Performance evaluation of machine learning techniques in surface roughness prediction for 3D printed micro-lattice structures

This study focuses on predicting the surface roughness of additively manufactured A286 steel micro-lattice structures using machine learning techniques, evaluating various models for their effectiveness in enhancing surface quality control. Machine learning models are utilized to assess the effects of post-processing techniques, including stress relieving and heat treatment, on the surface roughness and micro-Vickers hardness. The micro-lattices, produced through 3D printing technology, are evaluated for lattice volume, surface roughness, and hardness in three conditions: as-printed, stress-relieved, and heat-treated. Several machine learning algorithms, including Bayesian Ridge Regression, K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Random Forest, AdaBoost, XGBoost, and an Artificial Neural Network (ANN), are applied to predict surface roughness based on lattice parameters and post-processing conditions. The models are optimized using cross- validation and hyperparameter tuning, with performance evaluated through metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R²). Among the models, Random Forest demonstrated robust predictive performance with an R² of 0.8823 on the training set and 0.7459 on the test set. The ANN model achieved an overall R² of 0.805, with a training R² of 0.99, showcasing its capability to capture complex nonlinear relationships within the data while suggesting a need for further optimization to enhance generalization. A comparison of machine learning predictions with experimental results highlights the impact of post-processing on surface roughness and hardness. This work enhances control over additive manufacturing processes by providing a predictive approach for optimizing the surface characteristics of 3D-printed metallic micro-lattices.