Solving a Real-World Structural Optimization Problem With a Distributed SMS-EMOA Algorithm

Abstract

This paper addresses a real-world optimization problem in civil engineering. It lies in the dimensioning of a 162m long bridge composed of 1584 bars so that both its weight and its deformation are to be minimized. Evaluating each possible configuration of the bridge takes several seconds and, as a consequence, running a metaheuristic for several thousands of evaluations would require many days on one single processor. Our approach has been to develop a distributed master/worker version of SMS-EMOA, an indicator-based multiobjective algorithm. By combining the Java implementation of the algorithm in jMetal with the Condor distributed scheduler, we have been able to use more than 350 cores to obtain accurate results in a reasonable amount of time.