SUMMARY

This thesis presents a process for digital twin part registration in a closed-loop machining system through 3D scanning technology and demonstrates its application in a novel approach to near-net-shape (NNS) voxel volume registration. Unlike methods commonly used in medical imaging and point clouds, which are focused on consolidating separate sets of data into a single, overlaid volume, the registration proposed minimizes the variance of selected distance field values. This allows for an optimization algorithm to drive a floating volume to be positioned inside a larger reference volume such that the inner/outer displacement between the two is as uniform as possible.
The proposed similarity metric for voxel-based models is demonstrated on a simple, uniform grid structure and is optimized through a genetic algorithm. Additionally, simulated annealing and particle swarm optimization are used for comparison as well as different schemes to increase accuracy. These optimization algorithms drive the computation of a 6 degree-of-freedom transform to fit a target part geometry within its NNS rough casting volume. Registration results from a test suite of parts are reported and examined. While a complete development of the ideas presented is beyond the scope of this research, the proposed method’s feasibility is validated by showing how toolpaths and ultimately g-code can be generated based off the relative positioning of the target part and NNS geometries.