Application of Bayesian Optimization for the optimized development of slurries for Additive Manufacturing

Additive manufacturing (AM) by vat photopolymerisation (VPP) enables the flexible production of ceramic and metallic components through a layer-by-layer build-up followed by debinding and sintering. In this process, slurries consisting of a photosensitive binder system and ceramic or metallic powder are cured locally by selective application of light. In order for the slurry to be processed in VPP systems, its viscosity must be sufficiently low. At the same time, the slurries must harden well for shaping. Both properties are generally worsened by a higher ceramic or metallic filler content. However, in order to accelerate the debinding and sintering process and to increase the process reliability, the aim is to maximize the filler content. This conflict of objectives requires a precise adjustment of the large amount of slurry constituents. Hence, an experimental slurry development and optimization is very expensive and time-consuming. Therefore, the application of artificial intelligence (AI) seemed to be a promising approach. In this work, Bayesian optimization was used to iteratively optimize the slurry composition. Using this approach for ceramic slurries, it was possible to achieve in less than 40 optimization steps an aluminum oxide (Al2O3) slurry suitable for vat photopolymerisation with a volume fraction of 65 % ceramic powder, the highest currently known fraction for Al2O3 in VPP slurries