Study And Application Of Automatic Recognition Of Defects Based On Radiography

In order to insure the safety of work pieces, especially those parts relevant to vehicle safety like railway casts, wheel rims, a control of their quality is required. With the unique advantages, radial detection technology is widely used in the non-destructive testing for the internal defects of industrial products. However, the evaluation of testing results still relies on artificial assessment. This method is heavy workload, low efficiency, and the results of evaluation are influenced by the subjective factors of staffs. And it is difficult to meet the needs of the modern production. So in order to improve efficiency and reduce human-induced undetected miscarriage of justice, and to make the results of evaluation objective, standardized and scientific, the research on automatic defects identification is valuable.Defects of work pieces generally fall into two categories: one category is assembly errors, such as whether the key parts are left out, or whether the installation location is correct etc; the other is the material internal defects of work pieces such as pores, cracks, includes or shrinkages. This article is mainly against the research on the identification of the second type of defects in X-ray images of casting, and the ultimate goal is to develop the automatic defects identification and classification software based on the X-ray images of casting to implement the recognition of the four typical types of defects (pores, cracks, includes shrinkages), with the focus on the identification of pores. Here X-ray images refer to two types, that is, CT, DR images.In this paper, firstly, the fractal characters and the scale-free range of X-ray images of work pieces are analyzed. Secondly, focusing on the identification of defects, the implements of image preprocessing, defects segmentation, defects classification are given. In the image preprocessing part, a method to remove horizontal or vertical strip-artifacts in DR images is proposed. And using the method, great results are acquired for the strip-artifacts in DR images of work pieces with the same thickness. In the defects segmentation part, a method based on the box fractal dimension and the box pixel-points is investigated to locate defects in CT or DR images, and then the traditional chain code tracking method is improved. And the exact defects location and the precise defects tracking are implemented. In the defects classification part, five gray statistical characteristics are selected and are used in the binary tree classifier to classify four different types of defects. Lastly, a set of automatic defects identification and classification software is developed based on X-ray images of casting. The results show that the software can achieve the recognition rate of 90% for the defects whose area bigger than 3×3.