Laser Powder Bed Fusion (LPBF) is a revolutionizing additive manufacturing (AM) that melts the powder particles to create innovative products, but it faces substantial challenges in monitoring and predicting the quality of the printed samples. Traditional density measurement methods, like the Archimedes technique, have been employed to determine the density of maraging steel samples produced under varying machine settings (MACH-S) process parameters, including laser power and scan speed. However, these machine-dependent parameters alone are insufficient for accurately predicting part density. To address this issue, an optical coaxial pyrometer in the AconityMINI system was utilized to capture thermal emissivity data, from which six statistical pyrometer-driven (PYRO-D) features, such as mean, median, and standard deviation, were extracted. The proposed feature-driven model then interprets machine settings and pyrometer data into physics-based (PHYS-B) features, including laser energy density, radiation pressure, intensity, temperature, and wavelength. With these three sets of input features, the study investigates the effectiveness of four supervised machine learning (ML) regression models: Random Forest (RF), K-Nearest Neighbor (KNN), Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP), for predicting the density of printed samples. The performance of these models was enhanced through three hyperparameter optimization (HPO) techniques: Random Search (RS), Grid Search (GS), and Bayes Search (BS), alongside a feature selection (FS) method to refine the input feature dimensions. The findings indicate that RS is the most effective HPO technique across all models and highlight the significance of MACH-S and PHYS-B features in improving model predictions over PYRO-D features, achieving an R2 score of 0.948 and a mean-squared error (MSE) of 0.007. Finally, the feature-driven model incorporating optimal machine, pyrometer, and physical features was utilized for density prediction through ML. Future research will aim to extend this approach to predict layer-wise part quality by incorporating a broader range of process parameters, focusing on understanding influencing factors such as porosity and surface roughness in AM printed parts.
