Conditional neural field for spatial dimension reduction of turbulence data: a comparison study
2510.25135v1
physics.flu-dyn, cs.LG
2025-10-31
Авторы:
Junyi Guo, Pan Du, Xiantao Fan, Yahui Li, Jian-Xun Wang
Abstract
We investigate conditional neural fields (CNFs), mesh-agnostic,
coordinate-based decoders conditioned on a low-dimensional latent, for spatial
dimensionality reduction of turbulent flows. CNFs are benchmarked against
Proper Orthogonal Decomposition and a convolutional autoencoder within a
unified encoding-decoding framework and a common evaluation protocol that
explicitly separates in-range (interpolative) from out-of-range (strict
extrapolative) testing beyond the training horizon, with identical
preprocessing, metrics, and fixed splits across all baselines. We examine three
conditioning mechanisms: (i) activation-only modulation (often termed FiLM),
(ii) low-rank weight and bias modulation (termed FP), and (iii) last-layer
inner-product coupling, and introduce a novel domain-decomposed CNF that
localizes complexities. Across representative turbulence datasets (WMLES
channel inflow, DNS channel inflow, and wall pressure fluctuations over
turbulent boundary layers), CNF-FP achieves the lowest training and in-range
testing errors, while CNF-FiLM generalizes best for out-of-range scenarios once
moderate latent capacity is available. Domain decomposition significantly
improves out-of-range accuracy, especially for the more demanding datasets. The
study provides a rigorous, physics-aware basis for selecting conditioning,
capacity, and domain decomposition when using CNFs for turbulence compression
and reconstruction.
Ссылки и действия
Дополнительные ресурсы: