,slot width and slot radius of the three-phase asynchronous motor stator are used as design variables,and the magnetic density of the stator teeth and yoke are used as the output variables to determine the reasonable variation range of the physical structures.The process of two-dimensional model modeling and finite element analysis of three-phase asynchronous motor is written into APDL command flows,and the electromagnetic field distribution of the motor under different structural parameters is obtained.This thesis takes each group of structural parameters and their corresponding electromagnetic field distributions as a sample.The parametric modeling calculation obtains a large number of samples and performs data processing on them to provide samples for the training of subsequent deep learning models.Secondly,the hyperparameters required to build deep learning models are determined through in-depth research on the calculation process and network structure of convolutional neural networks,long and short-term memory networks and their variants.Then,Python is used to build three deep learning models on Tensorflow,a special platform for deep learning,and the models are trained with samples.The optimal model is selected by comparing the models with the evaluation indexes,and three different test functions are introduced to test the performance of the optimal model.Finally,the optimal model is used to predict the electromagnetic field distribution,and the predicted datas are visualized into an image through the simulation software,and then compared it with the image obtained by the actual commercial software simulation.By comparing the magnetic potential datas at different positions in different structural schemes,it can be seen that the datas predicted by the neural network prediction model have high accuracy.