Research On CNN Recognition Of Motor Operation Statu

As an important mechanical equipment,motors are widely used in various fields.It is an inevitable factor that temperature rise while the motor is running.The temperature is one of the important parameters to indicate the motor’s running state.It is an essential measure to avoid the motor’s high temperature by monitoring the temperature of the motor.Therefore,this dissertation carries out an in-depth study of the running state of the motor.The motor’s running state is considered in this paper.The running thermal image data of motor is collected with the thermal imager.For improving the recognition accuracy of motor running status,the wavelet transform is used to enhance the images.For increasing the speed of recognition,the image segmentation is adopted.The work of this article is divided into several parts:First of all,the reasonable experimental equipment is selected to build an experimental platform for motor’s running state recognition in this dissertation.About 10,000 data images of motor’s running state are collected by the thermal imager.The acquired images are divided and marked them into three categories,namely normal operating state,stop state and excessive temperature state of motor’s running.In the experiment,randomly selects 20% of the collected samples as the test set and the remaining sample data as the training set.Then,to solve the problem that traditional machine learning will be affected by manual extraction of feature vectors.The convolutional neural network(CNN)method for motor’s running state recognition is used to reduce the loss in feature extraction and greatly reduce the manpower.The image pre-processing method is used to solve the problems of blankness and mode conversion due to image acquisition.And the processed data images are input to the CNN model.In order to make the features extracted by the model more significant,wavelet transform(WT)is used to enhance the data images.Then the features are extracted by the convolution layer,and the average pooling reduces the computation.After that,the output is turned into a one-dimensional input to the fully connected layer,and the Dropout function is added to reduce the generation of model overfitting.The extracted features are input to the Softmax classifier to identify the motor’s running status.Finally,in order to solve the problem that the accuracy of motor’s running state recognition and the computation time of the model.The CNN model is improved in this dissertation.And the dimension of the convolution kernel is changing in the CNN model.In order to make the features enhanced,the convolution kernel dimension is increased to strengthen the feature values after convolution and improve the recognition effect.It is verified through experiments that the right number of convolutional kernel dimensions are chose to reduce the model training time and improve the model recognition accuracy.In addition,to further improve the effectiveness of motor identification,the convolutional kernel matrix type is adjusted.The validity of the random upper triangular matrix for motor operation status identification is verified by experimental comparison.To solve the problems of excessive data volume and computational pressure of model operations.The image segmentation is used to divides large-size images into independent smaller images by MTATLAB,which greatly reduces the computational effort of the model and improves the recognition speed.