Info-gap Decision Theory For Engineering Design. Or: Why 'Good' is Preferable to 'Best'

Designers face an ineluctable trade-off of system-performance against robustness to informationgaps in the designer’s knowledge base. A design which optimizes performance, based on the best available models and data, will have no immunity to deficiencies in those models and data. Immunity is obtained only by relinquishing aspiration for high performance. Info-gap uncertainty is ignorance or incomplete understanding of the systems and processes being optimized. This is broader than usually treated with probability theory. The strategy advocated here is one of robust-satisficing. The robustness function is the greatest horizon of info-gap uncertainty within which the performance is guaranteed to meet the aspirations. For fixed design: as the aspirations become more demanding, the immunity to uncertainty becomes lower and finally vanishes when maximal performance is demanded. Robustness can be recovered only by retreating from maximal aspiration. For fixed aspirations: one design is preferred over another if the first entails greater robustness than the latter. This search for robustifying designs is feasible (contains a non-empty search set) only when the aspiration is suboptimal. Sub-optimal designs can have greater robustness than performance-optimal designs, when evaluated at the same performance requirement. We consider four examples: designing the shape of a cantilever, maneuvering a dynamic system, identifying a system model, and supervising a go/no-go decision. One theorem is presented which establishes the theoretical foundations of this analysis.