Fault Diagnosis Of Wind Turbine Rolling Bearing Based On Machine Learning

As a major energy consumer in the world,China attaches great importance to the effective use of wind energy resources,however,the high failure rate of wind turbines restricts the rapid and efficient development of the wind power industry.As a key component of the wind turbine,bearings-related failures occur from time to time due to long-term operation.Therefore,the construction of a bearing fault diagnosis method with high accuracy is one of the necessary conditions to improve the reliability of wind power equipment,and it is also of far-reaching significance for the full use of energy resources and the protection of ecological environment.Based on this practical requirement,starting from the analysis of vibration signal of wind turbine bearings,this paper constructs a fault diagnosis model of rolling bearings of wind turbines from the two perspectives of direct recognition of vibration signals and indirect identification of key features,and develops corresponding diagnostic systems.Aiming at the problem of low recognition rate of bearing fault diagnosis under the background of strong noise,a deep residual shrinkage network model is proposed.In the model,the cuckoo search algorithm is introduced to optimize the parameters such as the number of layers,the weight value of each layer,the bias value,the size of the convolution kernel of each layer,and the learning rate,etc.,with the accuracy of the classification results of the test set as the objective function and the minimum error in the feedback loop as the fitness function.Experiments on CWRU bearing data set show that the accuracy of the optimized network on the test set reaches98.33%,and its parameters do not need to be manually adjusted,realizing end-to-end bearing fault diagnosis.To address the issue of inaccurate bearing fault diagnosis and recognition when data distribution is inconsistent,a rolling bearing fault diagnosis model based on wavelet packet and Transformer network is proposed from the perspective of extracting signal features first and then identifying key features.In the model,the results of wavelet packet transform were input as new input codes into the modified Transformer network for recognition.A new data augmentation method was used,combined with the K-means algorithm to achieve granular partitioning,which solved the impact of inconsistent data distribution on the diagnostic results.The bearing fault diagnosis based on fault feature recognition was achieved,and the superiority of the model was verified on the CWRU bearing dataset.In addition,a Transformer based early fault detection model for bearings is proposed.In this model,it is proposed to use the values generated from the One hot encoding output of the Transformer classification model as early fault detection indicators to detect early faults.Bearing fault detection based on fault feature recognition is implemented,and the superiority of the model is verified on the IEEE PHM Challenge bearing dataset.Finally,this paper uses MATLAB to create a graphical user interface and develops a system that can realize data display,feature extraction,traditional fault diagnosis,intelligent fault diagnosis and fault online detection,so as to guide the industrial field.