A machine learning-based recommendation framework for material extrusion fabricated triply periodic minimal surface lattice structures

Abstract

Lattice structures (LS) have been utilized in various fields, from engineering to biomedical sciences. In the lattice structures, the triply periodic minimal surface (TPMS) LS attains benefits in terms of higher productivity and less material usage, a step towards greener 3D printing. However, no automated system exists that can effectively recommend LS parameters to reduce material waste, which is often neglected in traditional methods. To overcome these challenges, this study presents a machine learning (ML) and Deep Learning (DL) based framework recommending TPMS LS according to specific requirements. Initially, a dataset of 144 samples was created using the Material Extrusion (ME) technique. The four TPMS LS were chosen (Split-P, Gyroid, Diamond, and Schwarz) and manufactured with Polylactic acid (PLA). This dataset was used to train both ML and DL algorithms. ML algorithms included Bayesian regression (BR), K-nearest neighbors (KNN), Random Forest (RF), Decision Tree (DT), and DL algorithm convolutional neural network (CNN). These models were used to predict the key parameters of TPMS LS, including wall thickness, unit cell type, loading direction, and unit cell size. Extensive testing was performed to evaluate the performance of the algorithms, employing R-squared values and root mean square error (RMSE). The results showed that the machine learning models, specifically the RF and DT algorithms, performed the best, achieving R-squared scores of 0.993 and 1.0 and RMSE scores of 0.1180 and 0.0795, respectively. The deep learning model, CNN, achieved an RMSE value of 0.46 and an R-squared score of 97%. This study not only contributes to a better understanding of automated TPMS lattice structures but also has significant implications for sustainable design and innovation, particularly in enhancing efficient and environmentally friendly 3D printing technologies.