Research On Fault Diagnosis Method Of Wind Turbine Generator System

As a source of rich,clean renewable energy,wind power plays a critical role in alleviating environmental pollution and advancing the transformation of the energy structure.The secure and reliable operation of wind power equipment is a prerequisite for ensuring that its complex mechanical and electrical systems are effective and create value,and scientific maintenance is an essential means of guaranteeing the long-term efficient and steady running of wind power equipment.The complex structure and harsh operating environment of wind turbines lead to frequent generator system failures,while the failure rate of bearings as a key component of generators is high.In order to enhance the power generation efficiency and operational reliability of wind turbines,it is of particular significance to conduct in-depth research on generator system fault diagnosis methods.This work is funded by "Health Assessment and Intelligent Decision Optimization for Mechanical Equipment Operation and Maintenance(No.2018-RC-25)",a project funded by Lanzhou Talent Innovation and Entrepreneurship Project.The 2.5MW direct-drive permanent magnet synchronous wind turbine is adopted as the study object,to address the problem of high failure rate of the generator system and its vulnerable component bearings,and carry out research on its fault diagnosis method based on the monitoring data of the wind turbine,the specific work and conclusions are as follows:(1)For the problem of hard to distinguish between normal and abnormal samples of wind turbine SCADA data,in order to reduce the wind turbine failure caused by the generator,a fault diagnosis method for wind turbine generators applying GMM and Cat Boost models is proposed.A few parameters in the high-dimensional SCADA data are taken as feature criteria,and the GMM is employed to eliminate the abnormal values,after which the normal data of the wind turbine is filtered out.Using the mutual information method to select the feature variables with strong correlation with the generator system state info,the generator fault information can be extracted accurately.According to the process from the beginning of the fault trend to the formation of the fault,the condition monitoring model and the fault diagnosis model are established respectively.Through the analysis and verification of a wind turbine real fault case in a wind farm,the approach can effectively monitor the generator condition and identify various types of faults,compared with other models,TR and MR are improved by 13.1%and 19.4% separately on average,and F1 score and MAF1 score are increased by 0.106 and 0.187 respectively on average.(2)To effectively diagnose generator system faults,and to address the problem that traditional data downscaling approaches cannot better retain sample information from wind turbine monitoring data,the ideas and methods of deep learning technology are introduced,and an integrated deep autoencoder neural network(DANN)model based on deep autoencoder and deep neural network is proposed,which can efficiently diagnose generator system faults.By using DANN model to autonomously learn highdimensional SCADA data features of wind turbines and transform them into a lowdimensional feature representation,and by constructing a fault mapping relationship between the low-dimensional feature data and the generator,the DANN model can identify faults that occur in the generator system.The technique was validated applying a real fault case of a wind turbine at a wind farm.Compared to other fault diagnostic methods,the DANN model can retain as much raw information of SCADA data as possible,and the identification for various conditions of the generator system is87.01%,it is indicated that the model can diagnose faults in the generator system of wind turbines.(3)The generator bearings are subjected to alternating stress and shock loads during the operation of wind turbines,and the vibration signal is noisy and the features are not sufficiently extracted for the problem,an OVMD-RF based generator bearing fault diagnosis method for wind turbines is proposed.Firstly,the wavelet noise reduction method was applied to process the generator bearing vibration signal,in order to eliminate the noise components in the raw signal.Then,the two essential parameters of the VMD are optimised using STOA search and the parameter optimised VMD method is employed to decompose the bearing vibration signal.Finally,the fused feature dataset is constructed with peak,cliff and envelope entropy to build a wind turbine generator with the aid of RF classifier bearing diagnosis model with the help of RF classifier to identify its faults.The usability of the OVMD-RF method is verified through simulation and instance analysis,the experimental results indicate that the diagnostic approach is an accuracy of 99.2% and can effectively identify bearing faults.