Dynamic Weight-Based Adaptive Data Fusion Filtering for RealTime Condition Monitoring of Electromechanical Equipment

Abstract

In electromechanical equipment condition monitoring, data accuracy and timeliness are critical for ensuring reliable operation. This paper introduces a dynamic weight-based adaptive data fusion filtering algorithm that integrates an Extended Kalman Filter (EKF) variant with wavelet denoising and real-time weight adjustment. The algorithm employs variance-based stability metrics and data update frequency for reliability assessment, dynamically allocating fusion weights via Equations (1)– (7). Experiments use a test platform with PCB 356A16 accelerometers, K-type thermocouples, and LEM LA55-P current sensors to simulate gear wear and motor rotor imbalance faults. Compared to baseline methods (Kalman Filter, Particle Filter), the proposed algorithm reduces RMSE by 35.2% and 28.7%, and MAE by 41.5% and 34.3%, respectively. Fault feature extraction shows a 37% increase in vibration peak index for mild gear wear and a 50% increase in current harmonic content for rotor imbalance. While runtime is slightly higher than Kalman Filter, it remains within practical limits, demonstrating superior real-time performance for fault early warning in complex industrial environments.