Health Status Assessment Research Of Wind Power Rolling Bearings Based On Signal Processing And Convolutional Neural Network

In the context of global energy conservation and emission reduction,wind power has become one of the important options to replace fossil energy.As the third major energy source in China,wind power has also become one of the most important renewable energy sources in the world,with its installed capacity increasing year by year.Affected by the random fluctuations of wind speed and wind direction,wind turbines need to experience frequent startstop and speed changes,which pose a severe test to various mechanical components of the wind power transmission chain.The failure shutdown caused by the transmission system is also the main reason of fan shutdown.As an important part of the wind power transmission chain,accurate health status assessment of rolling bearings is of great significance to ensure the efficiency and safety of wind power generation.In recent years,with the rapid development of big data analysis technology and artificial intelligence technology,as well as a large amount of data brought by the progress of sensor monitoring technology and data storage technology,the realization of intelligent rolling bearing fault diagnosis and degradation trend prediction has become a hotspot of current research.Based on this,this paper conducts a research on wind turbine rolling bearing health status assessment based on time series prediction.The main contents are as follows:(1)The mechanical components and signal acquisition system of the wind turbine transmission system simulation test bench are introduced,and the common time domain,frequency domain and time-frequency domain characteristics of rolling bearing signal processing are given.(2)Aiming at the problem of the limited receptive field of the traditional convolutional neural network convolution kernel,a convolutional neural network fault diagnosis model combining discrete Fourier transform(DFT)and efficient channel attention(ECA)is proposed.Firstly,the original vibration signal is converted from the time domain signal to the frequency domain signal by DFT,and then the frequency domain signal is convolved,and then,the convolved data is transformed into the time domain by inverse discrete Fourier transform(IDFT),and convolved with the original signal multidimensional data in the channel dimension,,and then the multi-channel features are assigned weights through an ECA mechanism,and the features are further learned through multiple convolution-pooling pairs,finally,the connection layer enables end-to-end fault diagnosis of rolling bearings.The experimental results show that the model in this paper not only achieves high accuracy on multiple datasets,but also has strong robustness and anti-noise performance.(3)Aiming at the problem of insufficient feature selection for the construction of degradation trend indicators at this stage,this paper obtains a multi-view feature set that can fully characterize the degradation state of rolling bearings through feature extraction and feature selection,and constructs a trend index that can present the degradation state of rolling bearings through feature fusion.The specific implementation is as follows.Firstly,by extracting the 797-dimensional features of the rolling bearing vibration signal in the time domain,frequency domain and time-frequency domain,a multi-view feature set of the rolling bearing is constructed,then the features are selected by the comprehensive health evaluation index and the Light GBM model.Finally,the dimensionality reduction and fusion of the filtered features are performed through Isometric Mapping(ISOMAP)to construct the bearing degradation index.The experimental results show that the method has higher comprehensive health evaluation index and better generalization performance.(4)In order to improve the feature extraction ability and prediction accuracy of the rolling bearing degradation trend prediction model,a neural network model was proposed,which embedded Dynamic Convolution(DC)into Temporal Convolution Network(TCN).Firstly,the traditional convolution method is replaced with dynamic convolution,and then two consecutive layers of convolution and 1×1 convolution are connected in parallel to form an improved residual block,and finally a prediction model is formed by stacking multiple improved residual blocks.The experimental results show that TCNDC has high accuracy in predicting the degradation trend index of rolling bearings,and it shows better generalization on multi-bearing data.