Fault analysis of offshore wind turbine gearbox based on FEEMD-MDIE-Chi2 and its influence on diagnostic accuracy

To address the problem of weak feature representation in early fault signals of wind turbine planetary gearboxes, existing research indicates that traditional Empirical Mode Decomposition (EMD) is highly susceptible to mode mixing. Although Ensemble EMD (EEMD) mitigates mode mixing through multiple decompositions, its high computational cost limits its suitability for real-time offshore wind monitoring applications. Single-scale entropy analysis is similarly inadequate, as it fails to capture multiscale nonlinear fault characteristics and early fault features. To overcome these limitations, this study proposes a signal processing framework that integrates Fast Ensemble Empirical Mode Decomposition (FEEMD) and Multiscale Diversity Index Entropy (MDIE) with a prediction model based on a Convolutional Neural Network (CNN) and Bidirectional Long Short Term Memory network (BiLSTM). Initially, the original vibration signals from planetary gearboxes are decomposed using FEEMD. MDIE is then applied to identify variation patterns of fault signals across different scales and extract fault-related information. Subsequently, the Chi-squared Test (Chi2) is used to rank features most strongly correlated with faults and remove redundant information. Finally, both the FEEMD-MDIE processed data and unprocessed data are input into the CNN-BiLSTM model for diagnosis. This process improves diagnostic accuracy by 24%. Moreover, when Chi2-processed data are applied, an additional 3.5% accuracy improvement is achieved, meeting the desired prediction performance.