Wind Turbine Drive Train Bearing Fault Diagnosis

Overcapacity of thermal power,our country is gradually reducing the proportion of thermal power in the energy structure,although in the future,thermal power will still be the dominant form of power generation in our country,but there is no doubt that the advent of wind power is coming,which can be perceived from a series of national policies.China is vast in territory,a large territory could meet the conditions required for wind power,basically every province has a wind farm,especially as in Neimeng,Xinjiang,Gansu and other provinces because of its excellent mining conditions.With the number of wind turbine increasing,the wind turbine failures will be more and more common,so to improve the monitoring and diagnosis of the wind turbine level is very significant.Bearings is an essential part of rotating machinery,and there are many bearings in wind turbines,and bearings often fail,so strengthing the fan bearing fault diagnosis is very necessary.In this paper,we chose deep groove ball bearings of wind turbine drive chain as research target,we put ewt and multi-feature extraction and pso-svm and pso-bp to together.Firstly calculate the characteristic frequency of the three faults,decompose the signal by ewt,compare and analyze the mode,get the mode's envelope spectrum,seek the mode that contain fault characteristics.Reducing noise of bearing normal vibration signal by wt.Finally extract fault features and build up pso-svm and pso-bp model,then training model and test model.Training model of bp and pso-bp with the neural network.Test the model with test samples,fault diagnosis correct recognition rate of pso-bp is 90%that better than bp of 75%.To further test the fault diagnosis algorithm,we train pso-bp model under 12000hz and 24000hz and 48000 hz.The results show that the fault diagnosis algorithm is successful.Training model of svm and pso-svm,fault diagnosis correct recognition rate of pso-svm is 92.5%that better than svm of 82.5%,this shows that the PSO in this paper is successful in optimizing SVM,which improves the accuracy of the diagnostic model.In order to further test the fault diagnosis algorithm,we train pso-svm model under 12000hz and 24000hz and 48000 hz.The results show that higher the sampling frequency,the higher the precision of the fault diagnosis model trained,and pso-svm is better than pso-bp.