Welding Defect Inspection And Recognition Based On X-ray Imaging

Aimed at the special request for welding defect test in X-ray imaging, the paper studies the main techniques in the inspection and identification process.The defects, such as blowholes and crack, occasionally appear in the welding process, which can affect the quality and security of products. In one word defect test is very important in industrial testing. The traditional method is short of standardization and accuracy. Film method has low detection efficiency and complicated operation, which can not save the data and realize automation. The on-line inspection technology is objective, normative and standard .This technology can overcome misjudgment by manual effectively if used X-ray real time inspection system.According to detective theory of imaging system, this paper establish an high resolution and definition X-ray real time inspection system on welding defect, which can test and evaluate property of defect automatically, at last evaluate the level in weld.There is much redundant background information for defects detection in the image. Therefore this paper use an automatic method of weld area based on the auto-adapted threshold segmentation with subtraction technique in this paper. This method can reduce the computation, increase the precision and extract the defect accurately. The effect is good. And it can be finished easily.According to image analysis and the basis of judgment of Euclidean distance, feature parameters are selected to be used in defect recognition, including Flatness, Sharp degree of sharp part, Ratio of perimeter to area, Fill degree index, Symmetry, Relative gray-scale.At last the neural network is applied here to recognize the weld defects from radiographic image in this paper. By the experiment, it is concluded that only one hidden layer in the network is the most stable and get optimum combination of training function and transfer function using the method that theory is combined with practical experience. The experiment proves that this method making global estimates least which has high identification rate in weld image and short time of recognition.