Real-time recognition of weld defects based on visible spectral image and machine learning

The quality of Tungsten Inert Gas welding is dependent on human supervision, which can’t suitable for automation. This study designed a model for assessing the tungsten inert gas welding quality with the potential of application in real-time. The model used the K-Nearest Neighborhood (KNN) algorithm, paired with images in the visible spectrum formed by high dynamic range camera. Firstly, projecting the image of weld defects in the training set into a two-dimensional space using multidimensional scaling (MDS), so similar weld defects was aggregated into blocks and distributed in hash, and among different weld defects has overlap. Secondly, establishing models including the KNN, CNN, SVM, CART and NB classification, to classify and recognize the weld defect images. The results show that the KNN model is the best, which has the recognition accuracy of 98%, and the average time of recognizing a single image of 33ms, and suitable for common hardware devices. It can be applied to the image recognition system of automatic welding robot to improve the intelligent level of welding robot.

A novel detection of weld defects by magneto-optical imaging under combined magnetic field

To overcome the shortcomings of existing magneto-optical imaging, such as the saturation of an image under a constant magnetic field and the ambiguity of an image under an alternating magnetic field, imaging using a combined magnetic field is presented in this research. Weld defect samples include a laser-cut groove, a wire-cut penetrating groove, a pit and a Z-shaped crack. Magneto-optical imaging experiments were carried out under different magnetic fields. Contour extraction and standard deviation calculations were carried out for all magneto-optical images and the maximum standard deviation of the laser-cut groove under an alternating magnetic field was 20.9, which was less than the maximum value of 37.4 under a combined magnetic field. The experimental results show that the contrast of a magneto-optical image obtained under the combined magnetic field is greater than that obtained under the alternating magnetic field for all defects. The proposed combined magnetic field could optimise the magneto-optical imaging effect for weld defects under the existing excitation method to a certain extent.

Research on CNN-based intelligent recognition method for negative images of weld defects

The existing technology of automatic classification and recognition of welding negative images by computer is difficult to achieve a multiple classification defect recognition while maintaining a high recognition accuracy, and the developed automatic recognition model of negative image defect cannot meet the actual needs of the field. Therefore, the convolutional neural network (CNN)-based intelligent recognition algorithm for negative image of weld defects is proposed, and a B/S (Browser/Server) architecture of weld defect feature image database combined with CNN is established subsequently, which converted from the existing CNN by the migration learning method. It makes full use of the negative big data and simplifies the algorithm development process, so that the recognition algorithm has a better generalization ability and the training algorithm accuracy of 97.18% achieved after training. The results of the comparison experiments with traditional recognition algorithms show that the CNN-based intelligent recognition algorithm for defective weld negatives has an accuracy of 92.31% for dichotomous defects, which is significantly better than the traditional recognition algorithm, the established recognition algorithm effectively improving the recognition accuracy and achieving multi-category defect recognition. At the same time, the CNN-based defect recognition method was established by combining the image segmentation algorithm and the defect intelligent recognition algorithm, which was applied to the actual negative images in the field with good results, further verifying the feasibility of CNN-based intelligent recognition algorithm in the field of defect recognition of welding negative images.

Stability of Weld Pool and Elimination of Weld Defects in Aluminum Alloy Plasma Arc Keyhole Welding at Continuously Varying Positions

Mathematical statistics were used to study the stability of weld pool and the elimination of weld defects in aluminum alloy plasma arc keyhole welding at continuously varying positions. In the mathematical model, the mass transfer position and spatial welding position were taken as the input, and the shape of the welded joints (symmetry/deviation) was taken as the output. The results showed that the fitted curves of the front, back, and average deviations of the weld seam were all similar to the actual curves. According to the optimum results obtained in the experiment and the mathematical models, the mass transfer position only needs to be adjusted once (near to 30°) during the continuously varying positions, from vertical-up to horizontal welding. A breakthrough from fixed environmental variables to dynamic environmental variables in the process control of the keyhole weld pool was realized, which enabled the Al-alloy keyhole weld pool to resist the disturbance caused by gravity during variable position welding. The deviation of the welded joints of the whole plate was smaller than 0.5 mm, and the mechanical properties of the weld reached at least 85% compared to those of the base material, thus meeting the requirements of Al-alloy welding.