Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
Applying a Multi-Armed Bandit (MAB) Problem-based Algorithm to optimize printing routes in GMAW Additive Manufacturing
Submission Author:
Americo Scotti , MG , Brazil
Co-Authors:
Rafael Pereira Ferreira
Presenter: Americo Scotti
doi://10.26678/ABCM.COBEM2023.COB2023-2392
Abstract
GMA Additive Manufacturing (GMA-AM) may face setbacks when using conventional trajectory strategies, such as incomplete or excessive material deposition and deviations in areas with frequent arc strikes and stops. To address these issues, Space-Filling strategies like the "Pixel" strategy have been developed, generating continuous trajectories that potentially minimise material accumulation and arc strikes and stops. However, computational efficiency remains a concern when dealing with complex geometries. This study proposed an upgraded version of the Pixel strategy, which incorporates reinforcement learning for trajectory planning. Grading up with the Multi-Armed Bandit problem solutions, the MAB-based-Pixel strategy aims at improving upon the existing Enhanced-Pixel strategy. Computational validation was performed. The results indicate that the MAB-based-Pixel algorithm achieves the shortest trajectory with fewer iterations than its predecessor, with the possibility to reduce the printing time. These findings establish the MAB-based-Pixel strategy as a promising solution for GMA-AM printing setbacks.
Keywords
3D printing, GMA-AM, path planning, Reinforcement Learning, Multi-Armed Bandit problem
