Fault Diagnosis Method of Differential Pressure Measurement System Based on Wavelet Energy Entropy and LightGBM-AM

The key differential pressure measurement system of the comprehensive cigarette test bench will cause reference air pressure imbalance and drift due to aging of the sealing loop, leakage of constant current devices and other abnormalities, affecting the accuracy of the measurement results. This study proposes a differential pressure measurement system fault diagnosis method that combines wavelet energy-information entropy characteristics and improved LightGBM algorithm. First, DB4 multi-scale wavelet basis decomposition and low-frequency signal reconstruction were performed on the collected differential pressure measurement signals of various types of fault tests. The energy distribution of each frequency band of the initial signal was calculated and the information entropy value of the reconstructed low-frequency signal measurement plateau was constructed to construct a joint feature set of fault samples. Furthermore, an attention mechanism is introduced based on the LightGBM algorithm to dynamically adjust the learning feature weights during the decision tree construction process to strengthen the impact of key features on decision-making. Comparative verification with the indicators of 8 classic classification algorithms such as SVM, KNN, and GBDT shows that under working condition measurement conditions, the classification accuracy of the improved LightGBM method is significantly improved compared with traditional LightGBM, SVM and other methods. The overall accuracy of various types of fault detection is 97.50%, which provides a highly interpretable and robust solution for real-time condition monitoring under complex working conditions.