arXiv subjects

Lilla Koloszar

Publications and source records attributed to Lilla Koloszar.

4 recordsLinked to original sources

arXiv2026

Reinforcement Twinning for Hybrid Control of Flapping-Wing Drones

Controlling flapping-wing drones requires controllers that handle time-varying, nonlinear, underactuated dynamics from incomplete, noisy sensor data. Recent advances in artificial intelligence (AI), particularly reinforcement learning (RL), have opened new perspectives for addressing such complex control problems through data-driven policy optimization from interaction with the environment. Yet purely data-driven methods are sample-inefficient, demanding extensive, sometimes unsafe exploration, especially without guiding physical models. This motivates hybrid AI-physics frameworks. This article proposes a hybrid model-free/model-based flight-control approach using the reinforcement twinning algorithm. The model-based (MB) component uses an adjoint formulation and an adaptive digital twin continuously identified from live trajectories; the model-free (MF) component uses RL. The two agents share knowledge via transfer learning, imitation learning, and shared experience between the real environment and the digital twin, coordinated by a policy referee that selects which agent acts in reality based on digital-twin performance and a real-to-virtual consistency ratio. The framework is evaluated for the longitudinal control of a flapping-wing drone, modelled as a nonlinear time-varying system driven by quasi-steady aerodynamic forces. The hybrid strategy is tested under three adaptive-model initializations: (1) offline identification from existing data, (2) random initialization with fully online identification, and (3) offline pre-training with biased parameters followed by online adaptation. In all cases, the hybrid framework improves performance, robustness, and sample efficiency over purely model-free and purely model-based approaches.

cs.RO↗

arXiv2025

Characterisation and extension of a rigid body dynamics solver coupled with OpenFOAM for flight performance analysis of flapping-wing drones

The extraordinary aerial agility of hummingbirds and insects continues to inspire the design of flapping-wing drones. To replicate and analyze such flight, computational fluid dynamics (CFD) simulations that couple flow solvers with rigid body dynamics are essential. While OpenFOAM offers tools for these multiphysics simulations, two key limitations remain: (1) a lack of thorough verification and performance characterization, and (2) the reliance on torque-based control for wing motion, which is impractical for parametric studies and real-time control. The developments are tested with a four and a five degrees of freedom flapping-wing drone equipped with a rigid, semi-elliptical wing. Ascending flight motions are simulated using the overset method, a moving background grid, and an LES model. Parametric studies demonstrate the independence of the grid and integration schemes, while profiling analyses identify the overset method as the computational bottleneck. The drone trajectories are compared with those from a literature quasi-steady model, and the body-wing interaction is analyzed in detail.

physics.flu-dyn↗

arXiv2024

Data-driven turbulent heat flux modeling with inputs of multiple fidelity

Data-driven RANS modeling is emerging as a promising methodology to exploit the information provided by high-fidelity data. However, its widespread application is limited by challenges in generalization and robustness to inconsistencies between input data of varying fidelity levels. This is especially true for thermal turbulent closures, which inherently depend on momentum statistics provided by low or high fidelity turbulence momentum models. This work investigates the impact of momentum modeling inconsistencies on a data-driven thermal closure trained with a dataset with multiple fidelity (DNS and RANS). The analysis of the model inputs shows that the two fidelity levels correspond to separate regions in the input space. It is here shown that such separation can be exploited by a training with heterogeneous data, allowing the model to detect the level of fidelity in its inputs and adjust its prediction accordingly. In particular, a sensitivity analysis and verification shows that such a model can leverage the data inconsistencies to increase its robustness. Finally, the verification with a CFD simulation shows the potential of this multi-fidelity training approach for flows in which momentum statistics provided by traditional models are affected by model uncertainties.

physics.flu-dyn↗

arXiv2022

Physics-constrained machine learning for thermal turbulence modelling at low Prandtl numbers

Liquid metals play a central role in new generation liquid metal cooled nuclear reactors, for which numerical investigations require the use of appropriate thermal turbulence models for low Prandtl number fluids. Given the limitations of traditional modelling approaches and the increasing availability of high-fidelity data for this class of fluids, we propose a Machine Learning strategy for the modelling of the turbulent heat flux. A comprehensive algebraic mathematical structure is derived and physical constraints are imposed to ensure attractive properties promoting applicability, robustness and stability. The closure coefficients of the model are predicted by an Artificial Neural Network (ANN) which is trained with DNS data at different Prandtl numbers. The validity of the approach was verified through a priori and a posteriori validation for two and three-dimensional liquid metal flows. The model provides a complete vectorial representation of the turbulent heat flux and the predictions fit the DNS data in a wide range of Prandtl numbers (Pr=0.01-0.71). The comparison with other existing thermal models shows that the methodology is very promising.

physics.flu-dyn↗