Design and Evaluation of Measurements With Info-Gap Uncertainty

Measurement systems and data analysis provide quantitative geometrical information for quality assurance, manufacturing and assembly control, reverse engineering etc. Measurement and analysis systems must provide reliable performance at a level suited to the application. Good performance is better than poor performance, but the need for reliability must temper the aspiration for high performance. To achieve this balance we must model and manage our severely deflcient information about, and understanding of, the processes we confront. The central idea in this talk is that: Performance must be traded-ofi against robustness to uncertainty. This is relevant both for measurement system design and for data evaluation: † A measurement system which is designed to achieve optimal performance will be extremely sensitive to deflciencies in the knowledge upon which the design is based. Immunity to information-gaps is enhanced only by decreasing performance requirements. † An estimation algorithm which is constructed to provide high fldelity to measurements will be extremely sensitive to imperfections in the knowledge underlying the algorithm. Robustness to info-gaps is enhanced only by reducing the fldelity requirement. Throughout the talk we will emphasize the conceptual tools from Information-gap decision theory