Intelligent fault diagnosis of rolling bearing under unbalanced samples based on simulation data fusion

Abstract With the rapid development of intelligent manufacturing, data-driven deep-learning techniques have been widely used in bearing fault diagnosis. However, the problem of unbalanced data samples usually occurs in actual production environments due to the difficulty of collecting comprehensive fault data covering multiple fault types and degrees, which directly affects the diagnosis performance. For this reason, this paper proposes a new method for simulation data-driven bearing fault diagnosis. In this paper, based on the vibration mechanism of rolling bearings, a fault signal simulation model that can accurately simulate different damage degrees of inner and outer rings is constructed. The model cannot only effectively extend the data set but also generate simulated signals that are highly consistent with accurate fault signals in terms of amplitude modulation (AM) characteristics in the absence of actual samples. This paper conducts experiments on the CWRU rolling bearing fault dataset by combining the generated simulation data with deep learning methods. The experimental results show that the model's classification accuracy reaches 98.7% and 93.7% in the case of a small number of samples (small sample scenario) and no actual samples (no sample scenario), respectively. In addition, we conducted experiments with multiple working conditions on a testbed built in the laboratory, and all of them also achieved excellent results.