A methodology is developed to detect defects in NDT of materials using an Artificial Neural Network and signal processing technique. This technique is proposed to improve the sensibility of flaw detection and to classify defects in Ultrasonic testing. Wavelet transform is used to derive a feature vector which contains two-dimensional information on various types of defects. These vectors are then classified using an ANN trained with the back propagation algorithm. The inputs of the ANN are the features extracted from each ultrasonic oscillogram. Four different types of defect are considered namely porosity, lack of fusion and tungsten inclusion and non defect. The training of the ANN uses supervised learning mechanism and therefore each input has the respective desired output. The available dataset is randomly split into a training subset (to update the weight values) and a validation subset. With the wavelet features and ANN, good classification at the rate of 94 % is obtained. According to the results, the algorithms developed and applied to ultrasonic signals are highly reliable and precise for online quality monitoring.
Keywords: Non-destructive testing, Defect classification, Wavelet transform, Artificial neural networks
[...] The progress in computational techniques, specifically the development of neural networks, has greatly stimulated the research into the development of automatic systems for the inspection and the classification of defects in engineering materials Previous Related Work In Ref. an evaluation of various types and configurations of neural network developed for the purpose of assisting in accurate flaw detection in steel plates is illustrated. The obtained results indicated that significant benefits may be obtained from the techniques demonstrated with no form of feature extraction employed. [...]
[...] From these data sets for each class are selected at random to create a separate test set consisting of 50 sets which is used for testing the trained network. The partial sample data for training the NN model is represented in Table 2. Table 2 : Partial Sample data for training the NN model Mean Variance Maximum Minimum Maximum Average Minimum Half Type of amplitude amplitude energy frequency frequency point Defect - Porosity - LF - TI 32.5188 - ND - Porosity - LF - ND - TI 88.8235 - Porosity - ND - LF - TI 102.6234 Before training the network, the above data were normalized suitably. [...]
[...] One class of signals from regions presenting no defect to identify signals from welds with defect or welds that presented no defects Experimental Procedure The Research work was carried out in the following steps: Collecting certain number of ultrasonic oscillograms for different type of defects and their digitalization. Extracting the features by using signal processing technique called wavelets. Training the artificial neural network (ANN) to classify defects. Testing the trained network for verification Material In the pulse echo technique, inspections were performed on nine test specimens made of stainless steel plates of 5 mm in thickness and 200 mm in length. [...]
[...] Conclusions In this paper, a novel method for classification of signals in NDT of materials is developed using ANN and DWT. The implemented configuration of ANN (An Artificial Neural Network with signal processing technique) showed a reasonable rate of success to classify patterns of ultrasonic signals obtained from four classes of defect in stainless steel plates extracted by pulse echo technique. The ANN model develops a fast and user friendly system, which assists practicing technicians by reducing the time spent in classifying the defect signals obtained through ultrasonic testing. [...]
[...] Fig.3: Ultrasonic signal for lack of fusion Fig.4: DWT coefficients representation for lack of fusion 1024 samples 256 samples Wavelet features Features for discrimination of detected echoes are extracted in discrete wavelet representation. In this study features are extracted from the each signal of the four classes. The extracted features from the signal and their relationship are as below: 1. Mean 2. Variance: m = ( 1 n m i=1 n i v = 1 n n i=1 i m n 3. [...]
APA Style reference
For your bibliographyOnline reading
with our online readerContent validated
by our reading committee