Research Of X-ray Digital Radiography To Automated Inspection Of Interior Assembly Structures Of Complex Products

Considering at the issue of inspection of interior assembly structures of complicated products, the paper synthetically studies technologies of X-ray digital radiography, image processing, pattern recognition and computer vision for 3D objects recognition to accomplish automated inspection of interior assembly structures of complex products.Based on Nyquist Sampling Theorem, the paper analyses the digitizing fundament of orientation structural information of 3D objects, studies one sampling method of spatial structures with whole-orientation image sequence and proposes one sampling rule that limited by the least structural size and the inherent resolution of the system to minimize the number of images in the sequence to speed up the recognition process. Furthermore, the paper studies the inherent relativities among the image sequence to achieve fast recognizing assembly constructions inside products with one or several images obtained in some orientations. The paper develops one scheme of improved X-ray DR system that used multiple view technique from a static camera and the issue of blind areas is resolved, in witch the optimizing principles of improving the spatial resolution, sensitivity and utilization ration of the X-ray energy are considered.Preprocessing of X-ray image is one of the most important processes to image pattern recognition. The paper adopts one method that piles up several images to remove yawps and extrude the distributing characteristics of the image in vertical and horizontal directions. During the process of segmenting objects, the paper presents an effective method based on the combination of 3-point in hardware and local area projection in software to accurately locate the products quickly.Feature extraction is one of the most basic problems in pattern recognition. For image recognition tasks, extracting the effective image features is a crucial step. As the most effective feature extraction approaches, subspace methods have received extensive attention owing to their appealing properties. The essence of subspace methods is to reduce a high-dimensional original sample space to a low-dimensional feature subspace that is benefit to classification. The paper firstly proposes principal component analysis entropy of hierarchical hybrid projection function of image as its recognition feature. Then using the whole-orientation features of a standard sample of the products and extracting its maximal system of linear independence as the feature basis for recognition. Finally, the paper utilizes one modified BP neural network to accomplish fast recognition.A parallel operation flow of the whole system is designed according to the engineering requirement and an automated recognition software is exploited under Visual C++ environment. The first domestic multi-view X-ray DR inspection system is developed that has been successfully applied to online automated inspecting of axial products containing complicated assembly structures. The vast experimental results show that wrong inspection ratio is less than 4.5% while without missing inspection and the average recognition takes less than 5s in inspecting progress. The system possesses self-study ability to new products and the function of inquiring database.