Collaborative optimization of complex systems: a multidisciplinary approach

Abstract

Engineering design of complex systems is a decision making process that aims at choosing from among a set of options that implies an irrevocable allocation of resources. It is inherently a multidisciplinary and multiobjective process; nowadays, the designer has to face the continuous growing complexity of engineering problems, but also, the increasing economic competition that have led to a specialization and distribution of knowledge, expertise, tools and work sites. Products become more and more complex and their design is usually of large scale, characterized by an important number of design variables, parameters, requirements, constraints and objectives. Consequently, multiobjective optimization (MOO) and multidisciplinary design optimization (MDO) are more and more used to provide one optimal solution-by the use of a "a priori" or interactive preferences modelling-or a set of optimal solutions -in which the designer will have to choose "a posteriori" the one to be developed. The objective of this paper is to present an original and efficient Collaborative Optimization Strategy for Multi-Objective Systems (COSMOS) designed to find the sub-set of the design space that contains the best solutionsPareto frontier -of the global system.

Collaborative multi-objective optimization for distributed design of complex products

“…For instance, goal programming and linear physical programming (LPP) approaches [9,10] are based on MOCO and allow the speci cation of multiple objectives at both levels, although this requires priorities to be set for the objectives at subsystem level. Other approach that allows the speci cation of multiple objectives at both levels is known as COSMOS [12], where a multi-objective evolutionary algorithm (MOEA) is used to generate a population of solutions to identify the trade-o s of the system problem. However, this requires that an optimization task is conducted not just for each subproblem, but also at the system level.…”

“…,c K ) ⊺ , where each element is a weight, such that c k ≥ 0 ∀ k=1,...,K . The interaction between the subsystems is captured by the linking functions, l 1 and l 2 , which leads to the following interacting equations 12 …”

“…Subsystems 1 and 2 share a decision variable x 0 , and the interactions are represented by two linking variables, namely 21 and 12 . The linking variable, 12 , is an output of Subsystem 1 and an input for Subsystem 2. Hence, Subsystem 1 is able to generate values for 12 since it has access to the linking function, l 1 , while Subsystem 2 needs to predict the values of 12 .…”

“…The linking variable, 12 , is an output of Subsystem 1 and an input for Subsystem 2. Hence, Subsystem 1 is able to generate values for 12 since it has access to the linking function, l 1 , while Subsystem 2 needs to predict the values of 12 . Once Subsystem 1 completes its own optimization task, the generated subsystem solutions are shared with Subsystem 2.…”

“…Once Subsystem 1 completes its own optimization task, the generated subsystem solutions are shared with Subsystem 2. The shared solutions, contain the values of x 0 and corresponding 12 , and represent the set of trade-o s preferred by Subsystem 1. These values are then used by Subsystem 2 to construct a supervised learning model where x 0 and 12 are used as training data.…”