Research On Faults Fusion Diagnosis Method Of Wind Turbines Bearing Based On Deep Learning

Wind power generation has been growing fast as a new energy source.Rolling bearings are essential components of wind turbines and they have a crucial role in ensuring the wind turbines operate efficiently.As wind turbines work in complex environments such as corrosion,sand,dust,humidity and heat,the rolling bearings being prone to damage,which has a serious impact on the running of the wind turbines.Therefore,it is very important to study how to diagnose the faults of rolling bearings in wind turbines under variable working conditions.The conventional method of using time-frequency for diagnosing bearing faults is not accurate enough for the needs of wind turbines operation and maintenance.Deep learning has become a popular technique for researching bearing faults diagnosis methods with the advancement of industrial automation.This paper proposes a deep learning-based fusion diagnosis method for wind turbines bearing faults and the main research contents are as follows:(1)To address the issues that large noise in wind turbines bearing faults data,the fault features are difficult to extract and the VMD decomposition parameters need to be artificially preset,this paper proposes a DBO-VMD-based fault feature extraction algorithm for wind turbines bearing.The algorithm uses dung beetle optimization algorithm(DBO)to optimize the parameters of VMD[K,α],and the fitness function of the DBO algorithm is chosen as the average envelope entropy of the IMF components.The kurtosis is used to select the appropriate IMF component for bearing fault signal reconstruction.The experimental results of CWRU and SQV datasets show that DBO-VMD can effectively extract fault features under complex and variable working conditions.(2)To address the problem of lack of wind turbines bearing fault datas samples in actual industrial operation,a TimeGAN-RepVGG-based wind turbines bearing faults diagnosis method is proposed.This method perform DBO-VMD decomposition on the original wind turbines bearing faults data firstly,then use TimeGAN to augment fault data based on the reconstructed signal,and finally,the wind turbines bearing faults diagnosis is realized based on the RepVGG.The experimental results show that when samples are unbalanced,the AP of the CWRU dataset is 99.1%,and the AP of the variable working conditions SQV dataset is 95.1%.(3)To address the problem of time domain information of fault data is ignored when training deep learning models using time-frequency maps,this paper proposes a dual-channel fusion model based on BiLSTM-RepVGG-ECA for wind turbines bearing faults diagnosis method under variable working conditions,which uses BiLSTM to extract the time domain features of fault data.And ensure the continuity of time domain data features while using an lightweight model RepVGG-ECA to extract the spatial information of the fault data.The experimental results show that the precision of BiLSTM-RepVGG-ECA fault diagnosis is 99.59%under the complex variable working conditions SQV dataset.(4)For the BiLSTM-RepVGG-ECA fusion model proposed in this paper,a wind turbines bearing faults diagnosis software was developed by PyQt5.The main functions of the software include the implementation of the DBO-VMD algorithm,the time-frequency analysis of the signal,and the training and testing of the BiLSTM-RepVGG-ECA model.The test results of the software show that it meets the needs of wind turbines bearing operation and maintenance and has good feasibility.