Abstract

From a certain point of view, parametric engineering design may be considered as an optimization problem. The design problem may be represented through a set of design parameters. The optimal solution is located by using a set of competing design parameters and its evaluation is based upon specific criteria. A significant number of techniques and methodologies have been proposed in order to perform this difficult task. The selection of the appropriate one(s) depends strongly upon the nature and the specific characteristics of the design problem under consideration. The majority of these techniques and methodologies rely on the definition of some initial conditions. Wrong, misleading or incomplete initial conditions may result to solutions characterized by local optimality or may need excessive computational time in order to converge to either an optimal or a sub-optimal solution. In the context of the current work, two different approaches are used for initializing the optimization process: genetic algorithms and pattern search. Genetic algorithms need an initial population of individual solutions before the genetic operations could be deployed, while the pattern search techniques use a starting (initial) point for the optimization process. These two initial conditions (initial population and initial point) may be defined either randomly or deliberately. The present paper introduces a case-based design (CBD) module as pre-processor to the design optimization. This CBD module is based on an artificial competitive neural network, which is submitted to unsupervised learning by examples based on past design solutions. The new design is represented through fuzzy preferences and weighting factors, which are compiled by the neural network for retrieving similar past solutions. The retrieved solutions are used in order to determine the initial conditions of the optimization method (the initial population for the genetic algorithm (GA) or the starting point for the pattern search). The optimal solution is then searched using the criterion of the maximum aggregated overall preference. A system, namely Case-DeSC, has been developed in the purpose of evaluating the proposed framework in the application area of parametric design of oscillating conveyors. The results show that the proposed optimization methods converge faster to more efficient solutions if case-based reasoning (CBR) is utilized for defining the initial optimization conditions.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call