Research Of Feature Extraction And Recognition Technology Of The Cylindrical Component’s Defect Based On The Ultrasonic Testing

The cylindrical component is widely used in aerospace, automotive, defense, ammunitionand other fields. In the production process, it is easy to cause defects such as porosity,inclusions, which can affect the final performance of the product. Therefore it needs to betested to ensure the normal use in the process of production or usage. In this paper, cylindermembers are treated as object of study. By using ultrasonic technology to research Echosignals’ processing and feature extraction techniques of defect feature in the inner memberand realize the type recognition and3-D reconstruction of defect.Firstly, analyzing echo signals’ noise characteristics of the cylindrical member in theprocess of automatic detection based on ultrasound. Using adaptive threshold noise reductionalgorithm of wavelet to process signals in order to solve this problem. By analyzing the affectof threshold for noise reduction, the threshold is determined by genetic algorithms. The resultsshow that the proposed method significantly improves the SNR of characteristic echo.Secondly, in view of identify of defects’ types, the identification scheme based on neuralnetwork is formulated. According to the echo characteristics of typical defects, means, energy,envelope variance, signal width and bandwidth are selected as the inputs of the neuralnetwork. Using the BP neural network to make training, identification and validation for thesample data and the correct rate of identification is higher than90%.Finally, after determining the position of the defect, using contour reconstruction methodto connect defect profile contour of the key point. After calculating the central point of thecontour, connecting the center of each contour plane and achieve the three-dimensionaldisplay of defect. The results show that the method can show a more complete form of variousdefects.