Constraint Management Methodology for Conceptual Design Tradeoff Studies

This paper presents a constraint management methodology, which facilitates tradeoff studies during conceptual design. This approach represents design models as constraints between variables, and uses the resulting constraint network to automatically derive computational procedures for performing userspecified tradeoff studies. By decomposing large constraint networks into smaller pieces that can be solved robustly, this approach can solve extremely large systems of non-linear equations present in practical system models. Design Sheet is a software implementation of this methodology; it allows the designer to interactively develop models, flexibly define tradeoff studies, and quickly explore large areas of design space to study how the different performance and cost criteria tradeoff with respect to one another. Design Sheet has been used on practical applications ranging from the system-level design of spacecraft using combined performance and cost models to the preliminary design of automotive bearings. This paper demonstrates the unique capabilities of Design Sheet in performing design tradeoff studies, using a thermal imaging system performance model developed for the DARPA MADE program. INTRODUCTION The design of complex systems requires assessing a wide range of alternative designs, so as to determine one that best meets competing requirements from multiple perspectives, such as performance, manufacturability, reliability and cost. As detailed evaluation of large numbers of design alternatives is not practical because of resource constraints, a hierarchical approach is used in deciding the design specifications. During the early conceptual phases of this process, decisions are made about which configurations are better than which others. During the detailed phases, when the major configuration options and design choices have already been made, the task is to determine the optimal values for all the detailed design parameters. In current industry practice, detailed simulation models and optimization algorithms are widely used for this purpose. Though design decisions made during the early stages have a far greater impact on the final design quality and cost, few decision support tools are available for conceptual design. During conceptual design, large areas of the design space are explored and candidate designs evaluated with respect to multiple criteria. Which tradeoffs are important at any stage during design, however, are not known a priori and depend on the results of other tradeoff analyses. Additionally, as cost is often a crucial criterion in system design, multiple attribute tradeoffs with combined performance and cost models are essential. Furthermore, models used for conceptual design tradeoff analyses are often based on incomplete and quickly evolving system descriptions. Therefore, the critical capabilities for a conceptual design system are flexibility in defining and performing tradeoff studies, ease of integrating models from multiple perspectives, and support for quick development and incremental modification of models. Several research efforts are aimed at developing collaborative design environments that integrate multiple models and tools used by engineers (e.g., Herman and Lu, 1992; Cutkosky, 1993; McGuire et al., 1993; Sriram and Logcher, 1993). However, these environments are only appropriate for the detailed stages of design. Conceptual design research, on the other hand, has mainly focused on the role of optimization and robust design (e.g., Choy and Agogino, 1992; Dixon, Orelup and Welch, 1993), without any emphasis on how to use integrated models. Currently, conceptual design decisions are often made without recourse to integrated models. In some cases, spreadsheets and specialized FORTRAN programs are used, but these