Single and multi-objective optimization of FDM-based additive manufacturing using metaheuristic algorithms

Abstract

The evaluation of additively manufactured components is often conducted by means of experiments assessing product quality, build time, dimensional accuracy and tolerances, production cost and tribological properties of parts. As it occurs to any other manufacturing process, the performance of additive manufacturing is strongly affected by its corresponding process parameters. This work examines the performance of different swarm-based evolutionary algorithms when it comes to single and multi-objective optimization problems related to additive manufacturing with emphasis to fused deposition modelling processes. Five problems adopted by the recent literature have been questioned regarding their number of independent variables and their predetermined optimization objectives. Two of these problems are of single objective, whilst three are of multi-objective optimization nature. The results obtained by the several independent executions of algorithms are compared by means of analogous indicators depending on the problem, i.e. convergence speed for single-objective problem, and quality of Pareto non-dominated solutions in the case of multi-objective optimization problems. The algorithms tested for single objective optimization, are the dragonfly algorithm (DA); the ant-lion algorithm (ALO); the grey-wolf algorithm (GWO); the moth-flame algorithm (MFO) and the wale optimization algorithm (WOA). For the multi-objective optimization problems, the multi-objective grey-wolf (MOGWO), the multi-objective ant-lion (MOALO), the multi-verse algorithm (MOMVO), the multi-objective dragonfly (MODA) the Pareto envelope-based selection algorithm (PESA-II) and the strength Pareto evolutionary algorithm (SPEA-II) have been tested. Even though all algorithms have been proven capable of providing optimal solutions to cope with volatile scenarios, the “No-Free Lunch” theorem has been validated supporting that algorithms do not perform the same when applied to different optimization problems.