Research On Separation Of Internal Excitation Response In Mechanical Vibration Signal

The use of mechanical vibration signals to estimate the vibration source is of great significance to the optimization of mechanical system design,vibration and noise control,condition monitoring and fault diagnosis.The working mechanical system is subjected to various external or internal vibration sources,and the vibration sources caused by component failures are all internal excitation sources.Obtain the response generated by the internal excitation from the vibration signal of the mechine,and evaluate it separately,which is more conducive to accurate and robust condition monitoring and diagnosis.In this thesis,by summarizing the nature of the internal excitation response and using its structural properties,three separation models based on low-rank recovery technology are studied,and the internal excitation response is obtained from the constructed observation matrix.Research on the nature of internal stimulus response and source number estimation method.Defines the internal stimulus response and summarizes the three properties of the internal stimulus response.Convolutional sparse coding is used to characterize the internal excitation structure and its distribution law.The low rank of the matrix is used to measure the similarity of the internal excitation response structure under constant speed and variable speed conditions.Time-frequency analysis technology is used to estimate the number of sources under constant speed conditions.Order tracking technology is used to estimate the number of sources under variable speed conditions on the angular domain order spectrum and envelope order spectrum.Research on separation based on low-rank sparse model.The principle of the model and the process of gradually realizing separation are described.The robustness of the model is studied from the three aspects of structure deformation,interference signal energy and the number of structure samples.Take the mixed signal of rolling bearing as an example to carry out the separation experiment.This model can initially realize the separation of the inner excitation response,but the separated inner ring fault response loses part of the modulation characteristics.In essence,the separated internal excitation response loses part of the amplitude information of the structure.Analyze the reasons for the poor performance and robustness of the model.Research on separation based on low-rank noise model.The statistical distribution of interfering signals in the process of single-low-rank recovery bisection is studied.A Gaussian noise term that is more in line with the interference signal distribution constraints is constructed,and a low-rank noise decomposition model is proposed.The influence of the regular parameter β of the noise term on the model is studied,and the parameter value range is determined.Using the mixed signals of rolling bearings,the separation performance of the model is compared with that of the low-rank sparse model.The fault response of the inner ring separated by the model has better modulation characteristics and higher signal-to-noise ratio.Research on separation based on multi-low rank noise model.Three low-rank noise models are constructed,and the solution methods are given.Compared with the previous two models,this model is a multi-objective optimization model,which can separate multiple internal excitation responses at one time.The model adds a priori constraints,and its results are more directional.When the number of internal excitation sources is underestimated,the separation effect decreases slightly.Take the simulated mixed vibration signal of rolling bearing as an example,and make a comparative analysis with the low-rank noise model.Use the measured signal as the research object to conduct a comparative experiment with the traditional EEMD-ICA blind source separation method.