Surrogate modeling of molten droplet coalescence in additive manufacturing

Patent metadata

Field	Value
Publication number	US-12318840-B2
Application number	US-202217864082-A
Country	US
Kind code	B2
Filing date	Jul 13, 2022
Priority date	Jul 13, 2022
Publication date	Jun 3, 2025
Grant date	Jun 3, 2025

Abstract

Techniques for modeling a droplet-based additive manufacturing process are disclosed. An example method includes obtaining training data, setting one or more hyperparameter values in a data-driven surrogate model architecture, and training, by a processing device, the surrogate model architecture on the training data to generate a trained surrogate model. The trained surrogate model is to be used in lieu of a physics-based model to make predictions about the results of an additive manufacturing process. The training data includes pairs of input data and output data, wherein the input data describes an initial state of a substrate and a molten droplet inside a moving subdomain prior to the molten droplet impacting the substrate and the output data describes a final state of the substrate inside that moving subdomain after the molten droplet has impacted the substrate and coalesced with previously deposited droplets making up the initial state of the substrate.

First claim

What is claimed is:

1. A method of modeling a molten droplet-based additive manufacturing process, comprising:
obtaining training data;
setting one or more hyperparameter values in a surrogate model architecture;
training, by a processing device, the surrogate model architecture on the training data to generate a trained surrogate model, wherein the trained surrogate model is to be used in lieu of a physics-based model to make predictions for predicting a state of a molten droplet deposition process,
wherein the training data comprises pairs of input data and output data, wherein the input data describes an initial state of a substrate and a molten droplet inside a moving subdomain prior to the molten droplet impacting the substrate and the output data describes a final state of the substrate inside that moving subdomain after the molten droplet has impacted the substrate and coalesced with previously solidified droplets and solidified when cooled, the previously solidified droplets making up the initial state of the substrate; and
using the trained surrogate model to predict post-coalescence substrate characteristics from pre-coalescence substrate characteristics, as part of generating a digital twin of a 3D printed part to be created by an actual 3D printer, wherein the post-coalescence substrate characteristics include local shape, temperature history, velocity, and pressure.

2. The method of claim 1 , wherein the training data is simulated using a physics-based solver.

3. The method of claim 1 , wherein the surrogate model architecture comprises at least one of machine learning, deep learning, operator learning, or a combination thereof.

4. The method of claim 1 , wherein the input data and output data comprise a three-dimensional array of voxels, wherein each voxel contains a local property of a printed part such as a local phase fraction for a single phase or a combination of phases, a local temperature, pressure, material velocity, among other possible quantities.

5. The method of claim 1 , wherein the processing device is a graphics processing unit (GPU).

6. The method of claim 1 , wherein one or more settings of the 3D printer are adjusted based on the predictions so as to improve the quality of a 3D printed part to be created by the actual 3D printer.

7. An apparatus for modeling a molten droplet-based additive manufacturing process, the apparatus comprising:
a memory to store training data; and
a processing device operatively coupled to the memory, wherein the processing device is to:
set one or more hyperparameter values in a surrogate model architecture;
train the surrogate model architecture on the training data to generate a trained surrogate model, wherein the trained surrogate model is to be used in lieu of a physics-based model to make predictions for predicting a state of a molten droplet deposition process,
wherein the training data comprises pairs of input data and output data, wherein the input data describes an initial state of a substrate and a molten droplet inside a moving subdomain prior to the molten droplet impacting the substrate and the output data describes a final state of the substrate and the molten droplet inside that moving subdomain after the molten droplet has impacted the substrate and coalesced with previously solidified droplets and solidified when cooled, the previously solidified droplets making up the initial state of the substrate; and
train and test the surrogate model, make predictions using the trained surrogate model, and predict post-coalescence substrate characteristics from pre-coalescence substrate characteristics, as part of generating a digital twin of a 3D printed part to be created by a 3D printer, wherein the post-coalescence substrate characteristics include local shape, temperature history, velocity, and pressure.

8. The apparatus of claim 7 , wherein the training data is simulated using a physics-based solver.

9. The apparatus of claim 7 , wherein the surrogate model architecture comprises at least one of machine learning, deep learning, operator learning, or a combination thereof.

10. The apparatus of claim 7 , wherein the input data and output data comprise a three-dimensional array of voxels, wherein each voxel contains a local property of a printed part such as a local phase fraction for a single phase or a combination of phases, a local temperature, pressure, material velocity, among other possible quantities.

11. The apparatus of claim 7 , wherein the processing device is a graphics processing unit (GPU).

12. The apparatus of claim 7 , wherein one or more settings of the 3D printer are adjusted based on the predictions so as to improve a quality of the 3D printed part to be created by the actual 3D printer.

13. A non-transitory computer-readable storage medium having instructions stored thereon that, when executed by a processing device, cause the processing device to:
obtain training data;
set one or more hyperparameter values in a surrogate model architecture;
train, by the processing device, the surrogate model architecture on the training data to generate a trained surrogate model, wherein the trained surrogate model is to be used in lieu of a physics-based model to make predictions for predicting a state of molten droplet deposition process,
wherein the training data comprises pairs of input data and output data, wherein the input data describes an initial state of a substrate and a molten droplet inside a moving subdomain prior to the molten droplet impacting the substrate and the output data describes a final state of the substrate and the molten droplet inside that moving subdomain after the molten droplet has impacted the substrate and coalesced with previously solidified droplets and solidified when cooled, the previously solidified droplets making up the initial state of the substrate; and
use the trained surrogate model to predict post-coalescence substrate characteristics from pre-coalescence substrate characteristics, as part of generating a digital twin of a 3D printed part to be created by an actual 3D printer, wherein the post-coalescence substrate characteristics include local shape, temperature history, velocity, and pressure.

14. The non-transitory computer-readable storage medium of claim 13 , wherein the surrogate model architecture comprises at least one of machine learning, deep learning, operator learning, or a combination thereof.

15. The non-transitory computer-readable storage medium of claim 13 , wherein the processing device is a graphics processing unit (GPU).

16. The non-transitory computer-readable storage medium of claim 13 , wherein the instructions further cause the processing device to use the trained surrogate model to predict post-coalescence substrate characteristics from pre-coalescence substrate characteristics, as part of generating a digital twin of a 3D printed part to be created by an actual 3D printer.

Assignees

• Palo Alto Res Ct Inc
• Xerox Corp

Inventors

• Taverniers Søren
• Behandish Morad
• Korneev Svyatoslav

Classifications

• G06N3/08

  Learning methods · CPC title

• B22F10/22

  Direct deposition of molten metal · CPC title

• B33Y50/02 Primary

  for controlling or regulating additive manufacturing processes · CPC title

• B22F10/38

  to achieve specific product aspects, e.g. surface smoothness, density, porosity or hollow structures · CPC title

• B22F10/30

  Process control · CPC title