We present a new framework for learning novel operational strategies and dynamically controlling the layering process in metal additive manufacturing. Metal additive manufacturing technologies such as powder bed fusion (PBF) are generally constrained by a fixed action powder spreading process. At every layer, the print platform is lowered by a fixed amount, and the same recoating action is performed. Ideally this would lead to consistent layering and identical properties each time, but frequently process variability disrupts this procedure, leading to inconsistent layers. This can be mitigated by intelligently controlling the powder spreading process, which we achieve via a shift to digital methodologies that can reveal new process strategies and dynamically update the printer commands. We employ Bayesian optimisation as a method to build and train surrogate models for real-time control. We then demonstrate the utility of this Smart Recoating approach within an integrated simulation framework driven by realistic Discrete Element Method powder spreading simulations. Our results inform new strategies for controlling the recoater and print stage displacements, and demonstrate the potential of a digital twin control system to mitigate process variation and achieve consistent print quality in each layer.
Read full abstract