Cross-domain learning in rotating machinery fault diagnosis under various operating conditions based on parameter transfer

Machine learning approaches work well with large labeled data sets. In the field of fault diagnosis, the need to analyze large amounts of data provides a foundation for machine learning to be applied. However, due to the changes of rotation speed, load, and other factors, data sets will be of different data distributions. Therefore, under conditions of different features and feature distributions, it is of great importance to improve the generalization ability of the machine learning model. In this paper, a deep convolutional neural network and the principle of parameter transfer are used to extract features and transfer parameters of rolling bearing data samples under different working conditions. This paper proposes to train the machine learning model on relevant data sets with different feature distributions, and improve the learning effect of the model under other conditions by means of parameter transfer. Different working conditions are simulated using the data collected from different experimental tables, and the transfer effects between different data collected from different experimental tables are discussed respectively for different damage degrees and rotating speeds. Experimental results show that, compared with direct verification, this method can effectively improve the performance of models under different working conditions.