Research On Diagnosis Method Of Rattle Noise Of Car

With the rapid development of my country’s automobile industry and the improvement of consumers’ requirements for product quality,automobile quality problems caused by rattle have become an important problem that plagues manufacturers.In order to solve this problem,the development of rattle of automobiles is being done at home and abroad,and the rattle performance of prototype cars needs to be tested,evaluated,diagnosed and adjusted.The current rattle sound diagnosis is mainly based on subjective judgment and highly dependent on the engineering experience of the R&D personnel.There are problems of time-consuming,inaccurate,easy to make mistakes,and easy to miss judgments.In the automobile industry in my country,rattle performance development or rattle control.The work is still in its infancy,and there is an extreme lack of experienced R&D personnel for abnormal noise performance.Therefore,the important significance of the research in this paper is to provide methods and tools for abnormal sound diagnosis in the development stage,so as to meet the engineering needs of automobile abnormal sound performance development.The thesis first introduces some related analysis methods of rattle percussion noise,such as traditional Fourier analysis,short-time Fourier analysis and wavelet analysis,and points out the shortcomings and defects of these methods for the characteristics of rattle noise.At the same time,in view of the non-steady state and instantaneous characteristics of car rattle noise,it is pointed out that speaker recognition technology is an ideal tool for analyzing the signal.This article proposes two methods respectively.The first is to judge whether the signal contains a percussive sound signal according to the characteristic factor;after judging that the signal contains a percussive signal,the second method is to try to identify the speaker.It is applied to the recognition of rattle,that is,the Mel Frequency Cepstrum Coefficient(MFCC)feature quantity and the Gaussian Mixture Model(GMM)are used as the recognition model,and the knock is based on the maximum likelihood discrimination.Diagnosis of the source of rattle sound.The rattle noise signal of a car is a non-stationary signal,and the time-domain analysis of the signal is also an essential basic method.The time-domain indicator analysis of the signal is intuitive and accurate.When the signal is abnormal,some timedomain indicators will appear obvious the change.This research will select crest factor,impulse factor,margin factor,kurtosis factor and form factor as the characteristics,simulate and generate steady and percussive rattle noise signals,study the change law of its characteristics,and determine the value range of its characteristic factors.And verify its correctness in subsequent experimental links.This article focuses on the rattle sound of the seat guide rod and the seat belt retractor,discusses the feasibility of applying the speaker recognition technology to the rattle sound,and points out the characteristics of the rattle sound of the percussion and the speaker recognition.The difference in extraction is that the feature vector of the signal calculated by MFCC is used as the input of the GMM model.After selecting the appropriate model parameters,the GMM model of each rattle sound is obtained by training,and the MFCC vector set of the sound signal is tested against the target GMM.Likelihood function value,it can be judged whether the sound transmission structure of the test sound signal is consistent with the target sound transmission structure,so as to determine what kind of rattle the test sound signal belongs to.Finally,the experiment verified the accuracy of the method.When the same kind of rattle is recognized,it can be identified 100%,and when different kinds of abnormal sound are identified,it can be100% rejected.