Defect identification method for aluminum plate based on electromagnetic acoustic technique

Currently, defect types were desired to obtain in order to guide the improvement of the production process. A defect identification method for the aluminum plate based on electromagnetic acoustic technique was proposed. Artificial defects such as flat and round bottom holes were used to represent the real defects in aluminum plates and multiple sets of echo signal samples were obtained by the bulk wave electromagnetic acoustic transducer (EMAT). The adaptive neighboring coefficients method based on wavelet was adopted, which effectively increased the signal noise rate (SNR). Multiply feature extraction methods were proposed to extract 45 kinds of signal features in the time domain, frequency domain and time-frequency domain. Optimum feature vector was obtained through the class distance separation criterion and sequential floating forward selection (SFFS) methods, which effectively deduced the dimension of the feature vector. The k-fold cross validation method was applied to choose the classifier's parameters. The experiment shows that the design of the support vector machine (SVM) classifier can identify the flat and round bottom holes effectively. The recognition accuracy rate is 96.7%.