Probabilistic engineering design optimization: Application to spacecraft and navigation systems.

摘要

In addition to designing engineering systems for optimal performance, developing systems that are robust to possible failures and/or non-ideal operating conditions is of great importance. In modern engineering practice, this is normally accomplished by incorporating conservative performance margins such that projected "worst-case" outcomes can be accommodated. In aerospace, safety or robustness considerations often dominate the design process, but this can lead to over-designed systems, lengthy development programs, and expensive final products.; Provided that decision makers accept the unavoidable possibility of failure, a superior approach based on system uncertainty and user utility modeling exists. System performance uncertainty, including unknown parameters and possible unit failures, is modeled using the best available information. The user's utility function, of arbitrary mathematical form, expresses the relative "goodness" of all possible outcomes. Once the axioms of decision theory are met, a maximum-utility search among the design space determines the optimal solution. Because these problems do not conform to standard mathematical assumptions, there is no guarantee of finding the best possible answer, but modern computer search techniques now provide the capability to converge toward the global optimum in reasonable time. As with traditional systems engineering, the optimal-decision process is iterative, since the computer search results are reviewed by designers who can further develop their risk and utility models.; This approach has been successfully applied to several system design tasks in this thesis. A tutorial aircraft landing control problem is used to demonstrate the basic procedure. New models for spacecraft reliability prediction have been combined with mission utility functions to predict the overall mission reliability of the Gravity Probe-B (GP-B) spacecraft and to find improved redundancy architectures. Uncertainty-based optimization has also been demonstrated to significantly improve the process of Receiver Autonomous Integrity Monitoring (RAIM) for Global Positioning System (GPS) navigation users. Similar uncertainty models applied to augmented Differential GPS (DGPS) systems can predict overall performance and integrity for large regions of users. Combining these with top-level objective models allows augmented GPS architectures to be optimized iteratively, as the latest experimental data updates the risk model and motivates additional system improvements.