Data-Driven Prediction and Uncertainty Quantification of Process Parameters for Directed Energy Deposition

Area of Science:

• Additive Manufacturing
• Materials Science
• Process Engineering

Background:

• Laser-based directed energy deposition using metal powder (DED-LB/M) enables flexible, software-defined manufacturing.
• Achieving specific track geometries in DED-LB/M requires precise knowledge of process parameters.
• Current prediction methods (analytical, numerical, machine learning) are insufficient, necessitating trial-and-error approaches.

Purpose of the Study:

• To develop a user-centric workflow for predicting optimal DED-LB/M process parameters for desired track geometries.
• To leverage Gaussian Process Regression (GPR) with uncertainty quantification (UQ) for predicting single track geometry.
• To enable inverse prediction of process parameters by combining GPR's UQ with user expertise.

Main Methods:

• Trained a Gaussian Process Regression (GPR) model using experimental data to predict single DED track geometry from process parameters.
• Incorporated uncertainty quantification (UQ) within the GPR model.
• Developed a workflow utilizing GPR's UQ and user expertise for inverse parameter prediction, minimizing deviation between target and actual track geometry.

Main Results:

• The GPR model was trained and validated on 379 cross sections of single DED tracks.
• The proposed workflow successfully proposed combinations of process parameters likely to yield desired track geometries.
• Demonstrated the workflow's benefit through two exemplary use cases.

Conclusions:

• The novel user-centric workflow effectively predicts DED-LB/M process parameters for desired track geometries.
• Integrating GPR with UQ and expert knowledge offers a significant improvement over traditional trial-and-error methods.
• This approach enhances the flexibility and efficiency of software-defined manufacturing in DED-LB/M.