arXiv (Cornell University) · 2022 · 259 citations · 52 references
Natural SciencesComputational EngineeringPhysicsComputational Fluid DynamicsComputational MechanicsFluid DynamicsDeep LearningDimensionality ReductionNeural NetworksNonlinear DynamicsPartial Differential EquationsReduced-order ModelingComputational ElastodynamicsConvolutional Neural NetworksHyper-reductionModel ReductionModel Reduction For Fluid DynamicsNonlinear Model ReductionNonlinear Partial Differential EquationsProper Orthogonal Decomposition
DL‑ROMs circumvent the limitations of conventional POD‑based ROMs—such as high‑dimensional linear trial manifolds, costly hyper‑reduction, and difficulty capturing nonlinear physics—by learning nonlinear trial manifolds and dynamics with deep neural networks, yet they remain computationally expensive to train due to the large number of network parameters. This work proposes to eliminate the costly training phase of DL‑ROMs by first reducing dimensionality with POD and then applying a multi‑fidelity pretraining stage that combines different physical models. The resulting POD‑DL‑ROM employs a POD‑reduced basis to train a compact neural network, which is pretrained on low‑fidelity models before fine‑tuning on high‑fidelity data, thereby enabling efficient learning of the reduced dynamics. Experiments on a variety of linear and nonlinear, scalar and vector, time‑dependent parametrized PDEs—including advection‑diffusion‑reaction, diffusion‑reaction, elastodynamics, and Navier‑Stokes—demonstrate the method’s broad applicability and substantial computational savings.
Deep learning-based reduced order models (DL-ROMs) have been recently proposed to overcome common limitations shared by conventional reduced order models (ROMs) - built, e.g., through proper orthogonal decomposition (POD) - when applied to nonlinear time-dependent parametrized partial differential equations (PDEs). These might be related to (i) the need to deal with projections onto high dimensional linear approximating trial manifolds, (ii) expensive hyper-reduction strategies, or (iii) the intrinsic difficulty to handle physical complexity with a linear superimposition of modes. All these aspects are avoided when employing DL-ROMs, which learn in a non-intrusive way both the nonlinear trial manifold and the reduced dynamics, by relying on deep (e.g., feedforward, convolutional, autoencoder) neural networks. Although extremely efficient at testing time, when evaluating the PDE solution for any new testing-parameter instance, DL-ROMs require an expensive training stage, because of the extremely large number of network parameters to be estimated. In this paper we propose a possible way to avoid an expensive training stage of DL-ROMs, by (i) performing a prior dimensionality reduction through POD, and (ii) relying on a multi-fidelity pretraining stage, where different physical models can be efficiently combined. The proposed POD-DL-ROM is tested on several (both scalar and vector, linear and nonlinear) time-dependent parametrized PDEs (such as, e.g., linear advection-diffusion-reaction, nonlinear diffusion-reaction, nonlinear elastodynamics, and Navier-Stokes equations) to show the generality of this approach and its remarkable computational savings.
52
Gradient-based learning applied to document recognition
Yann LeCun, Léon Bottou, Yoshua Bengio et al. · Proceedings of the IEEE · 1998 · 58.2K citations · Full text
Maziar Raissi, Paris Perdikaris, George Em Karniadakis · Journal of Computational Physics · 2018 · 16.9K citations · Full text
TensorFlow: A system for large-scale machine learning
Martı́n Abadi, Paul Barham, Jianmin Chen et al. · arXiv (Cornell University) · 2016 · 8.8K citations · Full text