Researchonthe Decision Method Of Fault Diagnosis For Complex System Based On The Gated Recurrent Unit

With the rapid development of large and complex equipment,reliability management has become one of the hot issues in the field of management.In the research of equipment reliability management,more and more attention has been paid to the effectiveness of fault diagnosis decision-making methods.However,the hidden system reliability rules in monitoring signals have not been fully mined and utilized.According to the reliability theory of complex mechanical system,based on the vibration signal and current signal data of the monitored equipment,this paper uses the idea and method of reliability management analysis of intelligent algorithm for reference to carry out causal analysis and fault diagnosis of complex system fault.Thus,the decision-making management method of fault diagnosis for complex system is studied,and the Enlightenment to major project management is discussed.As a typical complex system,wind turbine is the key infrastructure to realize the conversion of electric energy by using clean energy.Scientific maintenance and management is of great significance to its normal operation,guarantee the supply of renewable energy and promote the development of national economy.However,the data of fault state shows the characteristics of massive and diverse big data,and it has limitations in analyzing data and building models completely depending on traditional fault detection methods.With the development of intelligent algorithm,its ability to process and represent big data has been reflected in various fields well.Therefore,this paper takes the gearbox of wind turbine as an example,uses machine learning algorithm to learn the characteristics of each fault and realize fault classification on this basis.The main research contents are as follows:Firstly,the introduction of basic theory.In this paper,the content,method and application of complex system fault diagnosis are analyzed and summarized,and the structure and common fault characteristics of the gearbox of wind turbine as a large key equipment are introduced.This paper describes the test-bed used in this paper to simulate the operation mode of wind turbine,and briefly describes the data acquisition and operation conditions.The principle of several fault diagnosis models used in this paper is introduced in detail,including support vector machine,random forest,long and short term memory network,gating cycle unit and echo state network.This paper analyzes the loopholes and defects in the existing fault diagnosis decision-making methods,expounds the research ideas of fault diagnosis based on neural network feature extraction,and summarizes the framework of the whole paper.Secondly,single fault diagnosis modeling based on original sample and GRU.The vibration signals in three directions under six states of ring gear are collected by simulation test-bed.Then several typical machine learning models,such as support vector machine,random forest,long and short term memory network,echo state network andgated recurrent neural network are used to identify the original vibration data of the ring gear and analyze the learning effect of each model on the fault characteristics Visual analysis.The results show that in the absence of feature extraction,the echo state network has the strongest feature learning ability and the best classification effect.Generally speaking,the overall effect of the five models is unsatisfactory.Thirdly,single fault diagnosis modeling based on artificial feature extraction and GRU.On the basis of gear ring state recognition,the vibration signals collected are extracted by artificial features which include the time domain,the frequency domain and the time-frequency domain,and then classified by machine learning model.The results show that after feature learning,the fault recognition performance of the five models is generally significantly improved.The effect of fault classification is the best.Fourthly,multi fault diagnosis based on GRU.According to the multi fault characteristics of complex system,a fault diagnosis strategy based on multi-sensor feature layer fusion is developed to better fit the fault diagnosis of large-scale equipment in actual production.By monitoring the current signal and vibration signal,the comprehensiveness of the diagnosis source is increased.It is difficult to extract features manually to meet the needs of multi fault system diagnosis.On the basis of single fault diagnosis experiment of complex system,this paper proposes a method of combining artificial feature extraction and neural network feature extraction to form augmented feature vector,and then input it into the gating cycle unit for feature classification.The results show that to some extent,this method can obtain more accurate state estimation,make full use of sensor resources,maximize resource utilization,and improve the diagnosis accuracy of fault diagnosis method.