Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Driven, Parameterized Reduced-order Models for Predicting Distortion in Metal 3D Printing

Indu Kant Deo, Youngsoo Choi, Saad A. Khairallah, Alexandre Reikher, Maria Strantza

TL;DR

This work tackles distortion prediction in Laser Powder Bed Fusion under varying process settings. It presents a data-driven parameterized reduced-order framework based on POD-GPR and compares it with a parameterized Graph Convolutional Autoencoder (GCA). The POD-GPR method achieves distortion accuracy of about 0.001 mm and about 1800x faster runtime than high-fidelity FE simulations, enabling rapid process optimization. The GCA struggles with generalization due to limited data but shows potential for flexible distortion prediction with larger datasets.

Abstract

In Laser Powder Bed Fusion (LPBF), the applied laser energy produces high thermal gradients that lead to unacceptable final part distortion. Accurate distortion prediction is essential for optimizing the 3D printing process and manufacturing a part that meets geometric accuracy requirements. This study introduces data-driven parameterized reduced-order models (ROMs) to predict distortion in LPBF across various machine process settings. We propose a ROM framework that combines Proper Orthogonal Decomposition (POD) with Gaussian Process Regression (GPR) and compare its performance against a deep-learning based parameterized graph convolutional autoencoder (GCA). The POD-GPR model demonstrates high accuracy, predicting distortions within $\pm0.001mm$, and delivers a computational speed-up of approximately 1800x.

Data-Driven, Parameterized Reduced-order Models for Predicting Distortion in Metal 3D Printing

TL;DR

This work tackles distortion prediction in Laser Powder Bed Fusion under varying process settings. It presents a data-driven parameterized reduced-order framework based on POD-GPR and compares it with a parameterized Graph Convolutional Autoencoder (GCA). The POD-GPR method achieves distortion accuracy of about 0.001 mm and about 1800x faster runtime than high-fidelity FE simulations, enabling rapid process optimization. The GCA struggles with generalization due to limited data but shows potential for flexible distortion prediction with larger datasets.

Abstract

In Laser Powder Bed Fusion (LPBF), the applied laser energy produces high thermal gradients that lead to unacceptable final part distortion. Accurate distortion prediction is essential for optimizing the 3D printing process and manufacturing a part that meets geometric accuracy requirements. This study introduces data-driven parameterized reduced-order models (ROMs) to predict distortion in LPBF across various machine process settings. We propose a ROM framework that combines Proper Orthogonal Decomposition (POD) with Gaussian Process Regression (GPR) and compare its performance against a deep-learning based parameterized graph convolutional autoencoder (GCA). The POD-GPR model demonstrates high accuracy, predicting distortions within , and delivers a computational speed-up of approximately 1800x.

Paper Structure

This paper contains 7 sections, 6 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Methods
  3. LPBF simulation data
  4. POD-GPR
  5. Parameterized graph-convolution autoencoder
  6. Results
  7. Conclusion

Figures (4)

  • Figure 1: Schematic of dataset generation. LPBF simulations were generated for parameter dwell time. Distortion data was extracted for each simulation and arranged into a training snapshot matrix.
  • Figure 2: Illustration of parameterized graph convolutional autoencoder architecture
  • Figure 3: Graphical representation of the first four POD coefficients of final layer predicted by the GPR model for various dwell times in the test set ($dt = \{30, 45, 60, 75\}$ seconds) along with the 95% confidence interval.
  • Figure 4: (a) Maximum displacement predictions from the POD-GPR model compared with ground truth from finite element analysis, (b) Predictions from the parameterized GCA model