A dynamically balanced wavelet coefficient matching transient energy operator for state identification of rotating machinery

Currently, the state recognition of rotating machinery mostly relies on vibration signals as data sources. However, the actual operating environment of the equipment has an impact on the readings of the sensors, therefore, the diagnostic results are greatly affected by noise and interference. Until now, effective measures against noise and interference have not been found. We are looking for a new paradigm of neural network encoding traditional signal processing methods to attempt to solve diagnostic problems in noisy environments. We propose a method of dynamically balancing wavelet coefficients to match transient energy operators to enhance noise resistance. The first step is to design a self-learning wavelet threshold denoising mode for multi-step signal encoding and reconstruction to remove interference components. The second step is to embed the Teager energy operator into the model to enhance high-frequency components such as transient shocks and pulse excitations. The third step is to construct a joint attention fusion function of scale and channel dimensions to select discriminative elements. We validated the effectiveness of the proposed method using different rotating mechanical equipment in operating environments with varying levels of noise intensity, and the results showed that the model has strong noise robustness.