The Feature Extraction And Pattern Recognition Of Engine Abnormal Sound

Abnormal sound refers to under the action of external force,the surface friction or impact of two or more adjacent parts caused the sound.The main excitation source of automobile abnormal sound was the road,the driver's operation and component failure,etc.Cars,motorcycles have a lot of abnormal parts which could cause abnormal sound,such as the engine,suspension,body,chassis and so on.This article mainly discusses the motorcycle engine abnormal sound.As the consumers become more concerned about motorcycle comfortable,motorcycle NVH performance is also more and more attention by the vast majority of motorcycle companies.Therefore,Motorcycle engine has a sound detection before logoff link,to avoid the engine with abnormal problems flow to the market.At present,the engine abnormal sound detection mostly adopt the method of artificial auscultation,the workers must through auscultation equipment to determine whether it is an abnormal sound engine.Due to the differences in personal technical level and actual experience,this traditional abnormal sound recognition method has a lot of contingency and subjectivity.Moreover,There is a strong background noise in the method of artificial auscultation,and working long hours can also cause damage to the operator's body.Therefore,this paper proposes a method based on SVM classifier which through the engine sound signal for abnormal recognition,to avoid the many drawbacks of manual operation.Engine acoustic signal gathered from the workshop mostly contains the strong background noise,they must be done to deal with the noise before analyzing data.By learning wavelet analysis theory and signal simulation,it is known that the wavelet airspace correlation filtering method can effectively remove the background noise of engine acoustic signal.At the same time,the transient pulse component of signal also be well preserved,which is consistent with our requirement for analyzed signal with high signal-to-noise ratio.Signal has two kinds of descriptions both time domain and frequency domain,the signal characteristics in addition to frequency,there are energy,loudness,sharpness and other indicators to measure.This article use wavelet packet transform to make 3 layers decomposition for the signal,and defined the square of the wavelet coefficients as the energy of the filtered signal,it just is one characteristic of the signal.In addition,Analyzing the filtered signal by double spectrum method,and extracting the correlation information of spectrum method energy peak as another characteristic of the signal.This paper will through wavelet packet analysis and double spectrum analysis to extract feature vectors as the input into the support vector machine(SVM),make vector machine training.Select the appropriate classifier kernel function and the corresponding parameters,through sample training,the vector machine has the ability of recognizing engine abnormal.With the test sample known sound categories test vector machine's generalization ability,ensure that the trained vector machine can recognize the engine abnormalities.Using a set of sampling equipment collect the engine sound signal,filtering noise,extracting the feature vector,and then through the sample training so that the classification model has a high classification accuracy,the vector machine output is the type of engine abnormal sound.In order to facilitate the analysis,with MATLAB software designed a GUI program,to avoid the cumbersome data processing process.