Fault prediction of computer image recognition based on convolutional neural network

In order to solve the problem that traditional bearing fault diagnosis methods need a lot of professional knowledge, this paper proposes a fault prediction method of computer image recognition based on convolutional neural network.First of all, the concept V3 model is used as the pre training model, and the concept V3 model training method combining deep learning and transfer learning is designed; Then, the cross entropy is used as the loss function to evaluate the effect of model training, and the method and steps of fault diagnosis are given.The validity of the method is verified by the vibration data of bearings in normal and different fault states; Finally, the principal component analysis method is used to analyze the clustering effect of the characteristic parameters extracted by the inception V3 model on different fault modes.By comparing and analyzing the training times and training time of the inception V3 model with and without the transfer learning, the improvement effect of the transfer learning method on the training speed of the model is verified.The experimental results show that when the time domain waveform image data is used as the input of the model, the overall fault diagnosis accuracy of the model reaches 96.1%;When the spectral image data is used as the input of the model, the accuracy rate is 96.8%;When the envelope spectrum image data is used as the input of the model, the accuracy rate is 95.4%;With the same fault diagnosis accuracy, the training times and training time of inception V3 model are greatly reduced when using transfer learning.