Flaw detection in aluminium die castings using simultaneous combination of multiple views

Recently, X-rays have been adopted as the principal non-destructive testing method to identify flaws within an object that are undetectable to the naked eye. Automatic inspection using radiographic images has been made possible by incorporating image processing techniques into the process. In a previous work, we proposed a framework to detect flaws in aluminium castings using multiple views. The process consisted of flaw segmentation, matching and finally tracking the flaws along the image sequence. While the previous approach required effective segmentation and matching algorithms, this investigation focuses on a new detection approach. The proposed method combines, simultaneously, information gathered from multiple views of the scene; this does not require searching for correspondences or matching. By gathering all the projections from a 3D point, obtained from a sliding box in the 3D space, we train a classifier to learn to detect simulated flaws using all the evidence available. This paper describes our proposed method and presents its performance record in flaw detections using various classifiers. Our approach yields promising results: 94% of true positives detected with 95% sensitivity in real flaws. We conclude that simultaneously combining information from different points of view is a robust approach to flaw identification.