| With the continuous development of society,the problem of atmospheric environmental pollution is increasing day by day.Wind energy is one of the cleanest renewable energy sources,which plays an important role in improving the energy structure and is paid more attention to in all countries of the world.However,with the rapid development of installed capacity,the operation and maintenance of wind turbine is becoming more and more prominent.Because of the long-term work in the bad environment,the operation performance and health state of the components of the wind turbine show a declining trend with the change of the operating environment and time,and even produce faults,resulting in serious loss of electricity generation and economic loss.In wind turbine,the downtime caused by gearbox bearing fault is the longest,and rolling bearing is an important component.It has important theoretical research value and engineering application significance to study its fault diagnosis and prediction.In view of the fault diagnosis of wind power rolling bearing,the commonly used method is to combine manual feature extraction with classifier to realize fault recognition.The model is usually complex,and manual feature extraction is easy to lose key information.It can not guarantee generality and generalization ability.For this reason,this paper takes the rolling bearing of wind turbine as the research object,considering the problems of weak fault feature,difficult extraction and low diagnosis efficiency of rolling bearing of wind turbine,and puts forward a fault diagnosis algorithm based on improved convolution neural network(convolutional neural network,CNN)to realize automatic fault feature extraction and fault classification.During this fault diagnosis strategy,the structure and training algorithm of traditional CNN are improved.Experiments show that the model has advantages in automatic feature extraction,training speed,diagnostic accuracy,generalization ability and robustness.Considering the effect of CNN network parameters on fault diagnosis and the inability to optimally generalize all datasets,the network depth,initial learning rate,regularization coefficient,and SGDM momentum of this improved CNN model are selected using bayesian optimizer.Through experiments,the iterative process of Bayesian optimization is recorded in numbers and images,and the results of CNN parameter optimization are analyzed to verify the generalization and robustness of the optimized CNN model.The quick diagnosis of the fault of the wind power rolling bearing can save the troubleshooting time.However,predictive diagnosis of its running state can identify the symptoms of failure in time,determine the time of failure as soon as possible,and carry out predictive maintenance of the equipment.considering the superiority of LSTM in dealing with sequential sequences,a fault prediction strategy based on double-layer LSTM is proposed.Among the fault prediction strategies,two LSTM are used for short-term timing prediction and sequence-to-label classification,so as to predict the running state of each time step of the sequence in the future.The effectiveness of this model in fault prediction is verified by experiments.The research on the remaining life prediction of wind power rolling bearing can further optimize the operation and maintenance strategy and spare parts ordering,carry out maintenance according to the situation,and improve the reliability of wind turbine operation.Taking into account the problem that LSTM takes a long time to deal with longer sequence sequence,a prediction model of residual life of wind power rolling bearing based on CNN-LSTM model is proposed.Make full use of CNN fault feature extraction ability and LSTM advantages in processing sequence sequence sequence.Experiments show that this CNN-LSTM model can accurately describe the trend of residual life change and has high effectiveness in realizing residual life prediction. |
