Intelligent fault detection of zero-sample rolling bearings driven by combined time-frequency analysis and multimodal knowledge

To meet the demand for intelligent monitoring of rolling bearings and overcome the limitation of scarce fault samples in the real world, this paper proposes a zero-fault sample-based condition detection method that integrates the time-frequency analysis of rolling bearings with modal knowledge, which mainly includes: 1) The HHT envelope analysis and high-frequency filtering is introduced to reduce the interference of the base-frequency information of rolling bearings and to enhance the frequency of the fault information. 2) A novel zero-fault sample-based loss function is designed by combining the strong temporal sequence of rolling bearing monitoring data with the a priori knowledge of information mutualism to realize the effective training of the data-driven model. 3) An intelligent fault detection algorithm for rolling bearings is established based on a trained data-driven model. The proposed method is validated using a constructed rolling bearing experimental platform. The validation results show that the proposed method is able to build an effective intelligent fault detection model with zero fault samples of rolling bearings, which shows better fault detection performance than other supervised and unsupervised learning-based methods. The proposed anomaly detection method based on zero-fault samples can quickly establish its state detection model for important rolling bearings applied in engineering practice, providing a new perspective for data-driven fault diagnosis methods for rolling bearings.