AI for AM: machine learning approach to design the base binder formulation for vat-photopolymerisation 3D printing of zirconia ceramics
Fatih Tarak, Leah Okoruwa, Basar Ozkan, Farzaneh Sameni, Gerald Schaefer, Ehsan Sabet
Abstract
Additive manufacturing of ceramics, specifically through vat-photopolymerization, offers significant potential due to its high precision and ability to produce complex geometries. This study addresses the primary challenge in vat-photopolymerization: developing binder formulations that optimise both viscosity and mechanical properties while accommodating high ceramic loadings. This study introduces supervised machine learning (ML) algorithms as a novel approach to predict the viscosity and tensile strength of binder formulations. A comprehensive dataset was generated using a full factorial experimental design with three factors at three levels. Various ML algorithms were evaluated for their efficacy in regression applications. The leave-one-out cross-validation (LOOCV) method was employed to assess the performance of these ML algorithms due to the small dataset size. The ANN model with 5 hidden nodes delivered exceptional results, achieving mean absolute errors of 2.504 mPa·s for viscosity and 1.163 MPa for tensile strength, outperforming other ML models. ANN models were particularly adept at capturing the complex non-linear relationships. The inclusion of Okoruwa Maximum Saturation Potential (OMSP) area and peak position as input features significantly enhanced the predictive accuracy for both viscosity and mechanical properties. This research demonstrates the remarkable potential of ML algorithms to revolutionise the formulation process for VPP binder resins.