Reinforcement learning-based closed-loop control system for adaptive process parameter optimisation in 3D printing

Additive manufacturing has major issues regarding the consistency in quality, depending on the type of materials used and the conditions of the processes. In this paper, the controlled methodology is a reinforcement learning (RL)-based closed-loop control framework of real-time adaptive optimisation of 3D printing parameters. We have an autonomous system that regulates the power of the laser, the speed of printing, and the deposition paths to optimise the quality measurements. Experimental validation of three technologies of AM (FDM, SLM, DIW) shows the enhancement of the quality measures by 35–55% and the efficiency by 18–28% in relation to conventional methods (p < 0.001, Cohen d > 2.0). The architecture combines hierarchical control architecture with real-time sensing, with 92% first-time-right success (compared to 75% baseline) and 60% setup time savings. Findings indicate that RL controllers are useful in managing nonlinear dynamics, minimising sim-to-real-adaptation, and autonomously optimising systems, which is a substantial step towards intelligent manufacturing systems.