Related Experiment Video
Updated: May 28, 2026
09:17
Surrogate Model Development for Digital Experiments in Welding
Published on: March 28, 2025
A Finite Element Simulation-Informed Machine Learning Framework for Screening Average Thermal Stress Responses in
1Department of Materials Science and Engineering, Anhui University of Science and Technology, Huainan 232001, China.
Materials (Basel, Switzerland)
|May 27, 2026
Summary
This study introduces a machine learning framework to efficiently screen process parameters for 316L stainless steel in selective laser melting (SLM). The model identifies optimal laser power and preheating temperatures for reduced thermal stress and improved part quality.
Area of Science:
- Materials Science
- Additive Manufacturing
- Computational Modeling
Background:
- Selective laser melting (SLM) requires efficient process parameter screening for optimal material properties.
- 316L stainless steel is a widely used material in SLM, but its processing window is complex.
- Traditional screening methods are time-consuming and resource-intensive.
Purpose of the Study:
- To develop a finite element simulation-driven machine learning framework for efficient comparative process-window screening in SLM.
- To identify key process parameters influencing thermal stress in 316L stainless steel during SLM.
- To establish a surrogate model for ranking SLM process conditions.
Main Methods:
- Generated a simulation dataset using ANSYS covering laser power (LP), scanning speed (SS), heat-source diameter (HSD), and substrate preheating temperature (SPH).
- Trained nine regression models, selecting the Backpropagation Neural Network (BPNN) for its predictive accuracy and stability.
- Utilized Shapley additive explanations (SHAP) and partial dependence plots (PDPs) to analyze parameter influence.
Main Results:
- The BPNN model accurately predicted simulated average thermal stress (σavg).
- Laser power (LP) was identified as the dominant variable affecting σavg, followed by substrate preheating temperature (SPH).
- Optimal conditions near 180-200 W showed lower predicted σavg, with experimental results showing improved forming quality and stress trends.
Conclusions:
- The proposed finite element simulation-driven machine learning framework serves as an efficient surrogate tool for comparative parameter screening in SLM.
- The framework aids in identifying optimal processing parameters for 316L stainless steel, reducing thermal stress and enhancing part quality.
- The study validates the framework's utility within its defined assumptions and parameter range, paving the way for faster optimization in additive manufacturing.
