Optimizing resin shrinkage in additive manufacturing through supervised pathways: Leveraging machine learning in dentistry

This work aims to tackle the issue of poor dimensional stability in 3D printed resin dental devices, caused by polymerization shrinkage (PS). By optimizing the additive manufacturing (AM) process, the study seeks to improve the precision and fit of these devices, enhancing patient-specific solutions and overall comfort in dentistry. This study investigates the effects of five influencing process parameters layer thickness, exposure duration, print angle, infill density, and lift speed on PS in resin-based 3D printing. Using Response Surface Methodology (RSM), 32 optimal parametric combinations are generated corresponding to which PS is calculated. Three machine learning (ML) models are developed: an Artificial Neural Network (ANN) using Levenberg-Marquardt (lm) algorithm, another with scaled conjugate gradient (scg), and third with Support Vector Machine (SVM) using Radial Basis Function (RBF) to predict PS. The model’s effectiveness is enhanced through ML hyperparameter tuning and additional validation of selected model is performed using k-fold and bootstrapping method, making least RMSE and maximum R 2 value for ANN-lm at learning rate(lr) = 0.01, momentum rate(mr) = 0.3, with 5-1-10 architecture as the selected model characteristics. Later, model is optimized using Particle swarm optimization (PSO), ACO (Ant colony optimization) and Artificial Bee colony (ABC) for minimizing PS, where results states ANN-PSO converges to most optimal solution. Environmental factors are not analyzed, leaving room for further research. Also, a controlled infill design method is used in order to reduce resin entrapment and to retain structural integrity. This study makes an attempt to provide a novel, data-driven framework for enhancing the dimensional accuracy of 3D-printed dental devices, bridging a critical gap between process parameters and final product quality. It also offers a comparative analysis of several ML-based modeling and optimization methods which in turn fill available gap in research and contribute to the understanding of the influence of these techniques on the print accuracy.