Study On Blind Separation Techniques Of The Vibration Sources In Mechanical Systems

Vibration source identification plays a crucial role in optimal design,vibration and noise control,condition monitoring and fault diagnosis for mechanical systems,which also will be of great theoretical significance and applied value.However,the excitation situations as well as the positions of the vibration sources are complicated in practical mechanical systems and are hard to be predicted accurately.In addition,it is impossible to identify the system characteristics exactly.All these problems would bring lots of difficulties to existing vibration source identification methods.Moreover,blind source separation methods face a number of difficulties when applying to mechanical systems:the number of vibration sources is hard to estimate,and blind deconvolution is obviously a complicated task,and so on.Based on aforementioned problems,research on blind separation techniques of vibration sources in mechanical systems was carried out in this dissertation.Accordingly,the main works and contributions of this dissertation are summarized as follows:The difficulties of blind source separation(BSS)applying to mechanical systems are analyzed,and the BSS model of mechanical system based on modal analysis is extracted contrapuntally.The issue of dealing with estimation of blind deconvolution filter coefficients can be avoided.Moreover,the physical significance of BSS separation results is also clarified,namely,the separated signals correspond to the individual contributions of original excitation sources acting on mechanical system,and the estimated mixed matrix contains contribution information of various excitation sources for the dynamic responses.The influence factors on principal component analysis(PCA)for preprocessing are investigated,and experimental study is also conducted on the application of PCA to the wide band vibration signal preprocessing.Considering quasi cyclic wave characteristics of the modal response under different excitation,they are assumed as harmonic signals.Their correlation and independence conditions are analyzed theoretically,and kurtosis is introduced here for analyzing the independence quantitately.It could be found that harmonic signals with different frequencies exhibit time-delay incoherence and independent time structure characteristics.Moreover,the signals can be more independent when the harmonic signals correspond to the Fourier spectrum.The simulation of blind source separation of harmonic signals is carried out,and the simply-supported beam is chosen as an example for the experimental study of modal parameters blind identification.The results show that independent component analysis(ICA)would lead to the overlearning situation and more independent signals would be separated firstly,meanwhile the second order blind identification(SOBI)method would separate the signals more stably and accurately.Blind separation of the vibration sources in mechanical systems under environmental excitation is equivalent to the problem of modal parameters blind identification.Taking an elastic plate structure as research model,SOBI is used to identify the modal parameters blindly.Aiming at the over determination issue in modal test that the available sensing number is larger than that of excited modes,a mode selection method based on correlation analysis is presented.The modal assurance criterion is also calculated for the selected modes.The experiment results show that the proposed method can effectively remove the fake modes from the separated signals with similar frequency spectrum.Hilbert transformation technology is utilized to guarantee the accuracy of the identified damping coefficients.Based on the mode selection method,SOBI method can effectively and stably identify the modal parameters,and have some potential advantages over existing operational modal analysis(OMA)techniques.A BSS method based on phase elimination technique is proposed in this dessertation,which is aimed to deal with the strong coupling problem that the mechanical responses under working condition not only contain system parameters but also correlate to the excitation characteristics.The reconstructed response model could coincide with the BSS model by synchrophasing the contribution components of the same excitation source.The simulation investigation on vibration source blind separation is performed by using a three degree freedom model.Moreover,the proposed method is also verified on a complicated mechanical system-floating raft isolation structure.Simulation and experiment results show that the BBS method based on phase elimination technique can separate the individual contribution from different excitation sources,and the estimated mixed matrices can accurately characterize the individual contribution information of excitation sources with satisfactory robustness to noise disturbance.