High precision post-processing framework for industrial computed tomography detection

This paper presents a high-precision parameterless post-processing framework for industrial computed tomography (ICT) detection, which does not need parameter tuning for workpieces with different slice image conditions. The proposed framework contains three main steps: (1) adaptively calculate appropriate local window size for each pixel in all slice images by using inner and outer fitting energies to reduce the negative influence of inhomogeneity in subsequent contour extraction; (2) extract contours in all slice images via genetic algorithm-based self-optimizing multi-method combination to maximumly eliminate side effects of inhomogeneity, artifacts, noise, and low contrast; (3) use multi-resolution transformation to avoid point cloud breakpoints and dense final 3D post-processing results. To demonstrate the effectiveness of the proposed framework, its 2D contour extraction effects are compared with state-of-the-art algorithms, and its 3D detection accuracy is certified by comparing with the corresponding coordinate measuring machine (CMM) point clouds or CAD models of different industrial workpieces. The experimental results show that the framework's performance of dealing with negative effects in 2D contour extraction is better than a single state-of-the-art method, and that the deviation distributions in industrial workpiece detections are ≤ 0.02 mm for the simple workpiece and ≤ 0.05 mm for the complex workpiece.