Integrating explainable artificial intelligence in machine learning models to enhance the interpretation of elastic behaviors in three-dimensional-printed triangular lattice plates
Itthidet Thawon, Pana Suttakul, Ramnarong Wanison, Yuttana Mona, Korrakot Yaibuathet Tippayawong, Nakorn Tippayawong
Abstract
Three-dimensional printing provides significant manufacturing flexibility, facilitating the creation of intricate geometries. This process commonly involves fabricating porous products constructed from lattice structures comprising numerous unit cells. These lattice structures have gained popularity for their capacity to reduce weight and material consumption while preserving specific mechanical strength. However, optimizing their elastic properties requires a thorough understanding of how unit cell design influences the overall performance of these structures. This study introduces a novel approach combining finite element simulations with machine learning (ML) to predict and analyze the elastic moduli of two-dimensional triangular lattice structures. Using the Extreme Gradient Boosting algorithm, an ML model was developed to predict moduli based on simulation data, aiming to establish a precise relationship between relative density and elastic moduli. A key objective is to provide an efficient estimation tool for researchers and engineers, verified through comparisons with existing closed-form solutions and experimental data. To enhance trust in predictions, Explainable Artificial Intelligence techniques, including SHapley Additive exPlanations and Partial Dependence Plots, were employed to interpret the model's outputs and identify influential design variables. Findings revealed a non-linear relationship between relative density and elastic moduli, with ML models achieving high predictive accuracy. This approach offers valuable insights for optimizing lattice structures across diverse applications, emphasizing the critical design features influencing their elastic behavior.