Research On Fault Diagnosis Of Fan Bearing Based On Machine Learning

With the rapid development of wind power industry in China,the installed capacity of wind power generation is larger than that of other clean energy.With the expansion of the wind turbine,the wind turbine tower as a component supporting the engine room and the wind turbine system force is gradually increasing.These factors have brought some difficulties to the daily operation and maintenance of the wind turbine,and also increased the cost of the daily operation and maintenance of the wind turbine.In the wind turbine drive system,bearing is an indispensable part with high failure rate,so the health status of bearing is related to the operation safety of the whole wind turbine.Therefore,fault diagnosis of wind turbine bearing is of great significance.(1)In the fault diagnosis methods of wind turbine spindle bearing,the vibration analysis technology is the most widely used for fault signal feature analysis.That is,when the spindle bearing fails in the process of operation,the vibration analysis is used to analyze and process the fault signal,and finally the fault location and fault category are determined.At present,time domain analysis,frequency domain analysis and time-frequency domain analysis are more commonly used analysis methods.In this paper,the joint analysis of time and frequency of bearing vibration signals under different working conditions is carried out through short-time Fourier transform,and the time-frequency image data set of bearing is constructed.(2)The fault diagnosis algorithm based on classical model is optimized.In real life,due to the interaction of multiple components and environment,the working system usually has multiple time scale features.Fault diagnosis method based on single scale features is difficult to effectively mine and utilize the multi-scale information contained in vibration data,Therefore,this paper proposes to add an upper sampling layer to the classical convolutional neural network model to build a new network model for wind turbine bearing fault diagnosis.Taking the bearing time-frequency image as the research object,the fault diagnosis model built in this paper is simulated and verified.(3)Due to the influence of working condition alternation,the fault vibration signal of bearing has strong time-varying.Therefore,the vibration signals collected in different time periods have different characteristics,resulting in the increase of the difference between different samples,which will greatly reduce the generalization ability of the network model.In this paper,attention mechanism is added to the existing convolutional neural network structure to optimize the bearing fault diagnosis model.Through the experimental analysis,using the above method for bearing fault diagnosis,the average accuracy can reach 96.6%,which verifies the effectiveness and accuracy of the scheme in bearing fault diagnosis.