Multi-Fidelity Surrogate-Based Process Mapping with Uncertainty Quantification in Laser Directed Energy Deposition.

Area of Science:

• Additive Manufacturing
• Materials Science
• Computational Modeling

Background:

• Laser Directed Energy Deposition (L-DED) is a key additive manufacturing process.
• Process maps are crucial for understanding L-DED, but traditional methods are limited.
• Temporal process maps, considering time-varying parameters, are needed for advanced control.

Purpose of the Study:

• To develop a multi-fidelity Gaussian process (MFGP) surrogate for temporal process map design in L-DED.
• To establish temporal forward and inverse process maps incorporating uncertainty quantification (UQ).
• To improve optimization efficiency using MFGP coupled with Bayesian Optimization (BO).

Main Methods:

• Utilized a multi-fidelity surrogate model combining high-fidelity (FEM) and low-fidelity (analytical) models.
• Developed temporal forward process maps predicting melt pool depth from process parameters.
• Coupled MFGP with Bayesian Optimization (BO) for temporal inverse process map generation.

Main Results:

• MFGP successfully blended information from different fidelity models for temporal forward process maps.
• MFGP-BO significantly improved optimization solution quality compared to single-fidelity GP-BO.
• Demonstrated the realization of temporal forward and inverse process maps with UQ in L-DED.

Conclusions:

• The MFGP surrogate is effective for designing temporal forward and inverse process maps in L-DED.
• MFGP-BO offers a computationally efficient approach for optimizing L-DED processes.
• This work advances L-DED process control by enabling temporal map development with uncertainty quantification.