Research On Recognition Of Welding Spot Quality Of Car Body Based On Support Vector Machine

Resistance spot welding is the main connection technology in the manufacturing process of car body-in-white,so in order to ensure the assembly quality of car body-in-white and the rigidity and strength of the whole structure to meet the requirements of laws and regulations,the first thing is to ensure the qualit y of welding spots.Tradit ional destructive testing can not guarantee the qualit y of each welding spots because only some samples can be extracted for testing.As a widely used non-destructive test ing technology,ultrasonic testing technology has the advantages of low cost and simple operation.It can realize the quality detection of the white body welding spots with full coverage.SVM is a classification algorithm based on statistical learning theory and structural risk minimization.It can learn a lot of information about the qualit y of welding spots contained in the echo sequence of ultrasound,and then recognize the qualit y of welding spots intelligently.However,in practical applications,most of the welding spots are qualified,only a small number of welding spots are not qualified,so the identification of welding spots quality is a problem of unbalanced data set classification.In this paper,the characteristics of qualit y identification of welding spots are studied.Firstly,the basic theory of ultrasonic detection technology,the theoretical basis of support vector machine and the general processing method of unbalanced data se ts are introduced,which provides theoretical support for the later research content o f this paper.Secondly,the main parameters affecting the performance of SVM classificatio n model are introduced.Then a parameter selection method of SVM classificatio n model based on genetic algorithm(GA-SVM)is proposed.The penalty factor C and Gaussian kernel function parameter g of SVM classification model are binary coded,and the initial population is generated randomly.Then the selection,crossover and mutation operations are performed to generate a new generation of population,and the new generation of genet ic algorithm is started until the stopping criterion o f genetic algorithm is satisfied and the optimal C and G values are output.The experimental results show that GA-SVM can effectively reduce the randomness and time-consuming of tradit ional SVM classification model parameter selectio n methods.Then,the basic principle of K-Means algorithm is introduced,the advantages and disadvantages of SMOTE algorithm are analyzed,and an optimization method based on K-Means is proposed to solve the problem that it may destroy the distribution characteristics of welding spot data set samples and the boundary between fuzzy qualified welding spot and unqualified welding spot.In this method,K-Means is used to preprocess the data of unqualified welding spot.Based on the clustering center,the improved interpolation formula is used to replace the original interpolation formula,and the newly generated data of unqualified welding spot are controlled within the corresponding clustering range of unqualified welding spot.The experimental results show that the improved SMOTE algorithm based on K-Means(KM-SMOTE)can effectively improve the performance of classification model in dealing with unbalanced data sets.Finally,the characterist ics of ultrasonic welding spots detection technology are analyzed,and the attenuation rate,the number of bottom echoes,the number o f middle echoes and the distance between bottom echoes are selected as the characteristic values of welding spots quality identification.Five thousand welding samples were fabricated under simulated real welding environment,and data were collected by professional ultrasonic flaw detector.The collected ultrasonic data were analyzed and collated.The experimental results show that the proposed method can effectively identify the quality types of welding spots.