CT-based reverse engineering for precision inspection and shape analysis

CT-based reverse engineering is the process of extracting geometrical and structural information from the computed tomograph (CT) images of an existing workpiece. This dissertation focuses on problems of precision inspection and of three-dimensional (3-D) shape analysis for existing workpieces using their CT images.; Precision inspection is the process of extracting accurate dimension quantitative information of a workpiece. In this dissertation, a precision inspection system is developed for turbine blade dimension measurement using LMS methods to enhance accuracy prediction with adaptive property of increasing accuracy when additional verifiable data are available. A directional-derivative-based edge detector using facet model is applied first on CT images to extract edges with subpixel accuracy. Measurements of interested regions of the turbine blades are then computed and compared to the corresponding optical measurements. Templates are designed to automate this process. The system accuracy is within 0.003inch (0.0762mm). Two predictors are designed to improve the accuracy of unverified measurements based on the verified measurements. Experiments show increasing accuracy of the adjusted measurements as additional verified measurements are available. Quantitative analysis of a trapezoidal-shape workpiece is also performed. The results indicate that the CT system performance is affected by the structure and size of a workpiece.; Workpiece shape analysis is the process of extracting structural and topological information of a workpiece from its CT images. In this dissertation, an interactive shape analysis system is developed to detect surface points from a sequence of CT images, extract 3-D shapes from surface points, and construct a relational graph to represent the structural and topological relations between the extracted shapes. The directional-derivative-based edge detector using facet model is extended to the 3-D case as a surface detector. An interactive algorithm is developed to extract 3-D shapes by cross-sectioning the surface data and linking cross-sectional contours into 3-D shapes. Geometric features of each shape are computed and used to construct a view-independent hierarchical relational graph. The system was tested with real workpiece CT images. The output of the system can be used to facilitate methods for workpiece recognition, workpiece inspection, and high accuracy model construction.