Abstract:In this paper, we present a generic physics-informed generative model called MPDM that integrates multi-fidelity physics simulations with diffusion models. MPDM categorizes multi-fidelity physics simulations into inexpensive and expensive simulations, depending on computational costs. The inexpensive simulations, which can be obtained with low latency, directly inject contextual information into DDMs. Furthermore, when results from expensive simulations are available, MPDM refines the quality of generated samples via a guided diffusion process. This design separates the training of a denoising diffusion model from physics-informed conditional probability models, thus lending flexibility to practitioners. MPDM builds on Bayesian probabilistic models and is equipped with a theoretical guarantee that provides upper bounds on the Wasserstein distance between the sample and underlying true distribution. The probabilistic nature of MPDM also provides a convenient approach for uncertainty quantification in prediction. Our models excel in cases where physics simulations are imperfect and sometimes inaccessible. We use a numerical simulation in fluid dynamics and a case study in heat dynamics within laser-based metal powder deposition additive manufacturing to demonstrate how MPDM seamlessly integrates multi-idelity physics simulations and observations to obtain surrogates with superior predictive performance.

What problem does this paper attempt to address?

The main problem this paper attempts to address is the combination of multi-fidelity physical simulation with Diffusion Models to generate high-quality, physically consistent samples. Specifically, the paper proposes a general physics-informed generative model named MPDM (Multi-Physics Simulation Guided Generative Diffusion Models) to tackle the following two main challenges: 1. **Physical Consistency**: Standard diffusion models (such as DDMs) may generate predictions that inconsistently follow physical laws. For example, even the most advanced DDMs may produce artifacts in video generation that do not adhere to physical laws, lacking long-term consistency. 2. **Data Requirements**: Diffusion models typically require a large amount of training data, which limits their application in environments where obtaining high-quality and large-scale datasets is difficult or too costly. To address these issues, MPDM designs the following mechanisms: - **Multi-Fidelity Physical Simulation**: Physical simulations are divided into low-cost and high-cost categories. Low-cost simulations can directly provide context information for the diffusion model, while when high-cost simulation results are available, MPDM improves the quality of generated samples by guiding the diffusion process. - **Decoupled Training**: The training of the denoising diffusion model is separated from the training of the physical information conditional probability model, enhancing the model's flexibility. - **Bayesian Probability Model**: MPDM is based on a Bayesian probability model and provides theoretical guarantees to ensure that the Wasserstein distance between the generated samples and the true distribution has an upper bound. - **Uncertainty Quantification**: The probabilistic nature of MPDM offers a convenient method for uncertainty quantification in predictions. The paper demonstrates through case studies in fluid dynamics and thermodynamics in laser-based metal powder deposition additive manufacturing how MPDM seamlessly integrates multi-fidelity physical simulations and observational data to obtain surrogate models with superior predictive performance.

Multi-physics Simulation Guided Generative Diffusion Models with Applications in Fluid and Heat Dynamics

Arbitrarily-Conditioned Multi-Functional Diffusion for Multi-Physics Emulation

Diffusion-Generative Multi-Fidelity Learning for Physical Simulation

Compositional Generative Multiphysics and Multi-component Simulation

Accelerating Multiphase Flow Simulations with Denoising Diffusion Model Driven Initializations

Physics-Informed Diffusion Models

Physics-guided Deep Markov Models for Learning Nonlinear Dynamical Systems with Uncertainty

Generative downscaling of PDE solvers with physics-guided diffusion models

PhysDiff: Physics-Guided Human Motion Diffusion Model

Fast multi-fidelity gaussian processes with derivatives for complex system modeling

PF-DMD: Physics-fusion dynamic mode decomposition for accurate and robust forecasting of dynamical systems with imperfect data and physics

Inexpensive High Fidelity Melt Pool Models in Additive Manufacturing Using Generative Deep Diffusion

Unleashing the Potential of Multi-modal Foundation Models and Video Diffusion for 4D Dynamic Physical Scene Simulation

Text2PDE: Latent Diffusion Models for Accessible Physics Simulation

Generating Multiphase Fluid Configurations in Fractures using Diffusion Models

Pi-fusion: Physics-informed diffusion model for learning fluid dynamics

Multi-fidelity Physics-informed Generative Adversarial Network for Solving Partial Differential Equations

Enhancing High-Resolution Reconstruction of Flow Fields Using Physics-Informed Diffusion Model with Probability Flow Sampling

Multi-fidelity information fusion with concatenated neural networks

MPMNet: A Data-Driven MPM Framework for Dynamic Fluid-Solid Interaction

Physics Informed Distillation for Diffusion Models