A5081561708- Combustion and flame dynamics
- Model Reduction and Neural Networks
- Neural Networks and Reservoir Computing
- Fluid Dynamics and Turbulent Flows
- Neural Networks and Applications
- Wind and Air Flow Studies
- Aerodynamics and Acoustics in Jet Flows
- Meteorological Phenomena and Simulations
- Advanced Thermodynamic Systems and Engines
- Computational Fluid Dynamics and Aerodynamics
- Radiative Heat Transfer Studies
- Computational Physics and Python Applications
- Advanced Combustion Engine Technologies
- Energy Load and Power Forecasting
- Advanced Thermodynamics and Statistical Mechanics
- Quantum chaos and dynamical systems
- Probabilistic and Robust Engineering Design
- Plant Water Relations and Carbon Dynamics
- Image and Signal Denoising Methods
- Climate variability and models
- Fluid Dynamics and Vibration Analysis
- Heat Transfer and Boiling Studies
- Nonlinear Dynamics and Pattern Formation
- Wind Energy Research and Development
- Fluid Dynamics and Heat Transfer
The Alan Turing Institute
2020-2025
Imperial College London
2020-2025
University of Tasmania
2025
National Research Council
2023-2024
University of Cambridge
2014-2024
Turing Institute
2021-2024
Polytechnic University of Turin
2024
Politecnico di Milano
2022-2024
British Library
2023-2024
Vitenparken
2024
We apply adjoint-based sensitivity analysis to a time-delayed thermo-acoustic system: Rijke tube containing hot wire. calculate how the growth rate and frequency of small oscillations about base state are affected either by generic passive control element in system (the structural analysis) or change its base-state analysis). illustrate calculating effect second wire with heat release parameter. In single calculation, this shows changes oscillations, as function position tube. then examine...
The generation of indirect combustion noise by compositional inhomogeneities is examined theoretically. For this, the compact-nozzle theory Marble & Candel ( J. Sound Vib. , vol. 55 (2), 1977, pp. 225–243) extended to a multi-component gas mixture, and chemical potential function introduced as an additional acoustic source mechanism. Transfer functions for subcritical supercritical nozzle flows are derived, contribution compared entropy direct considering idealized downstream combustor...
The dynamics of turbulent flows is chaotic and difficult to predict. This makes the design accurate reduced-order models challenging. overarching objective this paper propose a nonlinear decomposition state predict flow based on representation dynamics. We divide into spatial problem temporal problem. First, we compute latent space, which manifold onto live. space found by series filtering operations, are performed convolutional autoencoder (CAE). CAE provides in space. Second, time...
In a thermoacoustic system, such as flame in combustor, heat release oscillations couple with acoustic pressure oscillations. If the is sufficiently phase pressure, these can grow, sometimes catastrophic consequences. Thermoacoustic instabilities are still one of most challenging problems faced by gas turbine and rocket motor manufacturers. systems characterized many parameters to which stability may be extremely sensitive. However, often only few oscillation modes unstable. Existing...
We propose a physics-constrained machine learning method-based on reservoir computing-to time-accurately predict extreme events and long-term velocity statistics in model of chaotic flow. The method leverages the strengths two different approaches: empirical modelling based computing, which learns dynamics from data only, physical conservation laws. This enables computing framework to output predictions when training are unavailable. show that combination approaches is able accurately...
We develop a versatile optimization method, which finds the design parameters that minimize time-averaged acoustic cost functionals. The method is gradient-free, model-informed, and data-driven with reservoir computing based on echo state networks. First, we analyse predictive capabilities of networks both in short- long-time prediction dynamics. find fully model-informed architectures learn chaotic dynamics, time-accurately statistically. Informing training physical reduced-order model one...
The prediction of the temporal dynamics chaotic systems is challenging because infinitesimal perturbations grow exponentially. analysis subject stability analysis. In analysis, we linearize equations dynamical system around a reference point and compute properties tangent space (i.e. Jacobian). main goal this paper to propose method that infers Jacobian, thus, properties, from observables (data). First, echo state network (ESN) with Recycle validation as tool accurately infer data. Second,...
Because of physical assumptions and numerical approximations, low-order models are affected by uncertainties in the state parameters, model biases. Model biases, also known as errors or systematic errors, difficult to infer because they 'unknown unknowns', i.e., we do not necessarily know their functional form a priori. With biased models, data assimilation methods may be ill-posed either (i) 'bias-unaware' estimators assumed unbiased, (ii) rely on an priori parametric for bias, (iii) can...
We highlight the work of a multi-university collaborative programme, PREMIERE (PREdictive Modelling with QuantIfication UncERtainty for MultiphasE Systems), which is at intersection multi-physics and machine learning, aiming to enhance predictive capabilities in complex multiphase flow systems across diverse length time scales. Our contributions encompass variety approaches, including Design Experiments nanoparticle synthesis optimisation, Generalised Latent Assimilation models drop...
In industrial settings, weakly supervised (WS) methods are usually preferred over their fully (FS) counterparts as they do not require costly manual annotations. Unfortunately, the segmentation masks obtained in WS regime typically poor terms of accuracy. this work, we present a method capable producing accurate for semantic case video streams. More specifically, build saliency maps that exploit temporal coherence between consecutive frames video, promoting consistency when objects appear...
Abstract In the current Noisy Intermediate Scale Quantum (NISQ) era, presence of noise deteriorates performance quantum computing algorithms. reservoir (QRC) is a type machine learning algorithm, which, however, can benefit from different types tuned noise. this paper, we analyze how finite sampling affects chaotic time series prediction gate-based QRC and recurrence-free (RF-QRC) models. First, examine RF-QRC show that, even without recurrent loop, it contains temporal information about...
Mantle processes control plate tectonics and exert an influence on biogeochemical cycles. However, the proportion of mantle sampled in-situ is minimal, as it buried beneath igneous crust sediments. Here we report lithological characteristics two sections from embryonic ocean drilled by International Ocean Discovery Program (IODP) in Tyrrhenian Sea. Contrary to at Mid Ridges (MORs) hyperextended passive margins, our findings reveal exceptionally heterogeneous fertile lithologies, ranging...
Data from fluid flow measurements are typically sparse, noisy, and heterogeneous, often mixed pressure velocity measurements, resulting in incomplete datasets. In this paper, we develop a physics-constrained convolutional neural network, which is deterministic tool, to reconstruct the full field data. We explore three loss functions, two of machine learning literature one designed by us: (i) softly constrained loss, allows prediction take any value; (ii) snapshot-enforced constrains at...
Temporal evolution of the solution separation two initially infinitesimally perturbed simulations for a turbulent flow. This metric provides assessment Lyapunov exponent as measure predictability time and dynamic content flow simulations.
Gas-turbine combustion chambers typically consist of nominally identical sectors arranged in a rotationally symmetric pattern. However, practice, the geometry is not perfectly symmetric. This may be due to design decisions, such as placing dampers an azimuthally nonuniform fashion, or uncertainties parameters, which break rotational symmetry chamber. The question whether these deviations from have impact on thermoacoustic-stability calculation. paper addresses this by proposing fast...
Thermoacoustic instabilities are often calculated with Helmholtz solvers combined a low-order model for the flame dynamics. Typically, such formulation leads to an eigenvalue problem in which appears under nonlinear terms, as exponentials related time delays that result from model. The objective of present paper is quantify uncertainties thermoacoustic stability analysis solver and its adjoint. This approach applied combustion test rig premixed swirl burner. adjoint solved by in-house...