Optimizing Spacecraft Design - Optimization Engine Development: Progress and Plans Steven L. Cornford, Martin S. Feather, Julia R Dunphy, Jose Salcedo Jet Propulsion Laboratory California Institute of Technology 4800 Oak Grove Dr Pasadena, CA 9 1 109 { Steven.L.Cornford,, Martin.S.Feather, Julia.RDunphy, Jose. Salcedo}@Jpl.Nasa.Gov 818-354-1701,818-354-1194,818-393-5365, 81 8-393-0995 Tim Menzies Lane Department of Computer Science & Electrical Engineering, West Virginia University PO Box 6109, Morgantown, WV 26506-6109 tim@enzies.com 304-367-8263 Abstract- At JPL and NASA, a process has been developed to perform life cycle risk management. This process requires users to identify: goals and objectives to be achieved (and their relative priorities), the various risks to achieving those goals and objectives, and options for rislc mitigation (prevention, detection ahead of time, and alleviation). Risks are broadly defmed to include the risk of failing to design a system with adequate performance, compatibility and robustness in addition to more traditional implementation and operational risks. The options for mitigating these different kinds of risks can include architectural and design choices, technology plans and technology back-up options, test-bed and simulation options, engineering models and hardware/software development techniques and other more traditional risk reduction techniques. Each of these risk mitigations has resource costs associated with them. The sum of all these mitigations is almost always unaffordable. Furthermore, fhere may be a variety of other constraints (mass, power, h d i n g profile, leveraged programs, etc.) that fkther constrain acceptable selections. The challenge is therefore to emerge with an “optimal” selection of mitigations that makes best use of available resources to reduce risk to the fkllest extent possible. For non-trivial design spaces, the search space of possible selections is huge. This precludes exhaustive search for the optimum, and therefore necessitates the adoption of heuristic search techniques. At JPL, we have explored application of several heuristic techniques for searching for, and refining, collections of risk mitigations, notably genetic algorithms, simulated annealing, and machine learning. The results of research and pilot applications of these techniques for finding best combinations of life cycle risk management solutions are discussed. TABLE OF CONTENTS .................................................................... 1. DDP’S RISK MANAGEMENT PROCESS .. 1 2. OPTIMIZATION NEEDS ......................... 2 3. GENETIC ALGORITHM EXPERIMENTS ... 3 4. MACHP~~LEARNINGEXPERIMENTS ..... 4 6. STATUS AND FUTURE WORK ................. 8 ACKNOWLEDGEMENTS .............................. 8 B~OGRAPHIES ............................................ 9 5. SIMULATED ANNEALING ;EXPERIMENTS 7 REFERENCES ............................................. 8 1. DDP’S RISK MANAGEMENT PROCESS This section summarizes the risk management process that we have developed and applied at JPL and NASA. Defect Detection and Prevention (DDP) is the risk management process that we have developed and applied to risk assessment, risk mitigation planning, and lifecycle risk management [l]. The primary purpose of DDP is to help expert users plan the design and development of complex systems. Risk management is central to their successful development, deployment and operation. Custom tool support [2] fiicilitates the practical application of the DDP process. DDP explicitly represents risks, the objectives that risks threaten, and the mitigations available for risk reduction. By linking these three concepts, DDP is able to represent and reason about the cost-effectiveness of risk reduction 1