What question did this study set out to answer?

The research aims to predict the relative density of L-PBF parts using machine learning models based on a comprehensive dataset.

June 26, 2026Open Access

Modeling and prediction of 316 L stainless steel relative density in L-PBF process using machine learning

Key Points

The research aims to predict the relative density of L-PBF parts using machine learning models based on a comprehensive dataset.
Developed six machine learning models, including KNN, AdaBoost-DT, and four MLPs optimized with various techniques.
Used input parameters: laser power, scanning speed, and hatch spacing from a dataset of 287 measurements.
Evaluated model accuracy with metrics such as R², RMSE, and average percent relative error.
The hybrid modeling framework significantly improved prediction accuracy compared to individual models.
Model accuracy metrics indicated strong performance despite dataset heterogeneity and outliers.
Models effectively captured the nonlinear relationships between process variables and relative density.

Abstract

This study investigates the prediction of relative density in L‑PBF parts using machine learning models trained on a literature-based dataset comprising 287 experimental measurements collected from previously published studies. The input parameters for the proposed models included laser power, scanning speed, and hatch spacing. To model the nonlinear relationship between process variables and relative density, k-nearest neighbours (KNN), adaptive boosting decision trees (AdaBoost-DT), and four multilayer perceptron (MLP) models optimised using Scaled Conjugate Gradient (SCG), Levenberg-Marquardt (LM), Bayesian Regularisation (BR), and Resilient Backpropagation (RB) were developed. The predictions from these six models were subsequently integrated into a single framework using a committee machine intelligence system (CMIS), which provided improved predictive performance. Model accuracy was evaluated using statistical metrics including the coefficient of determination (R²), standard deviation (SD), average percent relative error (APRE), average absolute percent relative error (AAPRE), and root mean square error (RMSE). The results showed that the proposed hybrid modelling framework can effectively capture the complex relationship between L‑PBF process parameters and relative density. However, the heterogeneity of the literature-based dataset and the presence of a limited number of outliers may introduce uncertainty into the predictions.

AIに質問

Bookmark

View Full Paper