Research On Composite Fault Diagnosis Method For Gearboxes Based On UBSS

As one of the most widely used mechanical structures,the fault diagnosis of rotating mechanical equipment is helpful to improve the safety and reliability of mechanical equipment,and is of great significance to the practical application of engineering.Modern machinery and equipment continue to develop in the direction of precision technology and complex structure,and the performance and industrial parameters of machinery and equipment have been greatly improved,followed by the requirements for high reliability and high robustness of modern machinery and equipment.Vibration signal is one of the main tools to analyze the state changes of mechanical equipment.In the face of complex working conditions and structural vibration coupling of modern mechanical equipment,its use performance also has obvious limitations.Aiming at the above problems,this paper introduces the underdetermined blind source separation(UBSS)method to study the typical weak or early fault diagnosis of rolling bearings and gears.The main research contents are as follows:(1)The time-frequency analysis theory(TFA)for weak or early faults has been studied.In view of the traditional Fourier transform(FFT)and continuous wavelet transform(CWT)time-frequency conversion process,the frequency resolution can not effectively match the distribution of modal components in the vibration signal,this paper proposes a vibration signal time-frequency analysis method based on adaptive cepstrum coefficient(AFCC)and parametric simultaneous extraction transform(PSET),and uses AFCC method to adaptively obtain the time-frequency resolution matching with the vibration signal segment,A triangular filter bank is constructed to preliminarily reduce the noise of the signal and improve the signal to noise ratio of the vibration signal;Then,the distortion frequency scale obtained from AFCC is combined with SET as a parameter to improve the efficiency of SET method,and the proposed PSET method is used to concentrate the time and frequency domain energy of the signal,reduce the impact of noise,and improve the sparsity of the main components in the signal.(2)The mixed matrix estimation method of blind source separation(UBSS)under the condition of underdetermined signal acquisition source is studied.The sensor installation position of rotating machinery equipment is limited,and there are coupling components in the collected multi-channel signals,and the vibration signal also presents the characteristics of multimode non-stationary due to the influence of the working environment.In order to solve the above problems,a time-frequency scatter diagram is introduced,and a vector detection(TFSVD)denoising algorithm based on time-frequency scatter diagram is proposed,which utilizes the central symmetry rule of the vibration signal modal components in the time-frequency scatter diagram,The extraction of signal modal principal components in time and frequency domain solves the noise impact in multi-channel blind source separation process,and improves the sparsity of multi-channel signal principal components.Then,the class cohesion algorithm(CCA)is used to estimate the parameters of the mixing matrix,which solves the problem of low estimation accuracy of the mixing matrix occasionally caused by the signal,and realizes the vibration source separation of rotating machinery under the condition of underdetermination.(3)The UBSS model proposed in this paper is experimentally verified.The experimental data of rolling bearings and gears under fault conditions are collected respectively,and the modal components of vibration signals are separated using the proposed method.The effectiveness of the method is verified by comparing the modal components obtained from the separation with the fault frequency obtained from the actual calculation.The progressiveness nature of the proposed method is further verified by comparing the performance with the parameter optimized variational mode decomposition(AVMD).