Machine learning for metal additive manufacturing: Predicting temperature and melt pool fluid dynamics using physics-informed neural networks (2008.13547v2)

Abstract: The recent explosion of ML and AI shows great potential in the breakthrough of metal additive manufacturing (AM) process modeling. However, the success of conventional machine learning tools in data science is primarily attributed to the unprecedented large amount of labeled data-sets (big data), which can be either obtained by experiments or first-principle simulations. Unfortunately, these labeled data-sets are expensive to obtain in AM due to the high expense of the AM experiments and prohibitive computational cost of high-fidelity simulations. We propose a physics-informed neural network (PINN) framework that fuses both data and first physical principles, including conservation laws of momentum, mass, and energy, into the neural network to inform the learning processes. To the best knowledge of the authors, this is the first application of PINN to three dimensional AM processes modeling. Besides, we propose a hard-type approach for Dirichlet boundary conditions (BCs) based on a Heaviside function, which can not only enforce the BCs but also accelerate the learning process. The PINN framework is applied to two representative metal manufacturing problems, including the 2018 NIST AM-Benchmark test series. We carefully assess the performance of the PINN model by comparing the predictions with available experimental data and high-fidelity simulation results. The investigations show that the PINN, owed to the additional physical knowledge, can accurately predict the temperature and melt pool dynamics during metal AM processes with only a moderate amount of labeled data-sets. The foray of PINN to metal AM shows the great potential of physics-informed deep learning for broader applications to advanced manufacturing.

• The paper introduces a PINN framework that integrates conservation laws to predict temperature and melt pool dynamics with reduced data reliance.
• It demonstrates that PINN predictions match high-fidelity FEM results even in sparse data scenarios.
• The study proposes a 'hard' boundary condition approach using a Heaviside function to enhance model stability and accuracy.

Predicting Temperature and Melt Pool Fluid Dynamics in Metal Additive Manufacturing Using Physics-Informed Neural Networks

The paper under discussion introduces a novel approach using physics-informed neural networks (PINNs) to predict the temperature and melt pool fluid dynamics in metal additive manufacturing (AM) processes. This method represents a significant advancement in the utilization of machine learning in manufacturing, addressing the inherent complexities and data scarcity in metal AM process modeling.

Primarily, the paper focuses on leveraging a PINN framework to reduce reliance on extensive labeled datasets, which are often expensive and challenging to obtain in metal AM due to the high cost of experiments and simulations. Traditional machine learning models depend heavily on large datasets, but the PINN approach integrates fundamental physical principles—specifically, the conservation laws of momentum, mass, and energy—directly into the learning process. This integration not only alleviates the demand for large datasets but also enhances the model's predictive accuracy by embedding domain-specific knowledge into the neural network.

The research identifies two significant applications of the PINN framework. First, the researchers apply the PINN framework to a classic one-dimensional solidification problem. This application demonstrates the capability of the PINN model to accurately predict outcomes using the energy conservation law without relying on labeled data, showcasing the framework's potential in sparse data scenarios. The paper contrasts this with the traditional finite element method (FEM), showing that even with few data points, the PINN provides predictions comparable to high-fidelity FEM simulations.

In the second application, the PINN framework is employed to predict temperature fields and melt pool fluid dynamics in a three-dimensional context during real AM processes based on the NIST AM-Benchmark test series. This application signifies the framework's robustness in handling more complex, multi-phase problems. The paper compares the predictive capabilities of the PINN model with both FEM results and experimental data, emphasizing the model's proficiency in achieving high accuracy with a reduced dataset, thus validating its practical applicability.

A key contribution of the paper is the development of a "hard" approach for imposing Dirichlet boundary conditions using a Heaviside function. This method guarantees precise enforcement of boundary conditions, enhancing the learning process's efficiency and accuracy compared to conventional "soft" approaches, which might struggle to maintain stability and accuracy, especially in domains with limited data.

The implications of this research are manifold. The integration of physics-informed methodologies within neural networks marks a paradigm shift, potentially influencing future developments in advanced manufacturing and scientific machine learning. The reduced reliance on vast amounts of labeled data, coupled with enhanced predictive accuracy, paves the way for broader adoption of machine learning in fields where data acquisition is non-trivial.

The paper also acknowledges certain limitations, such as the current inability to resolve ambient gas phases and the effects of evaporation in its models. Future research could extend the existing framework by incorporating multi-phase dynamics and evaporation models, potentially offering richer insights into metal AM processes at various scales.

Overall, this paper provides a compelling demonstration of physics-informed deep learning for metal AM, indicating promising directions for AI applications in complex manufacturing processes. The availability of the code and datasets upon publication further facilitates community engagement and validation, fostering advancements in this burgeoning field.