From Models To Experiments: Shallow Recurrent Decoder Networks on the DYNASTY Experimental Facility

Andrea Missaglia; Antonio Cammi; Carolina Introini; J. Nathan Kutz; Stefano Riva

arxiv: 2503.08907 · v2 · submitted 2025-03-11 · 💻 cs.LG · physics.comp-ph· physics.flu-dyn

From Models To Experiments: Shallow Recurrent Decoder Networks on the DYNASTY Experimental Facility

Stefano Riva , Andrea Missaglia , Carolina Introini , J. Nathan Kutz , Antonio Cammi This is my paper

Pith reviewed 2026-05-22 23:53 UTC · model grok-4.3

classification 💻 cs.LG physics.comp-phphysics.flu-dyn

keywords Shallow Recurrent Decoderstate estimationDYNASTY facilitynatural circulationsparse sensorsnuclear engineeringRELAP5reduced basis

0 comments

The pith

Shallow Recurrent Decoder networks reconstruct full system states from three temperature sensors in the DYNASTY facility.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests the Shallow Recurrent Decoder architecture on the physical DYNASTY experimental facility, which examines natural circulation flows relevant to Generation IV nuclear reactors. It trains the network on high-fidelity RELAP5 simulations and then feeds it real temperature readings taken at the facility to recover the complete spatio-temporal fields. The work moves the method beyond prior simulation-only tests to show that three sensors, even if randomly placed, can estimate coupled quantities that are difficult to observe directly. A reader would care because this points to a practical way to monitor complex engineering systems with very limited instrumentation.

Core claim

When trained on data compressed by a reduced basis from RELAP5 and driven by three temperature measurements from the DYNASTY loop, the Shallow Recurrent Decoder recovers the full dynamics of the facility, including fields that are not directly measured, thereby validating the architecture on real experimental data.

What carries the argument

The Shallow Recurrent Decoder architecture, which merges sparse sensor inputs with a reduced-basis representation of high-fidelity model data to perform state estimation.

If this is right

Only three sensors are needed to reconstruct the entire system state, including unmeasured coupled fields.
Sensors can be placed randomly without loss of reconstruction quality.
Training occurs on compressed model data, reducing computational cost.
The same network reconstructs multiple physical fields from measurements of a single easy-to-obtain variable.
Minimal hyper-parameter adjustment is required for the architecture to function on this engineering system.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The approach could support real-time monitoring and control in operating nuclear facilities where dense sensor arrays are impractical.
Similar sparse-sensor reconstruction may apply to other natural-circulation or thermal-hydraulic experiments beyond DYNASTY.
If the reconstruction remains accurate under changing operating conditions, the method could reduce instrumentation costs in new reactor designs.

Load-bearing premise

The RELAP5-generated data match the physical dynamics of the DYNASTY facility closely enough that training on the model transfers to real measurements.

What would settle it

Large reconstruction errors appear when the network output is compared against additional temperature or flow measurements recorded at the DYNASTY facility that were withheld from training and testing.

Figures

Figures reproduced from arXiv: 2503.08907 by Andrea Missaglia, Antonio Cammi, Carolina Introini, J. Nathan Kutz, Stefano Riva.

**Figure 1.** Figure 1: FIG. 1. SHRED architecture applied to the DYNASTY facility. Three thermocouples are used to measure the temperature in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: FIG. 2. DYNASTY natural circulation loop [ [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: FIG. 3. R5 nodalization of the DYNASTY experimental fa [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4. Train (75%) -validation (12.5%) - test (12.5%) split [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: FIG. 5. Verification of SHRED using synthetic data only for parametric scenarios at [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: FIG. 6. Validation of SHRED using real data for parametric scenarios at [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 7.** Figure 7: FIG. 7. Verification and Validation of SHRED using real data for forecasting scenarios for long times at [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗

read the original abstract

The Shallow Recurrent Decoder networks are a novel paradigm recently introduced for state estimation, combining sparse observations with high-dimensional model data. This architecture features important advantages compared to standard data-driven methods including: the ability to use only three sensors (even randomly selected) for reconstructing the entire dynamics of a physical system; the ability to train on compressed data spanned by a reduced basis; the ability to measure a single field variable (easy to measure) and reconstruct coupled spatio-temporal fields that are not observable and minimal hyper-parameter tuning. This approach has been verified on different test cases within different fields including nuclear reactors, even though an application to a real experimental facility, adopting the employment of in-situ observed quantities, is missing. This work aims to fill this gap by applying the Shallow Recurrent Decoder architecture to the DYNASTY facility, built at Politecnico di Milano, which studies the natural circulation established by internally heated fluids for Generation IV applications, especially in the case of Circulating Fuel reactors. The RELAP5 code is used to generate the high-fidelity data, and temperature measurements extracted by the facility are used as input for the state estimation. The results of this work will provide a validation of the Shallow Recurrent Decoder architecture to engineering systems, showing the capabilities of this approach to provide and accurate state estimation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This applies an existing SHRED method to real DYNASTY temperature data but the accuracy claim rests on untested RELAP5 fidelity for the unmeasured fields.

read the letter

The new piece is the first reported use of SHRED on in-situ temperature readings from the operating DYNASTY natural-circulation loop rather than pure simulation test cases. That step fills the gap the abstract itself flags. The setup is straightforward: three sparse sensors feed the decoder, training happens on a reduced basis from RELAP5 runs, and the goal is to reconstruct coupled fields that are hard to measure directly in a Generation IV relevant facility. That framing is clear and the choice of facility makes sense for the nuclear-monitoring audience. Credit for moving the method out of the computer and onto hardware data. The soft spot is exactly the one the stress-test note flags. The paper trains and validates against RELAP5 output, then feeds real temperatures as the only experimental input. Nothing in the abstract shows an independent experimental check on the reconstructed velocity or concentration fields. So the reported accuracy reduces to how faithfully RELAP5 already matches the physical rig, an assumption left untested by additional measurements. No quantitative error metrics, baselines, or uncertainty numbers appear in the provided text, which makes it impossible to judge whether the reconstruction is actually useful. This is a methods-application paper aimed at engineers who need sparse-sensor state estimation for coolant monitoring. Readers already working on data-driven nuclear instrumentation will find the DYNASTY case relevant, but anyone expecting a self-contained validation will need the full results and any cross-checks. It is coherent on its own terms and shows honest engagement with the prior SHRED literature, so it deserves a serious referee to evaluate the numbers and the RELAP5-proxy assumption rather than a desk reject.

Referee Report

2 major / 1 minor

Summary. The manuscript applies the Shallow Recurrent Decoder (SHRED) architecture to the DYNASTY experimental facility, which studies natural circulation in internally heated fluids for Generation IV reactor applications. High-fidelity training data are generated via RELAP5 simulations; real temperature measurements from three (possibly randomly placed) sensors in the physical facility serve as the only inputs to reconstruct the full spatio-temporal state, including unobservable coupled fields such as velocity and concentration.

Significance. If the accuracy claim holds, the work would constitute a first validation of SHRED on a real engineering experimental facility rather than purely synthetic test cases, confirming the architecture’s advertised advantages (three-sensor reconstruction, reduced-basis training, single-field observation to multi-field reconstruction, and minimal hyper-parameter tuning) in a nuclear-engineering context.

major comments (2)

[Abstract] Abstract: the central claim that the work 'will provide a validation' and demonstrates 'accurate state estimation' is unsupported by any quantitative metrics, error bars, baseline comparisons, or description of how accuracy was measured; the soundness of the result therefore cannot be assessed from the supplied text.
[Results] Main text (results section): the validation of reconstructed fields on the physical DYNASTY facility rests on RELAP5 serving as ground truth for unobservable quantities, yet no comparison is reported between SHRED reconstructions and independent experimental measurements of those same quantities (velocity, concentration, etc.); the accuracy claim therefore reduces to the untested assumption that RELAP5 faithfully reproduces the experimental dynamics.

minor comments (1)

[Abstract] The abstract is written in future tense ('will provide a validation'), which is atypical for a manuscript that reports completed experiments and results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to improve clarity and support for the claims.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that the work 'will provide a validation' and demonstrates 'accurate state estimation' is unsupported by any quantitative metrics, error bars, baseline comparisons, or description of how accuracy was measured; the soundness of the result therefore cannot be assessed from the supplied text.

Authors: We agree that the abstract lacks quantitative support. In the revision we will replace the future tense with present tense, add specific reconstruction error metrics (e.g., relative L2 errors on temperature and derived fields), report error bars where applicable, and briefly describe the accuracy evaluation procedure used in the results. revision: yes
Referee: [Results] Main text (results section): the validation of reconstructed fields on the physical DYNASTY facility rests on RELAP5 serving as ground truth for unobservable quantities, yet no comparison is reported between SHRED reconstructions and independent experimental measurements of those same quantities (velocity, concentration, etc.); the accuracy claim therefore reduces to the untested assumption that RELAP5 faithfully reproduces the experimental dynamics.

Authors: We acknowledge the limitation: independent experimental measurements of velocity and concentration are not available from the current DYNASTY sensor suite. We will add citations to prior RELAP5 validation studies against DYNASTY temperature data and explicitly state in the results section that validation of unobservable fields is performed against the RELAP5 model (itself validated on observable temperature). This addresses the assumption while remaining within the scope of available data. revision: partial

Circularity Check

0 steps flagged

No circularity: application of existing SHRED architecture to DYNASTY data with no self-referential derivations

full rationale

The paper applies the pre-existing Shallow Recurrent Decoder (SHRED) architecture to temperature measurements from the DYNASTY facility, using RELAP5-generated data for training. No new equations, derivations, or parameter fits are introduced that reduce claimed state estimation accuracy to quantities defined by the same inputs. The central claim is an empirical application and validation on engineering data; any limitations stem from untested assumptions about RELAP5 fidelity to physical measurements, which is a correctness concern rather than a circular reduction in the method. No self-citation load-bearing steps or fitted-input-as-prediction patterns appear in the provided text.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; the central claim rests on the unstated premise that RELAP5 simulations match the physical facility and that three temperature sensors suffice for reconstruction.

pith-pipeline@v0.9.0 · 5780 in / 1052 out tokens · 46947 ms · 2026-05-22T23:53:03.223027+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

SHRED combines LSTM... Shallow Decoder... SVD... three sensors... reconstruct coupled spatio-temporal fields
IndisputableMonolith/Foundation/DimensionForcing.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

parametric dataset... rank r=10... training on compressed data

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages

[1]

The heat sink is repre- sented by the upper finned tube, optionally cooled from below through an air fan

to simulate the internal heat source (a reasonable ap- proximation as the diameter/height ratio of the facility is large), as shown in Figure 2. The heat sink is repre- sented by the upper finned tube, optionally cooled from below through an air fan. Various heating configurations can be experimentally set up, enabling either simultane- ous activation of ...

work page
[2]

the SHRED surrogate model is first verified us- 5 ing synthetic sensors randomly positioned within the domain to assess its accuracy compared to the background model in case of a parametric dataset (Section IV A)

work page
[3]

then, the sensor positions will be constrained to the experimental configuration: the SHRED is still trained using the background knowledge of the model (using parametric snapshots), and the data collected on the facility will be used to reconstruct the state in the whole domain: therefore, the abil- ity to locally correcting the model information with th...

work page
[4]

The model knowledge is used for training up to a certain time, and then real data are used to reconstruct the whole state for times outside the training range (Section IV C)

lastly, the forecast capabilities of SHRED for time instants beyond the training dataset are investi- gated, considering a single transient simulated for a long interval of time. The model knowledge is used for training up to a certain time, and then real data are used to reconstruct the whole state for times outside the training range (Section IV C). For...

work page 2000
[5]

Michael W. Grieves. Virtually Intelligent Product Sys- tems: Digital and Physical Twins , pages 175–200. Progress in Astronautics and Aeronauting, 2019

work page 2019
[6]

Subhasish Mohanty and Joseph Listwan. Development of Digital Twin Predictive Model for PWR Components: Updates on Multi Times Series Temperature Prediction Using Recurrent Neural Network, DMW Fatigue Tests, System Level Thermal-Mechanical-Stress Analysis. Tech- nical Report ANL/LWRS-21/02, 1822853, 171255, Ar- gonne National Laboratory, September 2021

work page 2021
[7]

Argaud, B Bouriquet, F de Caso, H Gong, Y Maday, and O Mula

J.-P. Argaud, B Bouriquet, F de Caso, H Gong, Y Maday, and O Mula. Sensor placement in nuclear reactors based on the generalized empirical interpolation method. J. Comput. Phys. , 363:354–370, 2018

work page 2018
[8]

A real-time variational data assimilation method with data- driven model enrichment for time-dependent problems

Willy Haik, Yvon Maday, and Ludovic Chamoin. A real-time variational data assimilation method with data- driven model enrichment for time-dependent problems. Comput. Methods Appl. Mech. Eng. , 405:115868, Febru- ary 2023

work page 2023
[9]

Multi-physics model bias correction with data-driven re- duced order techniques: Application to nuclear case stud- ies

Stefano Riva, Carolina Introini, and Antonio Cammi. Multi-physics model bias correction with data-driven re- duced order techniques: Application to nuclear case stud- ies. Applied Mathematical Modelling , 135:243–268, 2024

work page 2024
[10]

Model Order Reduction in Fluid Dynam- ics: Challenges and Perspectives

Toni Lassila, Andrea Manzoni, Alfio Quarteroni, and Gi- anluigi Rozza. Model Order Reduction in Fluid Dynam- ics: Challenges and Perspectives. In Reduced Order Meth- ods for Modeling and Computational Reduction , pages 235–273. Springer International Publishing, Cham, 2014

work page 2014
[11]

An efficient digital twin based on machine learning SVD au- toencoder and generalised latent assimilation for nuclear reactor physics

Helin Gong, Sibo Cheng, Zhang Chen, Qing Li, C´ esar Quilodr´ an-Casas, Dunhui Xiao, and Rossella Arcucci. An efficient digital twin based on machine learning SVD au- toencoder and generalised latent assimilation for nuclear reactor physics. Annals of Nuclear Energy , 179:109431, December 2022

work page 2022
[12]

Data-Driven Science and Engineering: Machine Learning, Dynami- cal Systems, and Control

Steven L Brunton and J Nathan Kutz. Data-Driven Science and Engineering: Machine Learning, Dynami- cal Systems, and Control . Cambridge University Press, USA, 2nd edition, 2022

work page 2022
[13]

Efficient deep data assimilation with sparse observations and time-varying sensors

Sibo Cheng, Che Liu, Yike Guo, and Rossella Arcucci. Efficient deep data assimilation with sparse observations and time-varying sensors. Journal of Computational Physics, 496:112581, January 2024

work page 2024
[14]

Model Order Reduction: Volume 2: Snapshot-Based Methods and Algorithms

Gianluigi Rozza, Martin Hess, Giovanni Stabile, Marco Tezzele, Francesco Ballarin, Carmen Gr¨ aßle, Michael Hinze, Stefan Volkwein, Francisco Chinesta, Pierre Lade- veze, Yvon Maday, Anthony Patera, and J Farhat Char. Model Order Reduction: Volume 2: Snapshot-Based Methods and Algorithms . De Gruyter, 2020

work page 2020
[15]

Data- driven sensor placement with shallow decoder networks

Jan Williams, Olivia Zahn, and J Nathan Kutz. Data- driven sensor placement with shallow decoder networks. arXiv preprint arXiv:2202.05330 , 2022. 10

work page arXiv 2022
[16]

Nathan Kutz, Maryam Reza, Farbod Faraji, and Aaron Knoll

J. Nathan Kutz, Maryam Reza, Farbod Faraji, and Aaron Knoll. Shallow Recurrent Decoder for Re- duced Order Modeling of Plasma Dynamics, May 2024. arXiv:2405.11955 [nlin, physics:physics]

work page arXiv 2024
[17]

Williams, Francesco Braghin, Andrea Manzoni, and J

Matteo Tomasetto, Jan P. Williams, Francesco Braghin, Andrea Manzoni, and J. Nathan Kutz. Reduced Or- der Modeling with Shallow Recurrent Decoder Networks, February 2025. arXiv:2502.10930 [cs]

work page arXiv 2025
[18]

Nathan Kutz

Stefano Riva, Carolina Introini, Antonio Cammi, and J. Nathan Kutz. Robust state estimation from partial out-core measurements with shallow recurrent decoder for nuclear reactors, 2024

work page 2024
[19]

A parameterized-background data-weak approach to variational data assimilation: formulation, analysis, and application to acoustics

Yvon Maday, Anthony Patera, James Penn, and Masayuki Yano. A parameterized-background data-weak approach to variational data assimilation: formulation, analysis, and application to acoustics. Int. J. Numer. Methods Eng., 102, 2014

work page 2014
[21]

Technology Roadmap Update for Generation IV Nuclear Energy Sys- tems, 2014

Generation IV International Forum. Technology Roadmap Update for Generation IV Nuclear Energy Sys- tems, 2014

work page 2014
[22]

Relap5/mod3 code man- ual

CD Fletcher and RR Schultz. Relap5/mod3 code man- ual. Technical report, Nuclear Regulatory Commission, Washington, DC (United States). Div. of . . . , 1992

work page 1992
[23]

Long short- term memory

Sepp Hochreiter and J¨ urgen Schmidhuber. Long short- term memory. Neural computation , 9(8):1735–1780, 1997

work page 1997
[24]

Shallow neural networks for fluid flow reconstruc- tion with limited sensors

N Benjamin Erichson, Lionel Mathelin, Zhewei Yao, Steven L Brunton, Michael W Mahoney, and J Nathan Kutz. Shallow neural networks for fluid flow reconstruc- tion with limited sensors. Proceedings of the Royal Soci- ety A , 476(2238):20200097, 2020

work page 2020
[25]

Dynamics in natural circulation loop for internally heated molten salt

A Cammi, MT Cauzzi, L Luzzi, A Pini, M Ricotti, A Siviero, D Lecarpentier, C Peniguel, D Lathouwers, M Tiberga, et al. Dynamics in natural circulation loop for internally heated molten salt. Technical report, SAMO- FAR Project (H2020), 2019

work page 2019
[26]

Capability of the relap5 code to simulate natural circulation behavior in test facilities

Amit Mangal, Vikas Jain, and AK Nayak. Capability of the relap5 code to simulate natural circulation behavior in test facilities. Progress in Nuclear Energy , 61:1–16, 2012

work page 2012
[27]

Fundamentals of heat and mass transfer

Theodore L Bergman. Fundamentals of heat and mass transfer. John Wiley & Sons, 2011

work page 2011
[28]

Handbook of hydraulic resistance

IE Idelchik. Handbook of hydraulic resistance. Journal of Pressure Vessel Technology , 109(2):260–261, 1987

work page 1987
[29]

A Novel Approach for Parametric Dynamic Mode Decomposition: Applica- tion to the DYNASTY Experimental Facility

Stefano Riva, Andrea Missaglia, Carolina Introini, In Cheol Bang, and Antonio Cammi. A Novel Approach for Parametric Dynamic Mode Decomposition: Applica- tion to the DYNASTY Experimental Facility. In The 14th International Topical Meeting on Nuclear Reactor Thermal-Hydraulics, Operation, and Safety (NUTHOS- 14), Vancouver, USA, August 2024

work page 2024
[30]

Preliminary validation of the 1d model- ing of the dynasty natural circulation loop against results from water experimental campaign

Gabriele Benzoni, Carolina Introini, Stefano Lorenzi, and Antonio Cammi. Preliminary validation of the 1d model- ing of the dynasty natural circulation loop against results from water experimental campaign. Progress in Nuclear Energy, 155:104486, 2023

work page 2023
[31]

Model order reduction methods for data assimilation; state estimation and structural health monitoring

Tommaso Taddei. Model order reduction methods for data assimilation; state estimation and structural health monitoring. PhD thesis, MIT, 2016. LIST OF SYMBOLS Acronyms DYNASTY DYnamics of NAtural circulation for molten SalT in- ternallY heated FOM Full Order Model LSTM Long Short-Term Memory ML Machine Learning PDE Partial Differential Equations R5 RELAP5...

work page 2016

[1] [1]

The heat sink is repre- sented by the upper finned tube, optionally cooled from below through an air fan

to simulate the internal heat source (a reasonable ap- proximation as the diameter/height ratio of the facility is large), as shown in Figure 2. The heat sink is repre- sented by the upper finned tube, optionally cooled from below through an air fan. Various heating configurations can be experimentally set up, enabling either simultane- ous activation of ...

work page

[2] [2]

the SHRED surrogate model is first verified us- 5 ing synthetic sensors randomly positioned within the domain to assess its accuracy compared to the background model in case of a parametric dataset (Section IV A)

work page

[3] [3]

then, the sensor positions will be constrained to the experimental configuration: the SHRED is still trained using the background knowledge of the model (using parametric snapshots), and the data collected on the facility will be used to reconstruct the state in the whole domain: therefore, the abil- ity to locally correcting the model information with th...

work page

[4] [4]

The model knowledge is used for training up to a certain time, and then real data are used to reconstruct the whole state for times outside the training range (Section IV C)

lastly, the forecast capabilities of SHRED for time instants beyond the training dataset are investi- gated, considering a single transient simulated for a long interval of time. The model knowledge is used for training up to a certain time, and then real data are used to reconstruct the whole state for times outside the training range (Section IV C). For...

work page 2000

[5] [5]

Michael W. Grieves. Virtually Intelligent Product Sys- tems: Digital and Physical Twins , pages 175–200. Progress in Astronautics and Aeronauting, 2019

work page 2019

[6] [6]

Subhasish Mohanty and Joseph Listwan. Development of Digital Twin Predictive Model for PWR Components: Updates on Multi Times Series Temperature Prediction Using Recurrent Neural Network, DMW Fatigue Tests, System Level Thermal-Mechanical-Stress Analysis. Tech- nical Report ANL/LWRS-21/02, 1822853, 171255, Ar- gonne National Laboratory, September 2021

work page 2021

[7] [7]

Argaud, B Bouriquet, F de Caso, H Gong, Y Maday, and O Mula

J.-P. Argaud, B Bouriquet, F de Caso, H Gong, Y Maday, and O Mula. Sensor placement in nuclear reactors based on the generalized empirical interpolation method. J. Comput. Phys. , 363:354–370, 2018

work page 2018

[8] [8]

A real-time variational data assimilation method with data- driven model enrichment for time-dependent problems

Willy Haik, Yvon Maday, and Ludovic Chamoin. A real-time variational data assimilation method with data- driven model enrichment for time-dependent problems. Comput. Methods Appl. Mech. Eng. , 405:115868, Febru- ary 2023

work page 2023

[9] [9]

Multi-physics model bias correction with data-driven re- duced order techniques: Application to nuclear case stud- ies

Stefano Riva, Carolina Introini, and Antonio Cammi. Multi-physics model bias correction with data-driven re- duced order techniques: Application to nuclear case stud- ies. Applied Mathematical Modelling , 135:243–268, 2024

work page 2024

[10] [10]

Model Order Reduction in Fluid Dynam- ics: Challenges and Perspectives

Toni Lassila, Andrea Manzoni, Alfio Quarteroni, and Gi- anluigi Rozza. Model Order Reduction in Fluid Dynam- ics: Challenges and Perspectives. In Reduced Order Meth- ods for Modeling and Computational Reduction , pages 235–273. Springer International Publishing, Cham, 2014

work page 2014

[11] [11]

An efficient digital twin based on machine learning SVD au- toencoder and generalised latent assimilation for nuclear reactor physics

Helin Gong, Sibo Cheng, Zhang Chen, Qing Li, C´ esar Quilodr´ an-Casas, Dunhui Xiao, and Rossella Arcucci. An efficient digital twin based on machine learning SVD au- toencoder and generalised latent assimilation for nuclear reactor physics. Annals of Nuclear Energy , 179:109431, December 2022

work page 2022

[12] [12]

Data-Driven Science and Engineering: Machine Learning, Dynami- cal Systems, and Control

Steven L Brunton and J Nathan Kutz. Data-Driven Science and Engineering: Machine Learning, Dynami- cal Systems, and Control . Cambridge University Press, USA, 2nd edition, 2022

work page 2022

[13] [13]

Efficient deep data assimilation with sparse observations and time-varying sensors

Sibo Cheng, Che Liu, Yike Guo, and Rossella Arcucci. Efficient deep data assimilation with sparse observations and time-varying sensors. Journal of Computational Physics, 496:112581, January 2024

work page 2024

[14] [14]

Model Order Reduction: Volume 2: Snapshot-Based Methods and Algorithms

Gianluigi Rozza, Martin Hess, Giovanni Stabile, Marco Tezzele, Francesco Ballarin, Carmen Gr¨ aßle, Michael Hinze, Stefan Volkwein, Francisco Chinesta, Pierre Lade- veze, Yvon Maday, Anthony Patera, and J Farhat Char. Model Order Reduction: Volume 2: Snapshot-Based Methods and Algorithms . De Gruyter, 2020

work page 2020

[15] [15]

Data- driven sensor placement with shallow decoder networks

Jan Williams, Olivia Zahn, and J Nathan Kutz. Data- driven sensor placement with shallow decoder networks. arXiv preprint arXiv:2202.05330 , 2022. 10

work page arXiv 2022

[16] [16]

Nathan Kutz, Maryam Reza, Farbod Faraji, and Aaron Knoll

J. Nathan Kutz, Maryam Reza, Farbod Faraji, and Aaron Knoll. Shallow Recurrent Decoder for Re- duced Order Modeling of Plasma Dynamics, May 2024. arXiv:2405.11955 [nlin, physics:physics]

work page arXiv 2024

[17] [17]

Williams, Francesco Braghin, Andrea Manzoni, and J

Matteo Tomasetto, Jan P. Williams, Francesco Braghin, Andrea Manzoni, and J. Nathan Kutz. Reduced Or- der Modeling with Shallow Recurrent Decoder Networks, February 2025. arXiv:2502.10930 [cs]

work page arXiv 2025

[18] [18]

Nathan Kutz

Stefano Riva, Carolina Introini, Antonio Cammi, and J. Nathan Kutz. Robust state estimation from partial out-core measurements with shallow recurrent decoder for nuclear reactors, 2024

work page 2024

[19] [19]

A parameterized-background data-weak approach to variational data assimilation: formulation, analysis, and application to acoustics

Yvon Maday, Anthony Patera, James Penn, and Masayuki Yano. A parameterized-background data-weak approach to variational data assimilation: formulation, analysis, and application to acoustics. Int. J. Numer. Methods Eng., 102, 2014

work page 2014

[20] [21]

Technology Roadmap Update for Generation IV Nuclear Energy Sys- tems, 2014

Generation IV International Forum. Technology Roadmap Update for Generation IV Nuclear Energy Sys- tems, 2014

work page 2014

[21] [22]

Relap5/mod3 code man- ual

CD Fletcher and RR Schultz. Relap5/mod3 code man- ual. Technical report, Nuclear Regulatory Commission, Washington, DC (United States). Div. of . . . , 1992

work page 1992

[22] [23]

Long short- term memory

Sepp Hochreiter and J¨ urgen Schmidhuber. Long short- term memory. Neural computation , 9(8):1735–1780, 1997

work page 1997

[23] [24]

Shallow neural networks for fluid flow reconstruc- tion with limited sensors

N Benjamin Erichson, Lionel Mathelin, Zhewei Yao, Steven L Brunton, Michael W Mahoney, and J Nathan Kutz. Shallow neural networks for fluid flow reconstruc- tion with limited sensors. Proceedings of the Royal Soci- ety A , 476(2238):20200097, 2020

work page 2020

[24] [25]

Dynamics in natural circulation loop for internally heated molten salt

A Cammi, MT Cauzzi, L Luzzi, A Pini, M Ricotti, A Siviero, D Lecarpentier, C Peniguel, D Lathouwers, M Tiberga, et al. Dynamics in natural circulation loop for internally heated molten salt. Technical report, SAMO- FAR Project (H2020), 2019

work page 2019

[25] [26]

Capability of the relap5 code to simulate natural circulation behavior in test facilities

Amit Mangal, Vikas Jain, and AK Nayak. Capability of the relap5 code to simulate natural circulation behavior in test facilities. Progress in Nuclear Energy , 61:1–16, 2012

work page 2012

[26] [27]

Fundamentals of heat and mass transfer

Theodore L Bergman. Fundamentals of heat and mass transfer. John Wiley & Sons, 2011

work page 2011

[27] [28]

Handbook of hydraulic resistance

IE Idelchik. Handbook of hydraulic resistance. Journal of Pressure Vessel Technology , 109(2):260–261, 1987

work page 1987

[28] [29]

A Novel Approach for Parametric Dynamic Mode Decomposition: Applica- tion to the DYNASTY Experimental Facility

Stefano Riva, Andrea Missaglia, Carolina Introini, In Cheol Bang, and Antonio Cammi. A Novel Approach for Parametric Dynamic Mode Decomposition: Applica- tion to the DYNASTY Experimental Facility. In The 14th International Topical Meeting on Nuclear Reactor Thermal-Hydraulics, Operation, and Safety (NUTHOS- 14), Vancouver, USA, August 2024

work page 2024

[29] [30]

Preliminary validation of the 1d model- ing of the dynasty natural circulation loop against results from water experimental campaign

Gabriele Benzoni, Carolina Introini, Stefano Lorenzi, and Antonio Cammi. Preliminary validation of the 1d model- ing of the dynasty natural circulation loop against results from water experimental campaign. Progress in Nuclear Energy, 155:104486, 2023

work page 2023

[30] [31]

Model order reduction methods for data assimilation; state estimation and structural health monitoring

Tommaso Taddei. Model order reduction methods for data assimilation; state estimation and structural health monitoring. PhD thesis, MIT, 2016. LIST OF SYMBOLS Acronyms DYNASTY DYnamics of NAtural circulation for molten SalT in- ternallY heated FOM Full Order Model LSTM Long Short-Term Memory ML Machine Learning PDE Partial Differential Equations R5 RELAP5...

work page 2016