Propagating and Mitigating Uncertainty in the Design of Complex Multidisciplinary Systems

As humanity has developed increasingly ingenious and complicated systems, it has not been able to accurately predict the performance, development time, reliability, or cost of such systems. This inability to accurately predict parameters of interest in the design of complex multidisciplinary systems such as automobiles, aircraft, or spacecraft is due in great part to uncertainty. Uncertainty in complex multidisciplinary system design is currently mitigated through the use of heuristic margins. The use of these heuristic margins can result in a system being overdesigned during development or failing during operation. This thesis proposes a formal method to propagate and mitigate uncertainty in the design of complex multidisciplinary systems. Specifically, applying the proposed method produces a rigorous foundation for determining design margins. The method comprises five distinct steps: identifying tradable parameters; generating analysis models; classifying and addressing uncertainties; quantifying interaction uncertainty; and determining margins, analyzing the design, and trading parameters. The five steps of the proposed method are defined in detail. Margins are now a function of risk tolerance and are measured relative to mean expected system performance, not variations in design parameters measured relative to heuristic values. As an example, the proposed method is applied to the preliminary design of a spacecraft attitude determination and control system. In particular, the design of the attitude control system on the Mars Exploration Rover spacecraft cruise stage is used. Use of the proposed method for the example presented yields significant differences between the calculated design margins and the values assumed by the Mars Exploration Rover project. In addition to providing a formal and rigorous method for determining design margins, this thesis provides three other principal contributions. The first is an uncertainty taxonomy for use in the design of complex multidisciplinary systems with detailed definitions for each uncertainty type. The second is the modification of two simulation techniques, the mean value method and subset simulation, that can significantly reduce the computational burden in applying the proposed method. The third is a set of diverse application examples and various simulation techniques that demonstrate the generality and benefit of the proposed method.