Abstract:Reduced Order Models (ROMs) are of considerable importance in many areas of engineering in which computational time presents difficulties. Established approaches employ projection-based reduction, such as Proper Orthogonal Decomposition. The limitation of the linear nature of such operators is typically tackled via a library of local reduction subspaces, which requires the assembly of numerous local ROMs to address parametric dependencies. Our work attempts to define a more generalisable mapping between parametric inputs and reduced bases for the purpose of generative modeling. We propose the use of Variational Autoencoders (VAEs) in place of the typically utilised clustering or interpolation operations, for inferring the fundamental vectors, termed as modes, which approximate the manifold of the model response for any and each parametric input state. The derived ROM still relies on projection bases, built on the basis of full-order model simulations, thus retaining the imprinted physical connotation. However, it additionally exploits a matrix of coefficients that relates each local sample response and dynamics to the global phenomena across the parametric input domain. The VAE scheme is utilised for approximating these coefficients for any input state. This coupling leads to a high-precision low-order representation, which is particularly suited for problems where model dependencies or excitation traits cause the dynamic behavior to span multiple response regimes. Moreover, the probabilistic treatment of the VAE representation allows for uncertainty quantification on the reduction bases, which may then be propagated to the ROM response. The performance of the proposed approach is validated on an open-source simulation benchmark featuring hysteresis and multi-parametric dependencies, and on a large-scale wind turbine tower characterised by nonlinear material behavior and model uncertainty.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is an efficient and general - purpose method for handling parameter dependence in nonlinear systems. Traditional Reduced Order Models (ROMs) usually adopt projection - based dimension reduction methods, such as Proper Orthogonal Decomposition (POD). However, these methods have limitations when dealing with nonlinear or systems with large parameter variations. Multiple local ROMs need to be constructed to adapt to different parameter states, which leads to an increase in computational cost and a decrease in efficiency. The paper proposes a new Variational Autoencoder - boosted Reduced Order Model (VpROM), using Variational AutoEncoder (VAE) to replace traditional clustering or interpolation methods to infer the local reduced bases corresponding to different parameter inputs. This method can not only capture the nonlinear relationship between parameters and reduced bases, but also handle high - dimensional parameter dependence, and allows for uncertainty quantification of the predicted reduced bases. Specifically, the main contributions of the paper include: 1. **Nonlinear mapping**: By using VAE, the paper provides a more general method to establish a nonlinear mapping between parameter inputs and reduced bases, thereby generating a low - order model applicable to any parameter state. 2. **Uncertainty quantification**: The generative model characteristics of VAE allow for the estimation of the uncertainty of the reduced bases, which is very important in practical applications, especially in scenarios where model uncertainty needs to be considered. 3. **High - dimensional parameter handling**: Compared with traditional clustering or interpolation methods, VAE can better handle dependencies in high - dimensional parameter spaces, improving the robustness and accuracy of the model. 4. **Performance verification**: The paper verifies the effectiveness of the proposed method through two example problems, including a three - dimensional shear frame model with Bouc - Wen hysteretic nonlinearity and a finite - element model of a large - scale wind turbine tower undergoing plastic deformation. In summary, this paper aims to improve the deficiencies of traditional ROM methods in handling nonlinearity and parameter dependence by introducing VAE, and improve the efficiency and accuracy of the model.

VpROM: a novel variational autoencoder-boosted reduced order model for the treatment of parametric dependencies in nonlinear systems

VpROM: A novel Variational AutoEncoder-boosted Reduced Order Model for the treatment of parametric dependencies in nonlinear systems

A Reduced Order Model conditioned on monitoring features for estimation and uncertainty quantification in engineered systems

Physics-aware reduced-order modeling of transonic flow via β-variational autoencoder

Nonlinear Aerodynamics Reduced-Order Model Based on Multi-Input Volterra Series

Efficient Nonlinear Reduced-Order Model for Computational Fluid Dynamics-Based Aeroelastic Analysis

VENI, VINDy, VICI: a variational reduced-order modeling framework with uncertainty quantification

Data-Driven Variational Multiscale Reduced Order Models

Parametric Intrusive Reduced Order Models enhanced with Machine Learning Correction Terms

Nonintrusive projection-based reduced order modeling using stable learned differential operators

Data-driven Nonlinear Parametric Model Order Reduction Framework using Deep Hierarchical Variational Autoencoder

$β$-Variational autoencoders and transformers for reduced-order modelling of fluid flows

Enhancing non-intrusive Reduced Order Models with space-dependent aggregation methods

Data-driven reduced-order modeling for nonlinear aerodynamics using an autoencoder neural network

Towards optimal β -variational autoencoders combined with transformers for reduced-order modelling of turbulent flows

Deep Learning Reduced Order Modelling on Parametric and Data driven domains

Residual Data-Driven Variational Multiscale Reduced Order Models for Parameter Dependent Problems

A fast and accurate physics-informed neural network reduced order model with shallow masked autoencoder

SNF-ROM: Projection-based nonlinear reduced order modeling with smooth neural fields

Optimal Transport-Based Displacement Interpolation with Data Augmentation for Reduced Order Modeling of Nonlinear Dynamical Systems