Abstract:Partial differential equations (PDEs) are extensively utilized for modeling various physical phenomena. These equations often depend on certain parameters, necessitating either the identification of optimal parameters or solving the equation across multiple parameters to understand how a structure might react under different conditions. Performing an exhaustive search over the parameter space requires solving the PDE multiple times, which is generally impractical. To address this challenge, Reduced Order Models (ROMs) are constructed from a snapshot dataset comprising parameter-solution pairs. ROMs facilitate the rapid solving of PDEs for new parameters. Recently, Deep Learning ROMs (DL-ROMs) have been introduced as a new method to obtain ROM. Additionally, the PDE of interest may depend on the domain, which can be characterized by parameters or measurements and may evolve with the system, requiring parametrization for ROM construction. In this paper, we develop a Deep-ROM capable of extracting and efficiently utilizing domain parametrization. Unlike traditional domain parametrization methods, our approach does not require user-defined control points and can effectively handle domains with varying numbers of components. Moreover, our model can derive meaningful parametrization even when a domain mesh is unavailable, a common scenario in biomedical applications. Our work leverages Deep Neural Networks to effectively reduce the dimensionality of the PDE and the domain characteristic function.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to use deep learning for Reduced Order Modeling (ROM) in the parametric and data - driven domains. Specifically, the paper focuses on how to efficiently solve partial differential equations (PDEs) under different parameters and domain conditions, especially when the domain shape and parameter changes are large. Traditional reduced - order model methods have limitations in dealing with such problems, for example, they require manual definition of control points, are difficult to handle different numbers of sub - domains or holes, and are difficult to parameterize in the absence of available meshes. To overcome these challenges, the authors propose a Deep Learning Reduced Order Model (DL - ROM), which can automatically extract and utilize domain parameterization without the need for users to define control points, and can effectively handle domains with different numbers of components. In addition, this model can extract meaningful parameterization in the absence of domain meshes, which is very common in biomedical applications. The main contributions of the paper include: 1. **Automatic Domain Parameterization**: The feature functions are processed by a Convolutional Autoencoder to extract domain parameterization. This method does not require users to manually define control points and can adapt to domains of different shapes and complexities. 2. **Efficient Dimensionality Reduction**: Deep Neural Networks are used to effectively reduce the dimensions of PDEs and domain feature functions, thereby improving computational efficiency. 3. **Multi - Domain Support**: This method is not only applicable to a single domain, but can also be extended to multiple - domain cases, such as Fluid - Structure Interaction Problems, providing systematic geometric parameterization. Through these innovations, the paper aims to provide a more flexible and efficient reduced - order modeling method, especially suitable for application scenarios such as design optimization and practical simulation.

Deep Learning Reduced Order Modelling on Parametric and Data driven domains

A comprehensive deep learning-based approach to reduced order modeling of nonlinear time-dependent parametrized PDEs

PTPI-DL-ROMs: pre-trained physics-informed deep learning-based reduced order models for nonlinear parametrized PDEs

CROM: Continuous Reduced-Order Modeling of PDEs Using Implicit Neural Representations

Error estimates for POD-DL-ROMs: a deep learning framework for reduced order modeling of nonlinear parametrized PDEs enhanced by proper orthogonal decomposition

Domain Decomposition Strategy for Combining Nonlinear and Linear Reduced-Order Models

On Reduced-Order Modeling of Gas–solid Flows Using Deep Learning

Accelerating PDE-constrained Inverse Solutions with Deep Learning and Reduced Order Models

Nonlinear-manifold reduced order models with domain decomposition

Handling geometrical variability in nonlinear reduced order modeling through Continuous Geometry-Aware DL-ROMs

On the latent dimension of deep autoencoders for reduced order modeling of PDEs parametrized by random fields

CD-ROM: Complemented Deep-Reduced Order Model

Solving time-dependent parametric PDEs by multiclass classification-based reduced order model

Long-time prediction of nonlinear parametrized dynamical systems by deep learning-based reduced order models

Reduced order modeling for flow and transport problems with Barlow Twins self-supervised learning

Nonlinear dimensionality reduction then and now: AIMs for dissipative PDEs in the ML era

A Dynamic Mode Decomposition Based Reduced-Order Model for Parameterized Time-Dependent Partial Differential Equations

Deep learning-based reduced-order methods for fast transient dynamics

Model order reduction assisted by deep neural networks (ROM-net)

Non-intrusive data-driven reduced-order modeling for time-dependent parametrized problems