Research On Pattern Recognition Of Defect Ultrasonic Testing Signal

Ultrasonic non-destructive testing is one of the five non-destructive testing methods and has been widely used in the industrial field.This paper combines wavelet packet transform(WPT)and empirical mode decomposition(EMD)to analyze the advantages of non-stationary signals.The kernel principal component analysis method(KPCA)is used to process non-Gaussian distribution data and remove irrelevant and redundant with sample classification.The advantages of the remaining feature information,the improved particle swarm optimization algorithm can effectively converge to the advantages of the global optimal solution and the support vector machine can solve the practical problems of small samples,nonlinear,high-dimensional and local mini-points.The superiority of this method proposes a pattern recognition method for ultrasonic detection of defect depth.In this paper,the classification and identification methods of ultrasonic detection defect depth are studied.The main work is as follows:First,the basic theory of ultrasonic nondestructive testing is introduced.The ultrasonic parameters and the acoustic parameters of the medium,the transmission and reflection of the vertical plane of the ultrasonic wave,the ultrasonic detection method,the defect signal display method,and the quantitative,localization and qualitative analysis of the defects are introduced,for the subsequent ultrasonic inspection.The experimental implementation of the defect was theoretically prepared.Secondly,it mainly combines wavelet packet transform(WPT)and empirical mode decomposition(EMD)to analyze the advantages of nonlinear non-stationary signals and kernel-based principal component analysis(KPCA)in processing non-Gaussian distribution data,as well as removal and sample classification.The advantages of irrelevant and redundant feature information enable the extraction and optimization of the characteristics of the defect ultrasonic detection signal.Experiments show that the joint features extracted from the signals of each level after wavelet packet transform(WPT)and empirical mode decomposition(EMD)decomposition can fully characterize the defect signal,but it contains some redundant features and sample classification.Irrelevant,increases the computational complexity of the classifier and reduces the accuracy of the classifier.The feature optimization method based on kernel-based principal component analysis can effectively overcome the blindness of selected features,eliminate redundant feature information from joint feature groups,and find a set of special vector matrices in data space that can explain data variance.In addition,the complexity of the classifier calculation is reduced,and the accuracy and detection efficiency of the classification are effectively improved.Thirdly,the support vector machine algorithm and the improved particle swarm optimization algorithm are introduced.Firstly,the VC theory and structural risk minimization principle of statistical theory are introduced.The optimal classification surface of support vector machine,linear learning machine and nonlinear mapping of SVM are described.The support vector machine kernel function selection,algorithm model and Model parameters are described in detail.Furthermore,based on the standard particle swarm optimization algorithm,an improved particle swarm optimization algorithm is proposed for the selection of support vector machine parameters.Finally,based on the IPSO-SVM model,the process of deep defect classification and recognition mainly includes the acquisition and processing of corrosion defect depth signal data,and the feature extraction of defect signal data based on empirical mode decomposition(EMD)and wavelet packet transform(WPT).The kernel principal component analysis method further optimizes the extracted original features,the parameters and penalty factors of the kernel function based on the improved particle swarm optimization algorithm based on the support vector algorithm,and the optimized particle swarm optimization algorithm based on the support vector machine algorithm.The six steps of quantitative identification of the corrosion defect depth signal are introduced.Then the experimental system design of the corrosion defect detection,the working principle of the experimental system and the hardware introduction in the experimental system are introduced.The deep defect signal obtained by the experiment is subjected to zero-mean processing and trend item removal.84 joint time-domain dimensionless statistical features are obtained by wavelet packet transform(WPT)and empirical mode decomposition(EMD),and the kernel principal component method is adopted.optimize.The optimized samples and the unoptimized samples were classified and identified based on GA-SVM,PSO-SVM and IPSO-SVM respectively.The experimental results were compared and obtained after optimization by the kernel principal component analysis method(KPCA).The eigenvalues can effectively characterize the samples compared to the unoptimized original eigenvalues.The classification using the optimized samples is more accurate than the unoptimized sample classification,and the improved particle swarm optimization based support vector machine model(IPSO-SVM)In the classification of defects of different depths,the support vector machine model based on genetic algorithm optimization support vector machine model(GA-SVM)and standard-based particle swarm optimization has higher classification accuracy,using improved particles.The cluster algorithm optimization support vector machine significantly improves the classification ability and generalization ability of the classifier.It is an effective method to optimize the parameters of the support vector machine.The IPSO-SVM classifier model is suitable for the classification of different depth defects.The experimental results show that the ultrasonic detection defect depth pattern recognition method proposed in this paper is feasible and effective,and can be applied to pattern recognition of defect depth.