Deep Learning-based Fault Diagnosis And Analysis System For Wind Turbine Equipment

As an alternative energy source,wind energy has the advantages of being clean,non-polluting,renewable and ‘green’.Wind energy solves many of the pollution problems caused by traditional fuel oils,and is abundant and inexpensive.The economic and social benefits of these advantages have stimulated the development and use of wind energy,but wind technology is not yet mature in terms of manufacturing and the maintenance of required equipment.Current wind turbine fault identification methods,including model diagnosis,knowledge diagnosis and machine learning,all suffer from limitations.These include difficulties in establishing accurate models,general feature extraction capabilities,low accuracy and a reliance on large amounts of historical data,all of which can prevent timely and effective identification of wind turbine faults and lead to significant increases in maintenance and repair costs.In this project,a high-performance wind turbine equipment management and fault diagnosis system is proposed and implemented to help address the above problems.The main work and results of this research are as follows.Firstly,as the current wind turbine data acquisition system transfers raw data packets from the collector to the server periodically through an interface,this research designs and implements an efficient method of parsing and converting raw data to improve the extraction of feature values and enhance storage capacity in the database after parsing and converting the raw data.A high-performance hybrid neural network fault diagnosis method,based on a bidirectional gated recurrent unit and a one-dimensional convolutional neural network,is then proposed to address existing issues in wind turbine fault diagnosis.These include difficulties in the modeling process,general feature extraction capabilities,poor generalisation capabilities,low a priori knowledge and low accuracies.The output is used as the input for the bidirectional gated recurrent unit,which can obtain both forward and reverse cumulative dependence information to further extract the long-term dependence features of the sequence.Initial experimental results show that compared to mainstream fault diagnosis methods used in the industry,the proposed method 1D-CNN＿Bi-GRU has a higher classification accuracy and robustness in solving small sample problems.The method also shows excellent performance and generalization ability when run on real data sets,with a diagnosis accuracy of over90%.Finally,this research combines Spark,a general-purpose computing engine for large-scale data processing,with Spring Boot and My Batis frameworks,the current mainstream web application frameworks.Based on the initial research results,this combination streamlines the complete process from data monitoring to fault diagnosis,by delivering an efficient and high-performance method to monitor the status of wind turbines,diagnose adults and provide operation and maintenance personnel with technical support,thus saving maintenance costs.The implementation of this system has important research value and practical significance for ensuring the efficient and continuous operation of wind turbines by detecting faults as soon as possible and reducing equipment maintenance costs.