Toolpath design for additive manufacturing using deep reinforcement learning

Additive Manufacturing (AM) processes offer unique capabilities to build low-volume parts with complex geometries and fast prototyping from a variety of materials. Metal-based AM has become increasingly more popular over the last decade for manufacturing and repairing functional parts in automotive, medical and aerospace industries. Despite the great potential in metal-based AM market, the state-of-the-art practices involve rigorous trial and errors before achieving consistent parts with the desired geometric and material properties, which is mainly due to the sensitivity of the build on process parameters. While the influence of process parameters such as laser power, powder parameters, and scan speed on the microstructure and final properties of the AM build are extensively studied in the literature, the influence of toolpath strategies yet to be fully investigated. Authors in [Steuben et al., 2016] considered three different toolpath patterns for building a part using a fused deposition modeling process and demonstrated that the pattern has a significant effect on the ultimate strength and elastic modulus of the build. Akram et al. [Akram et al., 2018] formulated a microstructure model using a Cellular Automata (CA) and demonstrated a strong correlation between the toolpath pattern (i.e., unidirectional and bidirectional) and the grain orientations.