UBC Theses and Dissertations
Robust CT scanning of logs with feature-tailored voxels Angus, Edward J.
The greatest cost in sawmills is contained in uncut logs. Significant increases in profit stand to be made if logs are properly processed so that the amount of defect and feature-free products is maximized. This has driven research into internal sensing such as computed tomography (CT) for use in sawmills. Traditional CT systems from the medical industry are ill-suited for industrial use and provide distorted reconstructions when used with less than perfect or partially incomplete data. Industrial CT systems in areas such quality control have embraced so-called Algebraic Reconstruction Techniques (ART) for robustness to errant data, but these systems are difficult to provide reconstructions rapidly enough for sawmill use. A system is proposed here for an ART scanning arrangement specific for log-scanning use. The image space's voxel pattern is defined to reflect the geometry of a log's internal features. This greatly reduces the number of unknowns without a loss of information and allows for faster reconstruction. The geometry of the voxels makes exact calculation and storage of the series solution basis fast and practical for multi-slice scans. Data scaling and normalization eliminate unnecessary voxels in the reconstruction. It also makes the reconstruction scheme tolerant of rigid body motion radially and circumferentially about the cone beam source. The voxel numbering scheme means that knot features are contained in voxels that are numerically close for quick registration. A camera-based detector system was implemented to collect radiographs of logs. Logs were translated and rotated in an x-ray cone beam and their position was monitored by an optical encoder. The detector was activated at the proper intervals to yield radiographic data. The series solution basis for this helical movement was constructed. An iterative solver and selective low-pass filtering was found to provide good reconstruction results. Segmentation was implemented to demonstrate use of the reconstructions for lumber processing. Eccentric spiral motion tests demonstrated the effectiveness of the data scaling and normalization process. Results are provided that point towards efficient automated registration of features.
Item Citations and Data
Attribution-NonCommercial-NoDerivs 2.5 Canada