Configuration Design Optimization Method

This work presents a method (CDOM) for optimizing multiple system level assembly characteristics of complex mechanical assemblies by placement of their components. It addresses any component shape (including non-convex, hollow, sharp edges) connected together by functional constraints. This method finds multiple solutions to the Engineering Configuration Design Problem (ECDP) and proposes a simple cooperation scheme with the engineer using a Genetic Algorithm working on population of sets instead of population of individual points. In this paper four areas are addressed: • defining the assembly components and their relations (ECDP), • defining the Configuration Design Optimization Method (CDOM) • propose two strategies to improve the CDOM performance • show application of the CDOM to one academic and two engineering test cases INTRODUCTION Packing problems have been among the most exciting human pastimes for centuries. Puzzles, Tangram, Pentominos and other Polyminos reflect this excitement through Human history. Nowadays, these problems find applications in industry. Loading ships, trucks and trains, designing plants, electronic boards, placing mechanical components, cutting stocks, scheduling, are some of the numerous problems that can be classified as packing optimization problems. Among the geometrical packing problems, those involving spheres, squares and rectangles are the best known because they are the easiest to formulate and thus are the most studied. Unfortunately, even these simple cases are known to be difficult problems in the mathematical sense. So far, nobody has yet produced a global formal solution for them and even some mathematicians believe that this event will never happen (Lewis and Papadimitriou, 1978). Today's alternative is to accept a nearly optimal solution using computers and heuristics. These heuristics are case dependent and those guarantying good solutions are subject to combinatorial explosion. Moreover, engineering packing problems are more difficult to solve than basic packing problems since mechanical components are usually freeform and 3 dimensional. Common approaches involve different techniques such as dynamic programming, tree search or very specialized heuristics like finding optimal sub-configuration by assembling the objects in groups of two or three. In order to tackle the inherent complexity of simple packing problems, some nondeterministic methods are beginning to emerge. They involve the use of Simulated Annealing and Genetic Algorithms. Corcoran and Wainwright (1992) programmed a Genetic Algorithm to solve a bin packing problem using up to 500 packages and thus broke all the limits ever considered for such a problem. Wodziak (1994) and Grignon, Wodziak and Fadel (1996) used a mixed combinatorial and continuous formulation to address the truck-packing problem but their methods are dedicated to square like object shapes and thus are difficult to extend to engineering problems in which the objects are free form. Szykman and Cagan (1995, 1996) used a simulated annealing method to solve component layout problems. These non-deterministic approaches were enhanced by either local search (Hart, 1994), by specialized search heuristics (Yin and Cagan, 1998) or even combined with each other like in (Moscato, 1992). Eventually, Sachdev et al. (1998) proposed a framework of intelligent agents to integrate optimization techniques such as GA and SA to address layout problems. Free form packing refers to the packing of objects and containers having a non-regular shape. This case was not studied until recently because of several reasons. First, the