AI for AM in zirconia ceramic vat photopolymerization: A machine learning approach to stabilising slurry suspension and maximizing dimensional accuracy of the 3D print process

Abstract

The stability of ceramic suspensions and the dimensional accuracy of green bodies are critical factors influencing both printing process and the quality of the final component. Therefore, assessing dispersant agents and stability is essential during the suspension development stage. A novel power-law-based stability assessment approach was developed and employed for evaluating dispersant performance. The conventional stability testing method is based on visual monitoring and requires prolonged durations, whereas the newly introduced technique demonstrated significantly greater accuracy within a much shorter timeframe, which is in complete accordance with the conventional stability testing methods. Consequently, among six dispersants examined, DisperBYK-145 was selected based on the results of both evaluation methods. Additionally, the dimensional errors of green-body zirconia ceramics were analysed considering printing parameters and slurry additives. Optimal parameters were determined as 5.5 mW/cm² light intensity, 0.1 wt% photoinitiator, and 4 wt% white pigment (TiO₂) to minimise geometric errors. Given its significance, supervised machine learning algorithms were employed to predict dimensional errors based on the selected parameters. An artificial neural network model with 12 hidden nodes exhibited the best performance, aside from averaging-based ensembles. This study introduces a stability assessment method and machine learning models to predict the dimensional accuracy of zirconia ceramic suspensions in vat-photopolymerisation.