Successful printing of highly filled suspensions (≥ 50 Vol.%) via vat photopolymerization (VP) depends on viscosity and filler optical properties. This study aims to introduce a machine learning (ML) based end-to-end pipeline for selecting material formulation and printing parameters conducive to successful VP printing of highly filled suspensions.
Predicting the printability of highly filled suspensions was framed as a classification-based ML task. A small training data set was sequentially generated through physical experiments by using a hybrid process that combines two-step landmark selection and intuitive sampling for a glass microparticle-based suspension. The trained ML models were then used to map a broad design space, identifying potential material formulations and processing conditions.
An artificial neural network with a multi-layer perceptron (MLP) architecture, featuring one hidden layer of 10 neurons, best captured the printability of the highly filled system. Coupled with probability estimation, the MLP identified areas in the design space with the highest chances of successful printing. The model identified numerous new formulations that would be extremely time-consuming to discover using experimental methods or physics-based models.
This study introduces the first instance of ML-based modeling for printing highly filled polymer suspensions via the VP process. Based on data from physical experiments, the model captures the real-time complexities of the printing process, paving the way for more efficient and effective printing of highly filled polymer suspensions.
