Research On Blind Separation Technology Of Vehicle Structural Vibration And Noise

Along with development of automobile industry，people also propose higher requirements for thecar’s performance and comfort, so vibration and noise problems increasingly appear. As a complexmechanical system, there are often many vibration and noise sources on the vehicle. In order toimproving vehicle’s NVH performance, it needs experiments to find the vibration and noise sources.But the signals received by the sensors are often the mixtures of many source signals, and this affectsus to identify the vibration and noise sources.The usage of blind source separation (BSS) in mechanical vibration and noise signal processingand fault diagnosis is still in the initial stage. In this article, a further study on the existing BSSalgorithms and its theory was done. Blind source separation technology was used for time-domainmodal parameter identification; single channel blind separation and multi-correlated sources blindseparation method were studied. The main contents of this paper are summarized as follows:(1) The mathematical model, assumptions and uncertainty of BSS were studied in details. Thedetails concepts and separation criteria related to blind source separation algorithm were summarized.Three typical signal separation algorithms for mechanical vibration and noise sources separation werestudied.(2) Through comparing the similarly between time-domain modal parameter identification modeland blind source separation model, the structure modal parameter identification method using BSSwas studied. Four degrees of freedom structure model and a simple beam model were analyzed formodal parameter identification, including natural frequency, modal damping ratios and mode shapesof the matrix. The recognition accuracy of different algorithms was compared. The separationperformance and the precision of modal parameters identification for signals with strong noise byusing different blind source separation methods were studied, and a high order modal frequencyidentification method was proposed.(3) Combining EMD and BSS method, a single-channel BSS method for vibration signal wasproposed. Using EMD method to decompose the acquisition signal, and then new observed signalswere reconstructed by choose the instinct modal function which we are interested in. The number ofsource signals was estimated by using the source number estimation method base on power spectraldensity. The separated signals were gained by doing BSS to reconstruction observed signals and theoriginal observed signal. Through simulation analysis, measured vibration signal analysis of a breakdown truck, and vibration signal analysis of a turbocharger system, the correctness and validityof the method were verified.(4) For multi-correlated vibration and noise signals, the separation performance of waveletpacket sub-band blind separation method for correlated source signals was studied. The mixed signalswere decomposed to sub-bands by wavelet packet approach. The correlation of each sub-band wasevaluated by mutual information. And then new mixed signals were reconstructed by selectingsub-bands with less dependent components, do BSS to the reconstructed mixed signals to obtain theseparate matrix. Through simulation analysis, motor noise signal analysis and rotor vibration signalanalysis, the correctness and validity of the method was verified.