Research On Gearbox Fault Classification Method Based On Hidden Markov Model

With the development of science and technology in modern industrial production,the mechanical equipment as the main production tool is constantly developing in the direction of complexity,high speed,high efficiency,etc.At the same time,it is also facing a more severe working environment.Gearbox is the most widely used mechanical component in petrochemical,electric power,metallurgy,machinery,aerospace and other sectors,and it is also one of the most vulnerable components.Its working state has a significant impact on the operation of the entire mechanical equipment and even the entire production line..Therefore,effectively diagnosing the running status of the equipment and preventing the occurrence of unexpected situations in time are the problems that need to be solved urgently.To this end,this thesis takes gearboxes as the research object and conducts in-depth research on fault classification and diagnosis methods based on hidden Markov model(HMM).The main contents are as follows:1.Explains the background and research significance of the topic from two aspects of theory and engineering,focusing on the advantages and disadvantages of neural networks,support vector machines and hidden Markov models,and other computer-based intelligent algorithms in fault diagnosis applications.Based on the analysis of the geometric structure of the gearbox,the generation and transmission of vibration signals are introduced,the mechanism of the gear under different failure types is analyzed,the vibration signals under different failure modes are simulated,and the time domain of grasping the vibration signals of different failure types is understood And frequency domain characteristics.2.According to the non-linear characteristics of the vibration signal,the three parameters of time domain,frequency domain and time-frequency domain are used to extract different vibration signal characteristic parameter sets to describe the state of the gear.Principal component analysis(PCA),compensation distance evaluation technology(CDET)and improved compensation distance evaluation technology(MCDET)are proposed to reduce the feature set of the extracted sample feature parameters respectively.The characteristic index of the dimension is transformed into several comprehensive indexes representing most of the information of the gearbox running signal status3.Introduce the basic concepts and algorithms of Markov chain and HMM,and discuss the evaluation,decoding and learning problems and basic algorithms of HMM.The gearbox was extracted from the time domain,frequency domain and time-frequency domain under different working conditions and reduced,and the reduced feature parameter set was standardized and scalar quantized preprocessing,and then the hidden Markov training was performed.Model,and finally carry out classification and recognition tests on the models under different working conditions to obtain the best training model4.The results show that the improved compensation distance assessment technology(MCDET)reduces the recognition rate of the test samples for each working condition after feature reduction,which is higher than the other two feature reduction methods.