Ultrasonic Identification Of Steel Plate Defects

As an important material for the production of various large-scale equipment,steel plate has been widely used in power systems,automobile manufacturing,aerospace and aviation,medical equipment and other fields.However,in the production and later processing of steel plates,some quality problems will inevitably occur,such as cracks,pores and inclusions.These defects will directly affect the overall operation level of the equipment and even cause some security incidents.Therefore,it is of great significance to study the defect detection and identification technology of steel plates.Ultrasonic detection is used in defect detection of various equipment due to its advantages of being harmless to the human body and the environment,low cost,and high recognition accuracy.However,most of the currently used technologies are manual scanning and manual identification,which has low detection efficiency and poor reliability.Therefore,it is necessary to explore automatic defect detection and identification technology,the core technology of which includes signal processing and identification technology.The following are the main research contents of this paper:(1)A signal acquisition system for steel plate defects was constructed based on virtual instruments,and 1.4mm SUS316 thin steel plate was selected as the research object of this paper.The echo signals of these defects are collected as experimental data.(2)The noise reduction theory of ultrasonic echo signal is studied,and a signal noise reduction algorithm based on wavelet packet block threshold processing is proposed.On the basis of wavelet packet decomposition,a JamesStein block estimator is introduced to divide the signal into blocks,and the optimal estimation of the threshold of each block is performed,so as to perform threshold processing on different blocks.The reconstructed signal can effectively remove some noise signals.Finally,compared with the wavelet packet soft threshold processing and the noise reduction algorithm of literature[36],the signal processed by this algorithm has obvious advantages in both signal-to-noise ratio and root mean square error.(3)The feature extraction and identification of the denoised signal are studied.A feature extraction algorithm combining wavelet packet energy spectrum and independent component analysis(WP-ICA)is proposed.The energy of each frequency band of the signal is calculated as characteristic parameters,and these characteristic parameters are analyzed and optimized by independent components and input into the SVM classifier as characteristic samples to realize the training and identification of steel plate defects.Compared with the recognition result of the feature vector extracted from the wavelet packet energy spectrum,the recognition accuracy of the WP-ICA algorithm is improved by 9.77%.Finally,the recognition performance of SVM classifier and BP neural network are compared under the same number of samples,SVM classifier is 8.6s and 12.9% higher in speed and accuracy,respectively.