An advanced fault diagnosis approach for wind turbine planetary gearbox based on optimized multi-layer attention denoising autoencoders

Abstract Fault diagnosis of wind turbine planetary gearboxes is essential for maintaining their operational reliability; this paper introduces a novel method focused on fault diagnosis. Initially, raw vibration signals from the gearbox are transmitted to a data processing system where blind source separation and ensemble local mean decomposition are employed to extract sparse components. These sparse samples are used to optimize a deep fault diagnosis architecture based on multi-layer denoising autoencoders, which effectively extract fault features. By integrating the attention mechanism and chaotic quantum particle swarm optimization, the model enhances feature extraction, leading to improved fault classification accuracy. In two experiments based on different datasets, the diagnosis accuracy of fault types reaches 99.20% and 96.73%, respectively, while the diagnosis accuracy of corresponding fault severity is 99.12% and 93.60%. Experimental results validate the effectiveness of our method in the diagnosis of gearbox faults, demonstrating robust performance in complex operating environments.