Experimental data driven ANN model for permeability prediction in TPMS lattices

Abstract

Permeability plays a critical role in nutrient transport and in tissue ingrowth. However, experimental measurement of permeability across multiple lattice designs and porosity levels requires significant time and controlled testing conditions. In this study, a data-driven framework is developed to predict the permeability of triply periodic minimal surface (TPMS) lattice structures using machine learning techniques. Five lattice types are considered, namely Gyroid, Schwarz-P, Schwarz-D, Lidinoid, and New lattice-GD. Each lattice is evaluated at four porosity levels: 70%, 75%, 80%, and 85%. Experimental permeability tests are performed using a custom flow-through setup at flow rates ranging from 50 to 650 ml min−1, with increments of 50 ml min−1. The experimental dataset is used to train regression-based machine learning models that establish relationships between lattice geometry, porosity, flow rate, and permeability coefficient. The expected outcome of this work is a predictive model capable of estimating permeability for new TPMS designs without the need for repeated physical experiments. The model is expected to achieve high predictive accuracy while maintaining consistency with Darcy’s law of flow. The study also aims to identify dominant geometric features influencing permeability across different lattice types and porosity levels.