Printability prediction of food formulations for 3D printing using a Gaussian process regression model

Abstract

The aim of the study was to address the research gap in printability prediction for nutritious formulations in Food 3D Printing. To this end, a predictive model to reduce trial-and-error operations was developed for formulation optimization. Printability characterization assessments were employed based on 2D and 3D tests to forecast the printability of formulations using a machine learning (ML) strategy. Starch concentration and printing temperature of formulations were used as predictors of printability. The predictive model was developed based on different feature extraction methods combined with Gaussian Process Regression (GPR) algorithms. In addition, a complementary laboratory validation was performed to determine a comparative percentage error between the GPR model predictions and the experimental measurements from an additional dataset. The GPR model was able to predict printability, reaching a RMSE value between 0.013 and 0.48%. Which means the model fit the data with accuracy, providing a reliable tool for formulation optimization. The laboratory validation demonstrated close values to those obtained by the model, further confirming its effectiveness. The comparative percentage errors from the laboratory validation varied between assessments and formulations, with percentage errors as low as 0.01%. The model, with its ability to predict the printability of formulations with different starch compositions and printing temperatures, could be a valuable tool in the food industry. It could assist in examining both the quality of 3D-printed structures and the adaptation of these formulations and printers for producing food products with customized sensory and nutritional profiles, opening up new possibilities for food production and customization.