Faulty bearing diagnostic model based on multi-dimensional signal and multi-analysis domain

Deep learning and multidimensional signal fusion are utilized to fully extract fault features and integrate them into effective signals to cope with special cases in bearing fault diagnosis. Current mainstream data fusion methods only utilize vibration signals, and the vast majority of signal analysis is limited to the time domain. In addition, in the mainstream data fusion scheme, the fusion capability of the signal collector is relatively low, and the correlation and compatibility between the data cannot be guaranteed. In order to further improve the judging ability of signal features, this paper proposes a bearing fault diagnosis model based on multi-dimensional signals and multi-analysis domain. In this model, a multi-dimensional signal data model with multiple analysis domains is used for feature extraction and fusion. And the independent networks are classified according to their functions, and a single network is used to establish a data feature fusion system, while other networks extract features from different sensors. To ensure the fusion of signal acquisition from different analysis domains, multiple fusion nodes are added between the layers of the fusion network and an attention mechanism is introduced to self-weight the different features. Through experiments, technical comparisons were conducted to improve the efficiency of feature recognition and the accuracy of defect classification, and to verify the effectiveness and feasibility of the proposed method.