Fatigue Life Prediction Of Wind Turbine Tower Based On Acoustic Emission Signal

With the continuous development of economy and society,resources such as oil and gas are becoming increasingly scarce,and environmental problems caused by energy extraction remain severe.In response to energy shortages and environmental pollution,people have turned their attention to green and pollution-free clean energy sources,such as wind,solar,hydropower,and nuclear energy.Due to China’s abundant wind energy resources,the wind power industry is the most mature in the clean energy sector.As the main support structure in wind turbines,wind turbine towers are often subjected to periodic excitations from cyclic winds and blade rotors.Under these periodic excitations,small fatigue damage is easily generated inside the tower,which accumulates continuously under cyclic excitations,leading to the appearance of fatigue cracks,and eventually causing tower collapse.Therefore,monitoring the fatigue state of the tower is essential for the daily maintenance and upkeep of wind turbines.As a mature non-destructive testing technique,acoustic emission(AE)technology can be used to monitor the fatigue state of wind turbines,determine whether equipment needs repair through monitoring signals,and improve maintenance efficiency and reduce maintenance costs.In this study,AE signals were collected from Q355E steel,a commonly used material in wind turbine towers,from the beginning of fatigue to the complete fracture of the specimen,under laboratory conditions.The collected signals were subjected to feature extraction and fusion,and a prediction model was established based on the long short term memory neural network(LSTM)model to complete the life prediction study.The main work and results are as follows:(1)During the collection process,background noise caused by mechanical vibration or other reasons is inevitably collected.Therefore,before the formal experiment,the collected signals should be subjected to denoising.The threshold selection in traditional wavelet packet denoising algorithms is usually based on experience,leading to the phenomenon that some noise is not removed,while useful information is eliminated.Therefore,a particle swarm optimization algorithm was proposed to optimize the threshold in the wavelet denoising algorithm to enable the algorithm to adaptively set threshold parameters based on the signal and improve the denoising effect.Through simulation experiments,the results after denoising with the optimization algorithm have a higher signal noise ratio(SNR)and a lower root mean square error(RMSE)than other threshold denoising methods,with the best denoising effect.(2)In order to avoid using a single feature that cannot fully characterize material degradation performance,feature extraction was performed on the pre-processed signals in this study.The commonly used time-domain features and frequency-domain features were selected as extraction parameters.The correlation coefficient method was applied to select features,and principal component analysis was chosen to fuse the selected features to obtain a single degradation index.The first principal component after fusion was selected as the degradation curve that characterizes the material’s degradation performance,which laid the groundwork for the life prediction study.(3)The research is based on the Long Short Term Memory Neural Network(LSTM)model,which excels in processing time series,to establish a prediction model.In order to improve the prediction accuracy of the prediction model,the particle swarm optimization algorithm was selected to optimize the outage parameters in the model and the particle swarm optimization-long Short term memory neural network(PSO-LSTM)model was established as the final prediction model to complete the lifetime prediction research.In order to demonstrate the adaptability and effectiveness of the combined model,the combined model is compared with the single LSTM model and multiple regression model.The results indicate that compared with LSTM model and multiple regression model,the combined model has higher prediction accuracy.