A Novel Spatiotemporal Statistical Quality Control Scheme Using 3D Point Cloud Data

This dissertation proposes a novel spatiotemporal monitoring system for three-dimensional data such as point clouds created from 3D laser scanners. The research is divided into three sections. First, a methodology is proposed for measuring the uncertainty of 3D laser scanners to facilitate the reporting of the uncertainty budget for these systems as they gain usage in industry. Factors impacting laser scanner uncertainty were varied in a designed experiment. Analysis of the results yielded a model that was used in a Monte Carlo simulation to estimate the laser scanner uncertainty interval for several geometric features.

Second, a three-dimensional spatiotemporal monitoring method was proposed to overcome the challenge of projecting a complex shaped surface onto two dimensions. The proposed method divides the point cloud into regions of interest (ROIs) and uses the ROI summary data to calculate a multivariate generalized likelihood ratio (MGLR) test statistic for inclusion on a 3D MGLR control chart. This 3D MGLR control chart identified small defects more quickly than prior methodologies.

Finally, two metaheuristic algorithms, simulated annealing (SA) and genetic algorithm (GA), were implemented to optimize the 3D MGLR control chart to minimize the time needed to identify process shifts. The 3D MGLR control chart has four parameters that can be set: three defining the number of ROIs and one defining the number of past observations used to calculate the test statistic. While the GA found significantly better results than SA, it took almost twice as many iterations and took longer to run.