Fault Early Warning and Remaining Useful Life Prediction for Wind Turbine Main Shaft Systems Using IPSO-XGBoost and MSCNN-GRU-SA

Abstract In order to satisfy the demand of intelligent operation and maintenance for the main shaft system of a wind turbine, the present work introduces the limitations and problems of traditional methods, whose sensitivity to early defects is low and whose performance degradation prediction is difficult. This study raises a fault early warning and Remaining Useful Life prediction model that integrates intelligent optimization and deep learning. In the fault early warning stage, an improved Particle Swarm Optimization algorithm dynamically optimizes the hyperparameters of the eXtreme Gradient Boosting. In the Remaining Useful Life prediction stage, a Multi-Scale Convolutional Neural Network extracts temporal features from degradation signals, and a Self-Attention highlights critical degradation stages. Experimental data indicate that the elevated model helps to decrease the Root Mean Square Error and Mean Absolute Error of the Remaining Useful Life prediction to 8.7 hours and 6.3 hours, respectively. Correlation coefficients between the predicted sequences and the true sequences are 0.978 and 0.971. It is clear that the model continues to keep the highest prediction accuracy at the early life period, mid-life period, and late life period. The model is of great significance to the timeliness of early warning of fault, the accuracy of prediction, as well as the stability in complicated conditions.