Data-driven characterization of thermal models for powder-bed-fusion additive manufacturing

Area of Science:

• Computational modeling
• Additive Manufacturing (AM)
• Materials Science
• Mechanical Engineering

Background:

• Computational modeling is crucial for understanding additive manufacturing (AM) processes, predicting quality, and optimizing designs.
• A wide range of AM models exist with varying assumptions, making selection challenging for end-users due to a lack of standardized comparisons.
• Quantifying model uncertainty and evaluating the impact of incorporated physical factors are essential for reliable AM simulations.

Purpose of the Study:

• To quantify model uncertainty arising from different assumptions in additive manufacturing (AM) simulations.
• To evaluate the impact of including specific physical factors on simulation outcomes.
• To provide guidance for selecting appropriate AM models based on accuracy and computational cost.

Main Methods:

• Developed and compared multiple computational models: high-fidelity thermal-fluid flow, low-fidelity continuum heat transfer, and semi-analytical point heat source models.
• Ran simulations across various manufacturing process parameters.
• Validated model predictions using experimental data from the National Institute of Standards and Technology (NIST) Additive Manufacturing Metrology Testbed (AMMT) and employed data analytics for error characterization.

Main Results:

• Cross-comparison of simulation results demonstrated a significant influence of fluid flow on additive manufacturing (AM) process predictions.
• The importance of the powder layer's physical representation varied depending on the model's fidelity and assumptions.
• Data analytics revealed distinct error distributions for each model, reflecting their underlying assumptions.

Conclusions:

• The study provides a quantitative basis for understanding and comparing different computational models used in additive manufacturing (AM).
• Findings emphasize the critical role of fluid flow modeling and offer insights into the varying significance of powder bed representation.
• This work facilitates informed model selection for AM applications, improving simulation accuracy and guiding future model development.