Study On Coal Mine Mechanical Weld Flaw Recognition Technology Of Ultrasonic Signal

During ultrasonic non-destructive testing, feature extraction and recognition of ultrasonictesting signals for detection of mechanical flaws, reduce mechanical accident rate hasimportant significance. The selection of feature extraction method of ultrasonic signalsdirectly affects the flaw analysis of qualitative, quantitative and positioning. Mechanical weldflaw identification is essentially a pattern classification problem. Due to support vectormachine has excellent performance on solving the small sample classification problems, soget increasingly wide range of applications. Support vector machine set optimization, nuclear,and best promotion capacity to it, from the perspective of linear separable classification, itspurpose is to establish an optimal decision-making hyperplane, making the distance of theplane to the nearly two class samples on both sides of it maximize, thereby contributing to theclassification provides a good gen eralization. For nonlinear separable classification problems,according to Cover’s theorem, firstly transform it into linearly separable pattern classificationproblem to process.This thesis, from the measured ultrasonic signals of mechanical weld flaws area,combined with wavelet analysis, principal component analysis, and common classificationalgorithms to carry out the attempt to feature extraction and flaw identification work onmeasured ultrasonic signals. The main research content and results are as follows:(1) Introduction ultrasonic testing data acquisition system, collects the mechanical weldflaws in the experimental platform, analysis its waveform characteristics, and then useswavelet analysis and principal components analysis to do feature extraction.(2) Using KNN, BP neural network and SVM algorithm to do classification andidentification on standard data sets and extracted ultrasonic signals of mechanical weld flaws,and do a comparison of different algorithms and research.(3) For problems such as parameter selection is diffficult in SVM classification algorithm, using intelligent optimization algorithm for optimal parameter selection in SVM, byoptimizing the penalty factor C and RBF kernel parameter σ, do mechanical weld flawsultrasonic signal identification, improve the flaw recognition accuracy.(4) Using the method of wavelet analysis technology combined with principalcomponent analysis to do feature extraction on measured mechanical weld flaws ultrasonicsignals, and by comparing the classification results of SVM, ABC-SVM and RABC-SVMalgorithm to verify the feasibility of ABC-SVM algorithm proposed in this thesis, and higherspeed of the RABC-SVM.The experimental results show that the RABC-SVM algorithm is more suitble forsolving the mechanical weld flaw ultrasonic signals identification problem, and has certainadvantages. At the same time, this paper summarized data mining and its applications ofsignal processing in ultrasonic testing, then point out the further research directions.