Research On Fault Diagnosis And Remaining Life Prediction Technology For Wind Turbine Based On Machine Learning

Wind power industry has developed rapidly in recent years.As the basic output unit of the wind power industry,the economic benefits of wind farms depend on the downtime of wind turbines.The downtime of wind turbines is largely limited by fault location speed and availability of spare parts,which heavily depend on fault diagnosis technology and component residual life prediction technology of wind turbines,while wind farm is relatively weak in these two key technical links.Statistical analysis of wind power field operation shows that wind turbine failures and shutdowns are mainly caused by mechanical component failures of transmission system.Existing fault diagnosis technology and component residual life prediction technology require a large number of high-tech diagnostic analysts to manually perform fault diagnosis,which can not effectively support the healthy development of wind power industry.This paper conducts a detailed and in-depth analysis of the low efficiency of fault diagnosis in wind farms,focusing on fault diagnosis of the wind turbines transmission system and the residual life prediction technology of the bearings,proposes and realizes three methods for batching and automatic fault diagnosis of wind turbines.This method of predicting the residual life of the bearings has been studied,and the related work is as follows:The method of signal noise reduction is systematically studied.On the basis of detailed analysis of the causes and propagation paths of electromagnetic interference of wind turbines,specific noise reduction measures are proposed from four levels: noise reduction by shielding grounding and insulation isolation at sensor network level;noise reduction by designing anti-aliasing filter at monitoring data acquisition hardware platform level;over-sampling at software level and using wavelet transform and empirical mode decomposition(EMD),which realizes the noise reduction from sensor end to analysis data end.The proposed method has been successfully applied to the actual wind power field,and a good noise reduction effect has been achieved.A fault diagnosis method for wind turbines based on fuzzy expert system is proposed.This method extracts fault feature vectors from monitoring data by designing fault models.After the fuzzification process,feature vectors are used as the fact input of fuzzy expert system,and the wind turbines online fault diagnosis is completed under the action of the fuzzy knowledge base and the fuzzy inference engine.At present,this method has been successfully applied to the actual wind power field,which has realized the automation and batch operation of wind turbines fault diagnosis,and greatly improved the efficiency of wind turbines fault diagnosis.This paper presents a fault diagnosis method of visualizing fault features.The phase space reconstruction technology of chaotic system is introduced to transform the nonlinear time series reflecting the health status of wind turbine components into a high-dimensional analytical model.Singular value decomposition(SVD)is applied to reduce the dimension of the high-dimensional analytical model to three-dimensional space,and the nonlinear time series representing the health status of components is transformed into the moving trajectory of coordinate points in three-dimensional space,so that the fault of wind turbine components can be visually judged.Field application shows that the method does not need spectrum analysis,reducing the skill requirements of fault diagnosis analysts,and greatly improves the efficiency of fault diagnosis.Proposed the fault diagnosis method based on convolution neural network.The key of this method is to obtain complete training samples.However,with the development of large-scale wind power,it has not yet experienced a complete design life cycle,so it is unable to obtain the complete data of the actual operation of components.This paper constructs a time series corresponding to the fault characteristics in opposite direction,and forms a training sample together with a small amount of actual fault data on site to complete the training and testing of the convolutional neural network.Field application shows that the fault diagnosis method proposed in this paper can realize the batch and automation of wind turbine fault diagnosis,and has high practical value.A method for predicting the residual life of wind turbine bearings based on discrete cosine transform convolutional neural network is proposed.This idea is: Firstly,wavelet transform is performed on the vibration signal of bearings working process to obtain the time-frequency image that characterizes the bearing degradation state.Secondly,bilinear transform is introduced to reduce the dimension of time-frequency image features.Finally,discrete cosine transform is used to compress the dimension-reduced time-frequency image sparsely,and the compressed image is used as the training sample of convolution neural network.It reduces the network complexity and improves the network efficiency.The experimental results show that the proposed method has obvious efficiency advantages compared with other prediction methods,and can be implemented on the existing hardware platform of wind turbine monitoring system,which improves the utilization level of monitoring equipment and avoids secondary investment of wind farm.The research content of this paper is directly applied to actual production.According to different application scenarios of wind farms,three fault diagnosis methods proposed have been applied to more than 2000 wind turbines.A large number of field application results show that the method proposed in this paper can effectively reduce the average maintenance time and reduce the operation and maintenance cost of turbines,and it has significant economic benefits and promotion value.