A distributed design process and agent protocol for multidisciplinary optimization.

Abstract

Multidisciplinary optimization (MDO) problems are a specific class of concurrent engineering problems that involve distributed optimization of a design artifact. A large design problem is decomposed based on the disciplines required to solve the problem, and the disciplines interact to identify an optimal solution. This dissertation describes a distributed design process, Hierarchical Concurrent Engineering (HCE), for performing MDO. HCE involves sequentially achieving four design properties, leading to the identification of a feasible, desirable solution. The process can be repeated to improve the quality of the solution, ultimately identifying the optimal design. The process is implemented in the form of a protocol for a multi-agent system. The HCE design protocol regulates the interactions among design agents such that the agents follow the HCE process to identify a desirable, feasible solution. The protocol relies on constraint mediators to monitor and enforce dependencies among design-agent decisions. The design process and the multi-agent system draw upon the fields of constraint satisfaction, global optimization, and decision theory. Optimization performance is improved by applying several polynomial-time algorithms to eliminate decision alternatives that can never be part of a feasible, optimal solution. The benefits of HCE include: support for discrete ordered and unordered design variables, as well as continuous variables; (1) capability for top-down or peer-to-peer coordination strategies; and (2) ability to implement independent representation and reasoning mechanisms for each discipline. The HCE design process is applied to four case studies. The case studies present a range of requirements, including shared design variables, discrete and continuous design variables, constraint relationships defined by a family of curves or by table lookup, and top-down and peer-to-peer coordination strategies. For all four case studies, agents are created for each discipline involved, and the agents successfully identify optimal solutions. Performance analysis of the case studies shows that the efficient polynomial-time algorithms are effective in eliminating undesirable decision alternatives, especially for problems with discrete design variables. Traditional MDO approaches are not well suited for these problems, so HCE provides an important extension to the MDO field.