Convolutional autoencoders for the reconstruction of three-dimensional interfacial multiphase flows

2025-08-06

AI Summary

High-dimensional interface representation poses a critical bottleneck in data-driven modeling of three-dimensional multiphase flows. Method: This study proposes a convolutional autoencoder (CAE)-based model reduction framework focused on high-fidelity reconstruction of volume fraction fields. It systematically compares diffusion-based, sharp-interface, and level-set representations to quantify trade-offs between reconstruction fidelity and topological robustness. Crucially, the method decouples spatial reconstruction from temporal evolution: CAEs first learn low-dimensional latent embeddings, which then serve as structured inputs for downstream neural operators or neural ODEs. Contribution/Results: Evaluated on synthetic data and high-resolution direct numerical simulations of multiphase turbulence, the approach achieves significantly improved reconstruction accuracy—particularly for complex topological events such as droplet breakup and coalescence—while reducing latent dimensionality by over 99%. The framework establishes a generalizable, physics-informed dimensionality reduction paradigm for data-driven multiphase flow modeling.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Multi-modal Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

In this work, we perform a comprehensive investigation of autoencoders for reduced-order modeling of three-dimensional multiphase flows. Focusing on the accuracy of reconstructing multiphase flow volume/mass fractions with a standard convolutional architecture, we examine the advantages and disadvantages of different interface representation choices (diffuse, sharp, level set). We use a combination of synthetic data with non-trivial interface topologies and high-resolution simulation data of multiphase homogeneous isotropic turbulence for training and validation. This study clarifies the best practices for reducing the dimensionality of multiphase flows via autoencoders. Consequently, this paves the path for uncoupling the training of autoencoders for accurate reconstruction and the training of temporal or input/output models such as neural operators (e.g., FNOs, DeepONets) and neural ODEs on the lower-dimensional latent space given by the autoencoders. As such, the implications of this study are significant and of interest to the multiphase flow community and beyond.

Problem

• Investigates autoencoders for 3D multiphase flow reconstruction
• Compares interface representations for accurate flow fraction prediction
• Enables lower-dimensional modeling via autoencoder-derived latent spaces

Innovation

• Convolutional autoencoders for 3D flow reconstruction
• Examining interface representations for multiphase flows
• Combining synthetic and simulation data for training