-
A Comparison of Parametric Dynamic Mode Decomposition Algorithms for Thermal-Hydraulics Applications
Authors:
Stefano Riva,
Andrea Missaglia,
Carolina Introini,
In Cheol Bang,
Antonio Cammi
Abstract:
In recent years, algorithms aiming at learning models from available data have become quite popular due to two factors: 1) the significant developments in Artificial Intelligence techniques and 2) the availability of large amounts of data. Nevertheless, this topic has already been addressed by methodologies belonging to the Reduced Order Modelling framework, of which perhaps the most famous equati…
▽ More
In recent years, algorithms aiming at learning models from available data have become quite popular due to two factors: 1) the significant developments in Artificial Intelligence techniques and 2) the availability of large amounts of data. Nevertheless, this topic has already been addressed by methodologies belonging to the Reduced Order Modelling framework, of which perhaps the most famous equation-free technique is Dynamic Mode Decomposition. This algorithm aims to learn the best linear model that represents the physical phenomena described by a time series dataset: its output is a best state operator of the underlying dynamical system that can be used, in principle, to advance the original dataset in time even beyond its span. However, in its standard formulation, this technique cannot deal with parametric time series, meaning that a different linear model has to be derived for each parameter realization. Research on this is ongoing, and some versions of a parametric Dynamic Mode Decomposition already exist. This work contributes to this research field by comparing the different algorithms presently deployed and assessing their advantages and shortcomings compared to each other. To this aim, three different thermal-hydraulics problems are considered: two benchmark 'flow over cylinder' test cases at diverse Reynolds numbers, whose datasets are, respectively, obtained with the FEniCS finite element solver and retrieved from the CFDbench dataset, and the DYNASTY experimental facility operating at Politecnico di Milano, which studies the natural circulation established by internally heated fluids for Generation IV nuclear applications, whose dataset was generated using the RELAP5 nodal solver.
△ Less
Submitted 31 March, 2025;
originally announced March 2025.
-
Robust State Estimation from Partial Out-Core Measurements with Shallow Recurrent Decoder for Nuclear Reactors
Authors:
Stefano Riva,
Carolina Introini,
Antonio Cammi,
J. Nathan Kutz
Abstract:
Reliable, real-time state estimation in nuclear reactors is of critical importance for monitoring, control and safety. It further empowers the development of digital twins that are sufficiently accurate for real-world deployment. As nuclear engineering systems are typically characterised by extreme environments, their in-core sensing is a challenging task, even more so in Generation-IV reactor con…
▽ More
Reliable, real-time state estimation in nuclear reactors is of critical importance for monitoring, control and safety. It further empowers the development of digital twins that are sufficiently accurate for real-world deployment. As nuclear engineering systems are typically characterised by extreme environments, their in-core sensing is a challenging task, even more so in Generation-IV reactor concepts, which feature molten salt or liquid metals as thermal carriers. The emergence of data-driven methods allows for new techniques for accurate and robust estimation of the full state space vector characterising the reactor (mainly composed by neutron fluxes and the thermal-hydraulics fields). These techniques can combine different sources of information, including computational proxy models and local noisy measurements on the system, in order to robustly estimate the state. This work leverages the Shallow Recurrent Decoder (SHRED) architecture to estimate the entire state vector of a reactor from three, out-of-core time-series neutron flux measurements alone. Specifically, the Molten Salt Fast Reactor, in the EVOL geometry (Evaluation and Viability of Liquid Fuel Fast Reactor System project), is demonstrated as a test case, with neutron flux measurements alone allowing for reconstruction of the 20 coupled field variables of the dynamics. This approach can further quantify the uncertainty associated with the state estimation due to its considerably low training cost on compressed data. The accurate reconstruction of every characteristic field in real-time makes this approach suitable for monitoring and control purposes in the framework of a reactor digital twin.
△ Less
Submitted 19 September, 2024;
originally announced September 2024.
-
Roots in the semiring of finite deterministic dynamical systems
Authors:
François Doré,
Kévin Perrot,
Antonio E. Porreca,
Sara Riva,
Marius Rolland
Abstract:
Finite discrete-time dynamical systems (FDDS) model phenomena that evolve deterministically in discrete time. It is possible to define sum and product operations on these systems (disjoint union and direct product, respectively) giving a commutative semiring. This algebraic structure led to several works employing polynomial equations to model hypotheses on phenomena modelled using FDDS. To solve…
▽ More
Finite discrete-time dynamical systems (FDDS) model phenomena that evolve deterministically in discrete time. It is possible to define sum and product operations on these systems (disjoint union and direct product, respectively) giving a commutative semiring. This algebraic structure led to several works employing polynomial equations to model hypotheses on phenomena modelled using FDDS. To solve these equations, algorithms for performing the division and computing $k$-th roots are needed. In this paper, we propose two polynomial algorithms for these tasks, under the condition that the result is a connected FDDS. This ultimately leads to an efficient solution to equations of the type $AX^k=B$ for connected $X$. These results are some of the important final steps for solving more general polynomial equations on FDDS.
△ Less
Submitted 4 February, 2025; v1 submitted 15 May, 2024;
originally announced May 2024.
-
On solving basic equations over the semiring of functional digraphs
Authors:
Alberto Dennunzio,
Enrico Formenti,
Luciano Margara,
Sara Riva
Abstract:
Endowing the set of functional graphs (FGs) with the sum (disjoint union of graphs) and product (standard direct product on graphs) operations induces on FGs a structure of a commutative semiring R. The operations on R can be naturally extended to the set of univariate polynomials R[X] over R. This paper provides a polynomial time algorithm for deciding if equations of the type AX=B have solutions…
▽ More
Endowing the set of functional graphs (FGs) with the sum (disjoint union of graphs) and product (standard direct product on graphs) operations induces on FGs a structure of a commutative semiring R. The operations on R can be naturally extended to the set of univariate polynomials R[X] over R. This paper provides a polynomial time algorithm for deciding if equations of the type AX=B have solutions when A is just a single cycle and B a set of cycles of identical size. We also prove a similar complexity result for some variants of the previous equation.
△ Less
Submitted 15 May, 2025; v1 submitted 26 February, 2024;
originally announced February 2024.
-
Multi-Physics Model Bias Correction with Data-Driven Reduced Order Modelling Techniques: Application to Nuclear Case Studies
Authors:
Stefano Riva,
Carolina Introini,
Antonio Cammi
Abstract:
Nowadays, interest in combining mathematical knowledge about phenomena and data from the physical system is growing. Past research was devoted to developing so-called high-fidelity models, intending to make them able to catch most of the physical phenomena occurring in the system. Nevertheless, models will always be affected by uncertainties related, for example, to the parameters and inevitably l…
▽ More
Nowadays, interest in combining mathematical knowledge about phenomena and data from the physical system is growing. Past research was devoted to developing so-called high-fidelity models, intending to make them able to catch most of the physical phenomena occurring in the system. Nevertheless, models will always be affected by uncertainties related, for example, to the parameters and inevitably limited by the underlying simplifying hypotheses on, for example, geometry and mathematical equations; thus, in a way, there exists an upper threshold of model performance. Now, research in many engineering sectors also focuses on the so-called data-driven modelling, which aims at extracting information from available data to combine it with the mathematical model. Focusing on the nuclear field, interest in this approach is also related to the Multi-Physics modelling of nuclear reactors. Due to the multiple physics involved and their mutual and complex interactions, developing accurate and stable models both from the physical and numerical point of view remains a challenging task despite the advancements in computational hardware and software, and combining the available mathematical model with data can further improve the performance and the accuracy of the former.
This work investigates this aspect by applying two Data-Driven Reduced Order Modelling (DDROM) techniques, the Generalised Empirical Interpolation Method and the Parametrised-Background Data-Weak formulation, to literature benchmark nuclear case studies. The main goal of this work is to assess the possibility of using data to perform model bias correction, that is, verifying the reliability of DDROM approaches in improving the model performance and accuracy through the information provided by the data. The obtained numerical results are promising, foreseeing further investigation of the DDROM approach to nuclear industrial cases.
△ Less
Submitted 14 January, 2024;
originally announced January 2024.
-
An Algorithmic Pipeline for Solving Equations over Discrete Dynamical Systems Modelling Hypothesis on Real Phenomena
Authors:
Alberto Dennunzio,
Enrico Formenti,
Luciano Margara,
Sara Riva
Abstract:
This paper provides an algorithmic pipeline for studying the intrinsic structure of a finite discrete dynamical system (DDS) modelling an evolving phenomenon. Here, by intrinsic structure we mean, regarding the dynamics of the DDS under observation, the feature of resulting from the "cooperation" of the dynamics of two or more smaller DDS. The intrinsic structure is described by an equation over D…
▽ More
This paper provides an algorithmic pipeline for studying the intrinsic structure of a finite discrete dynamical system (DDS) modelling an evolving phenomenon. Here, by intrinsic structure we mean, regarding the dynamics of the DDS under observation, the feature of resulting from the "cooperation" of the dynamics of two or more smaller DDS. The intrinsic structure is described by an equation over DDS which represents a hypothesis over the phenomenon under observation. The pipeline allows solving such an equation, i.e., validating the hypothesis over the phenomenon, as far the asymptotic behavior and the number of states of the DDS under observation are concerned. The results are about the soundness and completeness of the pipeline and they are obtained by exploiting the algebraic setting for DDS introduced in [10].
△ Less
Submitted 19 December, 2022; v1 submitted 9 November, 2022;
originally announced November 2022.
-
Decomposition and factorisation of transients in Functional Graphs
Authors:
François Doré,
Enrico Formenti,
Antonio E. Porreca,
Sara Riva
Abstract:
Functional graphs (FGs) model the graph structures used to analyse the behaviour of functions from a discrete set to itself. In turn, such functions are used to study real complex phenomena evolving in time. As the systems involved can be quite large, it is interesting to decompose and factorise them into several subgraphs acting together. Polynomial equations over functional graphs provide a form…
▽ More
Functional graphs (FGs) model the graph structures used to analyse the behaviour of functions from a discrete set to itself. In turn, such functions are used to study real complex phenomena evolving in time. As the systems involved can be quite large, it is interesting to decompose and factorise them into several subgraphs acting together. Polynomial equations over functional graphs provide a formal way to represent this decomposition and factorisation mechanism, and solving them validates or invalidates hypotheses on their decomposability. The current solution method breaks down a single equation into a series of basic equations of the form AxX = B (with A, X, and B being FGs) to identify the possible solutions. However, it is able to consider just FGs made of cycles only. This work proposes an algorithm for solving these basic equations for general connected FGs. By exploiting a connection with the cancellation problem, we prove that the upper bound to the number of solutions is closely related to the size of the cycle in the coefficient A of the equation. The cancellation problem is also involved in the main algorithms provided by the paper. We introduce a polynomial-time semi-decision algorithm able to provide constraints that a potential solution will have to satisfy if it exists. Then, exploiting the ideas introduced in the first algorithm, we introduce a second exponential-time algorithm capable of finding all solutions by integrating several 'hacks' that try to keep the exponential as tight as possible.
△ Less
Submitted 4 April, 2024; v1 submitted 17 August, 2022;
originally announced August 2022.
-
Solving Equations on Discrete Dynamical Systems (Extended version)
Authors:
Alberto Dennunzio,
Enrico Formenti,
Luciano Margara,
Valentin Montmirail,
Sara Riva
Abstract:
Boolean automata networks, genetic regulation networks, and metabolic networks are just a few examples of biological modelling by discrete dynamical systems (DDS). A major issue in modelling is the verification of the model against the experimental data or inducing the model under uncertainties in the data. Equipping finite discrete dynamical systems with an algebraic structure of commutative semi…
▽ More
Boolean automata networks, genetic regulation networks, and metabolic networks are just a few examples of biological modelling by discrete dynamical systems (DDS). A major issue in modelling is the verification of the model against the experimental data or inducing the model under uncertainties in the data. Equipping finite discrete dynamical systems with an algebraic structure of commutative semiring provides a suitable context for hypothesis verification on the dynamics of DDS. Indeed, hypothesis on the systems can be translated into polynomial equations over DDS. Solutions to these equations provide the validation to the initial hypothesis. Unfortunately, finding solutions to general equations over DDS is undecidable. In this article, we want to push the envelope further by proposing a practical approach for some decidable cases in a suitable configuration that we call the Hypothesis Checking. We demonstrate that for many decidable equations all boils down to a "simpler" equation. However, the problem is not to decide if the simple equation has a solution, but to enumerate all the solutions in order to verify the hypothesis on the real and undecidable systems. We evaluate experimentally our approach and show that it has good scalability properties.
△ Less
Submitted 25 November, 2019; v1 submitted 30 April, 2019;
originally announced April 2019.
