Research on controllable deep learning of multi-channel image coding technology in Ferrographic Image fault classification

Utilizing computer technology to realize ferrographic analysis and intelligent fault diagnosis technology is fundamental research to ensure the normal operation of precision, complex and severe marine mechanical equipment. This work is focused on the issues that affect the feature extraction and accuracy of the computer model, such as fuzzy edge and complex surface texture of wear particle image in ferrographic image. This research creatively establishes a new multi-channel image encoder, which artificially encodes the information of the original ferrographic image. The new encoding image includes the original channel information, artificial edge enhancement information, and artificial surface enhancement information, to improve the edge and surface features of the image. The encoder can not only improve the visible edge and surface characteristics of the picture but also form the MCECNN model after connecting to the convolutional neural network. The model can successfully improve the accuracy, convergence speed, generalization, and controllability of the model, to solve the problem of online intelligent recognition in the field of ferrography to the greatest extent.