Researchers have developed a hybrid artificial intelligence framework that forecasts the tensile strength and surface roughness of 3D-printed polylactic acid (PLA) parts, and then uses that model to pinpoint the printing settings that deliver the best combination of both properties. The study, published in Discover Applied Sciences, pairs a random vector functional link (RVFL) neural network with an optimization algorithm called Henry Gas Solubility Optimization (HGSO), and reports that the tuned hybrid outperformed both a standalone RVFL model and a conventional artificial neural network on held-out test data.
The team examined how five process parameters shape the mechanical characteristics of printed PLA specimens: build angle, infill angle, layer thickness, printing speed, and nozzle temperature.
Three modeling approaches were built and compared: a baseline RVFL network, the RVFL network tuned by HGSO, and an artificial neural network (ANN). The RVFL-HGSO hybrid proved the most accurate for predicting surface roughness, achieving a coefficient of determination (R2) of 98.95 percent on the test dataset, alongside a root mean square error of 0.777, a mean absolute percentage error of 1.002 percent, and a mean absolute error of 0.514.
The performance gaps between models were substantial. Compared with the standalone RVFL model, the hybrid raised R2 by 4.98 percent on the test dataset while cutting mean absolute percentage error by 49.93 percent, mean absolute error by 53.90 percent, and root mean square error by 29.68 percent. The improvements were even larger against the conventional ANN: R2 increased by 9.93 percent, while mean absolute percentage error fell by 69.02 percent, mean absolute error by 73.95 percent, and root mean square error by 58.49 percent.
These figures indicate that tuning the RVFL network with the HGSO algorithm helps the model capture intricate, nonlinear correlations between printing parameters and the resulting part properties — relationships that simpler or untuned models may miss.
Beyond prediction, the researchers carried out multi-objective optimization using particle swarm optimization (MOPSO) to find the printing configuration that simultaneously minimizes surface roughness and maximizes tensile strength. The best possible solution identified was a build angle of 44.16, a nozzle temperature of 220, an infill angle of 14.93, a layer thickness of 0.22, and a printing speed of 45.18. At this setting, the model predicted a minimum surface roughness of 2.53 and a maximum tensile strength of 44.15.
The optimization stage goes a step further by suggesting concrete parameter values rather than leaving users to search the parameter space by trial and error.
The authors position the RVFL-HGSO framework as a foundation for better product design and mechanical property optimization through process parameter adjustment. As with any data-driven modeling study, the reported accuracy applies to the dataset and parameter ranges examined, and the predicted optimum reflects model estimates rather than guaranteed physical outcomes.
The article was published open access on 29 October 2026, with contributions from researchers including Getachew Semegn, Firi Ziyad, Habtamu Alemayehu, Desalegn Wogaso, and Mohammedsani Mahamud Hibatu.
Subject of Research: Machine learning prediction and multi-objective optimization of mechanical properties in 3D-printed PLA parts
Article Title: Predictive modeling and multi-objective optimization of 3D-printed PLA parts using a hybrid random vector functional link (RVFL) and henry gas solubility optimization (HGSO) framework
Article References: Semegn, G., Ziyad, F., Alemayehu, H., Wogaso, D., & Hibatu, M. M. (2026). Predictive modeling and multi-objective optimization of 3D-printed PLA parts using a hybrid random vector functional link (RVFL) and henry gas solubility optimization (HGSO) framework. Discover Applied Sciences. https://doi.org/10.1007/s42452-026-09796-3
Image Credits: AI Generated
DOI: 10.1007/s42452-026-09796-3
Keywords: 3D printing, PLA, machine learning, RVFL, HGSO, tensile strength, surface roughness, multi-objective optimization, particle swarm optimization, process parameters, artificial neural network, additive manufacturing
News Source: Denise Maddox. (October 29, 2026). AI Model Predicts Strength and Surface Quality of 3D-Printed PLA Parts. Scienmag.
