A low cost singular value decomposition based data assimilation technique for analysis of heterogeneous combustion data
Prajith Pillai, Ashton Ian Hetherington, Laura Saavedra Sago, Soledad Le Clainche Martinez · 2025
Read A low cost singular value decomposition based data assimilation technique for analysis of heterogeneous combustion data by Prajith Pillai, Ashton Ian Hetherington, Laura Saavedra Sago, Soledad Le Clainche Martinez (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-24064.txt.
Introduction
Prajith Pillai and co-authors apply a cheap compression method to combustion data in a 2025 paper. Standard mode extraction from flame simulations is costly in time and memory. Their low-cost singular value decomposition reads a smart subset of points instead of the whole field. Sensor style sampling picks the points, and the method can also merge unlike data sources.
Two flame cases test it: a simulated laminar coflow flame and a turbulent hydrogen flame from lab data, they report. Speed and memory gains pass tenfold, with compression over two thousand times. Merging sensor and simulation data rebuilds the flow with errors under one percent for key species.
Catalog record
| CREATOR | Prajith Pillai, Ashton Ian Hetherington, Laura Saavedra Sago, Soledad Le Clainche Martinez |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-24064
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A low cost singular value decomposition based data assimilation technique for analysis of heterogeneous combustion data Prajith Pillai, Ashton Ian Hetherington, Laura Saavedra Sago, Soledad Le Clainche Martinez Abstract This article applies low-cost singular value decomposition (lcSVD) for the first time, to the authors’ knowledge, on combustion reactive flow databases. The lcSVD algorithm is a novel approach to SVD, suitable for calculating high-resolution 2D or 3D proper orthogonal decomposition (POD) modes and temporal coefficients using data from sensors. Consequently, the computational cost associated with this technique is much lower compared to standard SVD. Additionally, for the analysis of full n-dimensional datasets, the method reduces data dimensionality by selecting a strategically reduced number of points from the original dataset through optimal sensor placement or uniform sampling before performing SVD. Moreover, the properties of data assimilation of heterogeneous databases of this method are illustrated using two distinct reactive flow test cases: a numerical database modeling an axisymmetric, time-varying laminar coflow flame with a fuel mixture of 65% methane and 35% nitrogen, using air as the oxidizer, and experimental data generated from a turbulent bluff-body-stabilized hydrogen flame. The computational speed-up and memory gains associated with the lcSVD algorithm compared to SVD can reach values larger than 10, with compression factors greater than 2000. Applying lcSVD for data assimilation to reconstruct the flow dynamics combining data from sensors with simulation measurements, we found errors smaller than 1% in the most relevant species modelling the flow. August 24, 2026 1 Introduction Improving the efficiency of combustion systems is essential for optimizing energy use, reducing environmental impact, and minimizing operational costs. Manufacturing industries are high energy consuming sector that depend heavily on combustion of fossil fuels. They are the main source of production of chemicals, iron and steel, paper and pulp, cement, food, beverages and many other consumables. These industries alone contribute to nearly 60 percentage of the total emissions [ 1 ] in the European Union. There have been significant efforts in replacing the fossil fuels with renewable synthetic fuels to reduce emissions and achieve energy sustainability [ 2 ] . Similarly, continuous studies have been going on in developing numerical models of reactive flows by capturing the physics behind them [ 3 ] . However, combustion science is a highly complex phenomena which requires a thorough knowledge and understanding of different physical concepts involved like thermodynamics, chemical reaction, heat and mass transfer, fluid dynamics and flame propagation [ 4 ] [ 5 ] [ 6 ] . The study of reactive flows is paramount in developing sustainable technologies that can contribute to the unified goal of reducing emissions and improving energy efficiency. The numerical resolution of these phenomenons using first principles is computationally exhaustive and the datasets generated are very large. The large size of these datasets make their analysis more complicated by increasing the computational cost and memory. One way to mitigate this issue is by using high computational power. Another approach is dimensionality reduction and feature extraction from these databases, following novel approaches as the one we introduce in this article. Reduced order models (ROM) have proven to be an effective alternative for extracting physical information from complex databases with minimal computational effort and high accuracy [ 7 ] [ 8 ] . Among the various reduced order modeling techniques, modal decomposition methods like Proper Orthogonal Decomposition (POD) [ 9 ] and Dynamic Mode Decomposition (DMD) [ 10 ] are the most widely used. POD is well-regarded for its ability to simplify complex systems by identifying orthogonal modes that capture the most energetic features of the flow [ 11 ] [ 12 ] [ 13 ] . On the other hand, DMD focuses on identifying dynamic modes associated with specific temporal frequencies, making it particularly useful to identify flow patterns driving the main dynamics in unsteady and turbulent flows [ 14 ] . A robust variant of DMD, known as Higher Order Dynamic Mode Decomposition (HODMD) [ 15 ] , was developed and used to enhance the modeling of reactive flow systems, proving to be highly efficient and accurate [ 16 ] . The application of Principal Component Analysis (PCA) for feature extraction and manifold identification has been demonstrated in turbulent combustion systems, providing insights into complex dynamics [ 17 ] . Additionally, an advanced technique known as local PCA has been utilized to pinpoint low-dimensional manifolds and determine optimal reaction species in turbulent systems [ 18 ] . Conventional and advanced deep learning models have also been used as feature extraction algorithm for flame reconstruction and synthetic data generations [ 19 ] [ 20 ] [ 21 ] . However, the primary limitation of these methods lies in their high computational costs during the training process when dealing with complex industrial databases. So new methods, more efficient in terms of reduced memory and time requirements are needed. Another challenge associated with combustion datasets is their heterogeneous nature. Sensors are generally used to measure fields of interest and often lack good spatial resolution. Some measurement techniques used during experiments are thermocouples for temperature measurement, particle image velocimetry (PIV) for measuring velocity of particles in flow field, chemiluminiscence for analysing reaction zones and gas analyzers for measuring combustion residues and species [ 22 ] [ 23 ] [ 24 ] . Thermocouples and gas analyzers generally retrieve information using sparse sensing and are associated with good temporal resolution and limited spatial resolution. Obtaining highly resolved spatial information from experiments is often restricted by the challenging combustion environment and cost inefficiency. Some algorithms capable of retrieving important features from experimental data and reconstructing high resolution datasets are reported in ModelFLOWs-app [ 25 ] . This software uses pure modal decomposition algorithms and hybrid modal decomposition and deep learning models for dataset reconstruction, repairing and forecasting. There are also models based on pure deep learning for prediction and forecasting of reactive flows [ 26 ] [ 27 ] . Unlike sparsely resolved spatial data obtained through experiments, we can generate highly resolved spatial information of combustion systems through large eddy simulations (LES) or direct numerical simulations (DNS) [ 28 ] [ 29 ] [ 30 ] . However, leveraging experimental data is crucial for validation and real-world applications, which requires techniques to reconstruct complex systems using sparse sensors. The QR-based Discrete Empirical Interpolation Method (QDEIM) [ 31 ] uses reduced-order models for efficient sensor placement, while tools such as PySensors [ 35 ] optimize sparse sensing workflows, simplifying their application in practice. These two datasets (experimental and theoretical) independently hold significant information about the combustion system. This mandates the need for a mathematical framework for data assimilation that can simultaneously analyze both datasets by extracting physical information, complementing data, correcting divergent tendencies, and addressing spurious measurements. Few physics-based data assimilation techniques employing different data sources and ensemble Kalman filter (EnKF) is reported in Ref. [ 32 ] [ 33 ] . In this work, we introduce, for the first time to the authors’ knowledge, the application of low cost singular value decomposition (lcSVD) [ 34 ] to identify patterns and perform data assimilation in reactive flows. The method presents a low-cost ROM capable of reconstructing POD modes and coefficients using data from sensors. Additionally, for