Research On Weak Fault Feature Extraction And Identification Method Of Single Loop Gear System

Recently,as the increasing of requirements for mechanical equipment in industrial application,the transmission system of mechanical equipment has developed in the direction of multi-degree of freedom and multi-loop transmission.As a result,the mechanical structure and dynamic characteristics of the transmission system have been complicated,which is difficult to ensure the accurately and stability.Therefore,real-time detecting has become an important method to solve the problems.The method of vibration singnal detecting has been wildely studied as it is easy-to-operate,low-cost,and theoretically efficient.However,for the complex transmission system,it is difficult to accurately analyze the characteristics of the motion by traditional methods,and it is difficult to accurately clarify the mechanism and physical characteristics of the fault signal,which affects the effect of fault identification.In this research,based on the objects which are early,middle and late stages of the single-cycle gear train root crack failure,the purpose is to develpe a monitoring method for the fault identification in different stages.Therefore,the study is divided into four major steps.Firstly,based on the dynamic features of the single-loop gear system and combining the features of gear meshing,the theory of the vibration signal in the normal and fault state are summarized as the theoretical basis for the fault identification;Secondly,the improved singular value decomposition method is developed to de-noise the signal;Thirldly,the improved multi-scale permutation entropy method is developed to extract the fault feature information of the system.Finally,the combined weighted Mahalanobis method for the identification is developed.In addition,the experiments of single-loop gear root crack testing was conduected to verify the method.The details are listed as follows:(1)The bond graph model of linear system of XP style and PX style in normal state of the single-loop gear system are established,and the bond graph model of non-linear system of XP style and PX style in fault state with non-linear factors such as time-varying meshing stiffness of the gear teeth,tooth surface friction,tooth surface damping and steady-state transmission error are established.Based on the models,the dynamic and response principle of the vibration signal in each system are concluded and clarified.(2)For the de-noising processing of the vibration signal of the single-loop gear system,the autocorrelation function method and the improved pseudo-nearest neighbor(Cao’s)method for the existing singularity is adopted to improve the SVD method for the de-noising.The signal-to-noise ratio(SNR)and root mean square error(RMSE)methods is adopted foe the evaluation in simulation analysis and experiments.The experimental signal SNR>26dB,RMSE<1.3,which shows that the improved SVD method has better noise reduction effect.(3)In order to ensure the completeness and accuracy of fault feature extraction,vibration feature information of single-loop gear trains is extracted by the multi-scale permutation entropy method(MPE)which is improve by the mutual information entropy method and Cao’s method.Then the effect is evaluated by Lyapunov exponent method.Finally,based on the verification by simulation and experiments,the complete information of the fault characteristics are accurately extracted by MPE.(4)To the accuracy and reliability of the fault identification of the root crack of the single-loop gear system,the optimal weighted of fault feature information is calculated by the combined weighted Mahalanobis method of AHP and information entropy.Then,the experiments of idenfication of tooth surface wear and tooth surface pitting faults is conducted to verify the effect.The results showed that the combined weighted Mahalanobis method identifies the weak feature information accurately and efficiently.(5)Finally,in order to verify the identification effect of single-loop tooth root crack failure,experiments in the condition of early,middle and late stages of the single-loop gear train root crack failure is conducted,the result of the accuracy rate has reached 99%,which showed the method is effect to single-loop tooth root crack faults identification.