Novel strategies for tool path planning of WAAM process using reinforcement learning

This thesis presents an innovative method for generating tool path planning for the Wire Arc Additive Manufacturing (WAAM) process with reinforcement learning. The present state of the art provides numerous techniques to tackle this task. Most methods employ mathematical algorithms, while there are no instances of AI being used to completely design paths. The suggested solution seeks to train a reinforcement learning agent on diverse shapes, employing a custom reinforcement learning environment that simulates the WAAM process and a user-defined reward function that teaches the desired behavior of the path. Particular attention is given to the environment, which dictates the physical constraints that the agent is subjected to during its activity. Additionally, this text includes a summary of how 3D geometry is digitized into meshes and describes the voxelization technique used to discretize the WAAM process environment. Experiments have been performed to evaluate the effects of various parameters throughout the training phase. The generated paths have been evaluated for coverage and the quality of the thermal profile they generate. This work presents an in-depth analysis of the numerous phases involved in the development of an intelligent agent to address this complex task and it serves as a foundational point for subsequent improvements in this field of research. The objective of this study is not to enhance the existing state-of-the-art approaches but to explore the feasibility of employing reinforcement learning to tackle such challenges.