Reduced-order modeling of a viscoelastic turbulent jet with hybrid machine learning models
Reduced-order modeling of a viscoelastic turbulent jet with hybrid machine learning models
Authors: Christian Amor, Adrián Corrochano, Marco Edoardo Rosti, Soledad Le Clainche Date: 2026-04-29 Paper ID: openalex:2604.26240
Summary
This paper presents a hybrid reduced-order modeling approach to address the high computational cost of simulating viscoelastic turbulent jets. By combining proper orthogonal decomposition (POD) with neural network predictors for mode coefficients, the model produces a compact, accelerated representation of complex viscoelastic fluid dynamics. Experimental results indicate that this framework effectively reconstructs long-term statistics, with deep neural architectures utilizing skip connections providing the best performance for multiscale dynamics.
Key Contributions
- Proposes a hybrid reduced-order model (ROM) integrating proper orthogonal decomposition (POD) and neural networks for efficient viscoelastic turbulent jet simulation.
- Demonstrates that neural network-based mode coefficient prediction successfully captures the long-term statistical behavior of viscoelastic turbulent flows.
- Establishes that while smaller models capture large-scale dynamics effectively, deeper models with skip connections are essential for accurately forecasting small-scale turbulent structures.
Open Questions & Future Work
Archivist Review
I approved the open question regarding the multi-scale turbulent dynamics bottleneck, as it captures a fundamental research challenge in fluid dynamics ROMs that persists despite the hybrid approach. The proposed hybrid methodology itself was rejected as a concept because it represents a domain-specific application of existing, well-understood hybrid ROM architectures.
Approved Open Questions
- Multi-scale turbulent dynamics bottleneck: Understanding how to effectively represent multi-scale turbulent dynamics in ROMs is essential for balancing predictive accuracy, model interpretability, and computational cost, especially for high-Weissenberg number viscoelastic flows where simulation is prohibitively expensive.
Rejected Candidates
- [concept] Hybrid reduced-order modeling for viscoelastic jets (
hybrid-reduced-order-modeling-viscoelastic-jet) - not_novel: The proposed method is a domain-specific application of well-established hybrid ROM (POD + NN) techniques and does not constitute a sufficiently novel, reusable architectural concept.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2604.26240
- paper_id
- 2604.26240
- paper_source
- openalex
- domain
- fluid-dynamics
- tags
-
- architectures
-
- datasets
-
- concept_slugs
-
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-05-02T05:50:22Z