Intelligent Machine Fault Diagnosis Using Convolutional Neural Networks and Transfer Learning

With the development of automated and integrated large-scale industrial systems, accurate effective fault diagnosis methods are required to ensure security reliability running mechanical equipment. Due time consumption poor generalization performance conventional machine learning-based methods, deep learning (DL)-based have wider application prospects due their end-to-end architectural properties. However, in DL models, problems such as a large number trainable parameters, complicated hyperparameter tuning, initialization instability increase difficulty model training limit higher performance. To address these disadvantages method, we proposed novel framework by applying convolutional neural networks (CNNs) based on optimization transfer (TL). TL can help achieve precision with less computational cost transferring low-level features fine-tuning high-level layers. In addition, data processing was implemented using continuous wavelet transformation (CWT) convert vibration signals into 2-D images, support vector machines (SVM) were employed replace fully connected layers for better classification. As result, method superior classical architecture trained from scratch. The is analyzed presenting testing reports, convergence curves, confusion matrixes. Moreover, experiments comprised cross-domain diagnosis, simulated composite detection, comparison seven datasets, including bearings, gearboxes, rotors, presented. Based results, it be observed that our achieved highest accuracy under various conditions.