Intelligent fault diagnosis for a wind turbine gearbox via multisensor information fusion with adaptive weights

Wind turbines generally operate under variable rotational speeds with strong environmental noise, which poses a great challenge to the generalizability of intelligent fault diagnosis for wind turbine gearboxes. In this paper, a novel fault diagnosis method that combines multisensor information fusion and adaptive weighting is introduced. More specifically, vibration signals collected from orthogonal positions are first transformed into time-frequency feature matrices via a synchro-squeezing S transform. Then, the matrices are spliced along the channel and mapped into a fused time-frequency feature image. Finally, a novel model based on a deep residual network and an attention mechanism is developed to identify mechanical faults. The proposed approach leverages a novel fusion framework that uses bidirectional multiresolution time-frequency analysis for multisensor measurements, followed by discriminant feature extraction through attention-guided deep learning techniques. The key innovation of this method lies in its intelligent weighting mechanism, which dynamically adjusts the sensor contributions while maintaining computational efficiency. Evaluation tests are conducted on a full-scale electrically closed wind turbine gearbox test rig operated at varying speeds. This conclusively demonstrates that the methodology has more stable convergence, higher accuracy, and greater generalizability than existing diagnostic models under challenging operational conditions.