[Paper Review] A Physics-Informed Machine Learning Model for Porosity Analysis in Laser Powder Bed Fusion Additive Manufacturing

Rui Liu, Sen Liu+1 more|arXiv (Cornell University)|Jan 13, 2021

Additive Manufacturing Materials and Processes45 references4 citations

TL;DR

This paper proposes a physics-informed machine learning (PIM) model that translates laser powder bed fusion (LPBF) machine settings into physical effects—such as laser energy density and radiation pressure—instead of relying on machine-specific parameters. By using these machine-independent physical effects to predict porosity levels (classified as pass, flag, or fail), the model achieves 10–26% prediction error across six learning methods, demonstrating generalizability and improved interpretability for porosity control in Inconel 718 parts on a Concept Laser M2 system.

ABSTRACT

To control part quality, it is critical to analyze pore generation mechanisms, laying theoretical foundation for future porosity control. Current porosity analysis models use machine setting parameters, such as laser angle and part pose. However, these setting-based models are machine dependent, hence they often do not transfer to analysis of porosity for a different machine. To address the first problem, a physics-informed, data-driven model (PIM), which instead of directly using machine setting parameters to predict porosity levels of printed parts, it first interprets machine settings into physical effects, such as laser energy density and laser radiation pressure. Then, these physical, machine independent effects are used to predict porosity levels according to pass, flag, fail categories instead of focusing on quantitative pore size prediction. With six learning methods evaluation, PIM proved to achieve good performances with prediction error of 10$\sim$26%. Finally, pore-encouraging influence and pore-suppressing influence were analyzed for quality analysis.

Motivation & Objective

• To overcome the machine-specific limitations of existing porosity prediction models in laser powder bed fusion (LPBF) additive manufacturing.
• To develop a data-driven model that interprets machine settings into underlying physical effects such as laser energy density and radiation pressure.
• To predict porosity levels using these physics-based inputs rather than direct machine parameters, enhancing model transferability.
• To identify physical effect ranges that suppress or encourage pore formation for actionable process optimization.
• To validate the model’s performance and generalizability across different LPBF systems, including Concept Laser M2 and ProX DMP 320A.

Proposed method

• The model maps machine-specific settings (e.g., laser power, scan speed) to physical effects like laser energy density and radiation pressure using physics-based transformations.
• Physical effects are normalized using min-max scaling, with real-value ranges provided for interpretability.
• Six machine learning algorithms (e.g., Random Forest, Gaussian Process) are evaluated for predicting porosity outcomes in three categories: pass, flag, fail.
• The model uses a classification framework rather than quantitative pore size prediction, focusing on quality assurance thresholds.
• Pore-encouraging and pore-suppressing physical effect regions are identified through correlation maps between physical effects and porosity metrics (e.g., max pore diameter).
• The method is validated on Inconel 718 parts printed on a Concept Laser M2 dual-laser system, with results generalized to other machines via physical effect equivalence.

Experimental results

Research questions

• RQ1 Can translating machine settings into physical effects improve the generalizability of porosity prediction models across different LPBF systems?
• RQ2 What physical effect ranges correspond to suppressed or enhanced pore formation in LPBF processes?
• RQ3 How does the performance of a physics-informed machine learning model compare to traditional machine-setting-based models in porosity classification?
• RQ4 To what extent can the model identify interpretable, actionable process parameter regions that minimize porosity?
• RQ5 Can the model be extended to other LPBF machines with different hardware configurations, such as ProX DMP 320A or EOS M290?

Key findings

• The physics-informed machine learning model achieved prediction errors between 10% and 26% across six learning methods, demonstrating strong performance in classifying porosity levels as pass, flag, or fail.
• The model identified that pore generation is primarily driven by low to average laser power intensity (below AVE 0.58×10⁵ W/mm²), high energy variance (SDEV > 0.4×10¹ W/mm²), and wide max-min energy ranges (beyond MIN [0.6, 1.00]×10⁵ W/mm²).
• Typical pore-suppressing regions for maximum pore diameter were estimated as AVE[0.58×10⁵, 1.00×10⁵], SDEV[0.00×10¹, 0.40×10¹], MIN[0.60×10⁵, 1.00×10⁵], and MAX[0.60×10⁵, 1.00×10⁵] W/mm², indicating optimal energy distribution ranges.
• The model’s use of machine-independent physical effects enables potential transferability to other LPBF systems, such as Concept Laser M/M3, EOS M270/280/290, and ProX DMP 300/320.
• The method provides interpretable physical insights into porosity mechanisms, identifying specific effect ranges that suppress pore formation, which can guide reverse process parameter design.
• Future work will extend the model to include additional physical effects (e.g., temperature, absorption rate) and validate its performance on diverse machines like ProX DMP 320A and EOS M290, confirming its generalizability.