Predicting wind turbine gearbox vibration amplitude by integrating convolutional neural networks and stochastic differential equations
Accurate prediction of wind turbine gearbox vibration amplitude is critical for fault early warning, as amplitude fluctuations encode degradation information. However, such predictions are challenging due to strong nonlinearity, high noise, and non-stationarity. This study proposes a hybrid framework integrating Convolutional Neural Networks (CNN) with Stochastic Differential Equations (SDE), where the CNN adaptively learns the time-varying drift and diffusion coefficients of the SDE, combining deep feature extraction with stochastic modeling. Validated using 6000 hours of field data from a 2 MW wind turbine gearbox, the CNN-SDE model achieves superior performance (RMSE = 0.0045, MAE = 0.0031, R 2 = 0.9944), outperforming ODE, SDE, standalone CNN, and LSTM baselines. The results confirm that the proposed model captures both deterministic trends and stochastic fluctuations, offering a robust approach for gearbox condition monitoring and preventive maintenance.
Authors
- Song Zhaocheng
- Hongsheng Su
- Zihan Zhou (ORCID: https://orcid.org/0009-0009-4602-516X)
- Zhensheng Teng
Institutions
- Lanzhou Jiaotong University (CN)
Publication Details
- Journal
- Wind Engineering
- Published
- 2026-11-16
- DOI
- https://doi.org/10.1177/0309524x261509736
- Primary Topic
- Machine Fault Diagnosis Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00