arXiv subjects

Stefano Discetti

Publications and source records attributed to Stefano Discetti.

At least 19 recordsLinked to original sources

arXiv2026

The balance between compactness and forecast accuracy of data-driven latent-space reduced-order models in controlled wake flows

Model-based active flow control requires predictive models that are accurate, stable, and fast enough for real-time optimisation. In controlled wake flows, this is often achieved through Reduced-Order Models (ROMs) that first compress high-dimensional velocity snapshots into a latent space and then learn a time- stepping predictor for the dynamics in the latent space. Here, we study how the choice of the spatial encoder affects the predictability of the resulting latent coordinates for wake flows under control inputs. Using two actuated 2D wake configurations, a simplified truck wake and the fluidic pinball, we compare Proper Orthogonal Decomposition (POD) against nonlinear Convolutional Autoencoders (CAEs) and two types of variational autoencoders for compression, and evaluate several temporal predictors based on Long Short-Term Memory networks. CAEs achieve higher compression efficiency and sharper short-term reconstructions, but they produce latent dynamics that are more irregular and with broadband spectral content. As a consequence, long-horizon forecasts degrade faster and show a higher probability of catastrophic divergence than POD-based models. POD yields smoother latent trajectories that are easier to learn and extrapolate, leading to more reliable predictions beyond the short- term regime. These results reveal a clear trade-off between compactness and forecast accuracy, and suggest that the stability of the latent dynamics prediction can outweigh maximal compression. This is particularly relevant for control strategies rooted in forecasts of the dynamics, such as model predictive control and reinforcement learning. The findings provide practical guidance for designing actuation-aware, hardware-feasible predictive ROMs for real-time flow control.

physics.flu-dyn↗

arXiv2026

Feature-based manifold model of actuated wakes

We propose a feature-based reduced-order model to predict the transient dynamics of bluff-body wakes under arbitrary time-varying actuation. Starting point is a control-oriented POD Galerkin modeling which is challenged by incorporating time-varying actuations as a free input. Our model includes three key enablers. First, POD modes are replaced by a more accurate feature-based manifold of same dimension. Second, a state space is distilled from dynamic features which encapsulate time-varying coherent structures. Third, this state space is augmented for the transient actuation response. Thus, a simple analytical manifold dynamics is obtained. The approach is applied to the fluidic pinball at Re=30, a canonical configuration of three identical circular cylinders arranged in an equilateral triangle and immersed in uniform flow under symmetric actuation. The model is validated against several representative actuation scenarios and accurately reproduces the transient dynamics without requiring unsteady training data, providing an interpretable, observable-based and control-oriented framework. The proposed description of actuated bluff-body flows is expected to be generalisable to other configurations.

physics.flu-dyn↗

arXiv2026

Data-efficient semi-supervised learning for flow estimation using unlabelled probe data

Estimating time-resolved velocity and pressure fields from Particle Image Velocimetry (PIV) remains challenging due to its limited temporal resolution in many applications. Data-driven approaches that combine snapshot PIV with high-frequency probe data have shown great promise in reconstructing the flow dynamics for advection-dominated flows; however, they typically exploit only the probe measurements directly synchronized with the PIV frames, leaving a large volume of probe-only data acquired between snapshots unused. In this work, we propose a framework that enriches the original PIV training dataset by time-marching a simple advection model and then exploits unlabelled probe data through a semi-supervised learning strategy. Two neural networks are trained to predict the temporal coefficients of Proper Orthogonal Decomposition (POD) modes of the flow fields, and their temporal derivatives, respectively. Unlabelled probe samples are leveraged to enforce temporal consistency and expand the coverage of flow scenarios beyond those captured by snapshot PIV, which is crucial for obtaining physically consistent temporal gradients required for pressure field reconstruction. A least-squares regularization step is further employed to reconcile the predictions and enforce consistency between temporal coefficients and their derivatives. The proposed approach is validated on both synthetic turbulent channel flow data and experimental PIV measurements of an airfoil wake. Results demonstrate that incorporating unlabelled probe data significantly improves the accuracy and temporal smoothness of velocity reconstruction, leading to more reliable pressure estimation via the Navier-Stokes equations, without increasing the experimental cost.

physics.flu-dyn↗

arXiv2026

Information decomposition for disentangled and interpretable manifold learning of fluid flows via variational autoencoders

We introduce an information-theoretic framework that uses variational autoencoders (VAEs) to extract compact, physically interpretable manifolds from high-dimensional flow-field data. To this end, the Kullback--Leibler (KL) divergence in the variational objective is decomposed into three complementary information-theoretic terms: the index-code mutual information, the total correlation, and the dimension-wise KL divergence. These terms explicitly regulate data compression, latent disentanglement, and geometric regularization. This establishes a principled basis for targeted latent-space design, allowing enhanced interpretability without sacrificing information capacity, a common drawback of heavily regularized VAE variants. The approach is evaluated on two synthetic unsteady flow datasets. First, we consider a flow around a cylinder in a channel with variable cylinder position, diameter, and Reynolds number. Later, we also consider the flow around a NACA 0012 airfoil at varying angles of attack and subjected to strong vortex gusts with variable intensity, position, and length scale. Comparisons with Principal Component Analysis, Isometric Feature Mapping, and $\beta$-VAE demonstrate clear advantages in disentanglement and physical interpretability. The learned latent coordinates successfully separate distinct physical effects. Moreover, the proposed method demonstrates strong robustness to variations in the loss-weighting parameters, despite involving a larger number of such parameters.

physics.flu-dyn↗

arXiv2026

On the turbulent wake of the actuated fluidic pinball: dynamics, bifurcations and control authority

We present the first comprehensive experimental and numerical study featuring the turbulent wake of the fluidic pinball for a large actuation range. The fluidic pinball is a cluster of three equal circular cylinders centered on the vertices of an equilateral triangle, pointing upstream in uniform flow. This configuration has become a canonical benchmark for control-oriented reduced-order modeling, for nonlinear control design and for a large kaleidoscope of drag reduction mechanisms. While the literature covers well the laminar two-dimensional Reynolds number regime, we focus on unexplored terra incognita: experiments of the symmetrically actuated turbulent regime at a Reynolds number of Re=9100. In other words, the upstream cylinder is kept stationary, while the two downstream cylinders rotate with equal and opposite angular velocities. A large range of base-bleeding and boat-tailing actuation parameters is investigated with time-resolved particle image velocimetry and aerodynamic force measurement with a companion Reynolds-averaged Navier-Stokes simulation. Our results indicate that the turbulent wake of the fluidic pinball can be approximated by a three-dimensional actuation manifold comprising two inverse pitchfork bifurcations. In the boat-tailing limit, a reduced control authority with a new low-frequency shedding state is observed.

physics.flu-dyn↗

arXiv2026

Divide and conquer complex flows. Part I: cluster and manifold-based local analysis

This work is a two-part study on the description and prediction of complex fluid flows through the partitioning of the flow domain. In this first part, we propose a framework for a global description of the dynamics of complex flows via clustered spatial representations of the flow, isolating and identifying local dynamics, retrieving different \acp{ST-CNM}. The key enabler is the partitioning of the domain based on a nonlinear manifold learning approach, in which spatial points are clustered based on the similarity of their dynamics, as observed in their compact embedding in manifold coordinates. The method receives as input time-resolved flow fields. The spatial manifold is computed through isometric mapping applied to the vorticity time histories at each spatial location. An unsupervised clustering method, applied in the manifold space, partitions the full flow domain into subdomains. The dynamics of each subdomain are then described with cluster-based modelling. The method is demonstrated on two flow-field datasets obtained with a direct numerical simulation of a fluidic pinball under periodic forcing and with two-dimensional particle image velocimetry measurements of a transitional jet flow. The spatial manifold-based flow partitioning identifies regions with similar dynamics in an automated way. For both cases, \ac{ST-CNM} identifies local dynamics that are not captured by a global approach. In particular, vortex shedding and vortex pairing dynamics are isolated in the jet flow experiment. The proposed fully automated domain partitioning method will benefit the structural description of controlled flows and unveil the actuation mechanisms at play.

physics.flu-dyn↗

arXiv2025

Drag reduction via separation control using plasma actuators on a truck cabin side

We investigate the drag reduction on a heavy-duty vehicle using dielectric-barrier discharge plasma actuators located on the A-pillars. An experimental campaign is carried out on a generalized truck model, the Ground Transportation System (GTS), which is known for its lateral separation bubbles on both sides of the truck's cabin. Measurements are performed for several yaw angles up to $7.5\degree$. Actuation is applied individually on the leeward and windward sides as well as simultaneously. Load cell measurements show that the plasma actuators effectively reduce the axial force on the GTS, with symmetric actuation achieving the highest reduction. Leeward actuation demonstrates greater control authority than the windward one; at large yaw angles the latter has a negligible effect on the axial force. Regarding side force, the leeward actuation produces a drop in its magnitude while windward actuation produces an increase. Interestingly, actuating symmetrically also augments the side force. Particle image velocimetry reveals that the plasma actuator causes a reduction in the length and width of the separation bubble on the cabin side, reducing the apparent frontal area of the truck and thus its drag. Under crosswind conditions, the stronger authority of the leeward actuator is explained by the larger separation bubble. The side force variation is driven by the net lateral suction force, which correlates with the size of the lateral recirculation regions controlled by the actuators.

physics.flu-dyn↗

arXiv2025

Latent-Space Non-Linear Model Predictive Control for Partially-Observable Systems

This work presents a scalable control framework based on nonlinear Model Predictive Control for high-dimensional dynamical systems. The proposed approach addresses the key challenges of model scalability and partial observability by integrating data-driven reduced order modelling, control in a latent space, and state estimation within a unified formulation. A predictive model is constructed via Operator Inference on a Proper Orthogonal Decomposition basis, yielding a compact latent representation that captures the dominant system dynamics. State estimation is achieved through an Unscented Kalman Filter, which reconstructs the latent space from sparse and noisy measurements, enabling closed-loop control. The input signals are computed directly in the reduced-order latent space, improving computational efficiency with negligible impact on predictive capability. The methodology is validated on the one- and two-dimensional Kuramoto--Sivashinsky equations, serving as benchmarks for chaotic and spatially-extended systems. Numerical experiments demonstrate that the proposed framework achieves accurate stabilisation. Overall, the framework provides a practical approach for nonlinear control of complex, high-dimensional systems where full-state measurements are often inaccessible or infeasible.

physics.flu-dyn↗

arXiv2025

A framework for realisable data-driven active flow control using model predictive control applied to a simplified truck wake

We present a data-driven active flow control framework designed for deployment with few non-intrusive sensors. The method builds upon Artificial Intelligence driven reduced-order predictive models based on Long-Short-Term Memory (LSTM) networks and efficient gradient-based Model Predictive Control (MPC). The model uses only surface-mounted pressure probes to infer the wake state, and is trained entirely offline on a dataset built with open-loop actuations, thus avoiding the complexities of online learning. Sparsification of the sensors needed for control from an initially large set is achieved using SHapley Additive exPlanations (SHAP). A parsimonious set of sensors is then deployed in closed-loop control with MPC. The framework is tested in numerical simulations of a two-dimensional truck model at Reynolds number 500, with pulsed-jet actuators placed in the rear of the truck to control the wake. The resulting LSTM-MPC achieved a drag reduction of 12.8\%.

physics.flu-dyn↗

arXiv2025

Sensor optimization for urban wind estimation with cluster-based probabilistic framework

We propose a physics-informed machine-learned framework for sensor-based flow estimation for drone trajectories in complex urban terrain. The input is a rich set of flow simulations at many wind conditions. The outputs are velocity and uncertainty estimates for a target domain and subsequent sensor optimization for minimal uncertainty. The framework has three innovations compared to traditional flow estimators. First, the algorithm scales proportionally to the domain complexity, making it suitable for flows that are too complex for any monolithic reduced-order representation. Second, the framework extrapolates beyond the training data, e.g., smaller and larger wind velocities. Last, and perhaps most importantly, the sensor location is a free input, significantly extending the vast majority of the literature. The key enablers are (1) a Reynolds number-based scaling of the flow variables, (2) a physics-based domain decomposition, (3) a cluster-based flow representation for each subdomain, (4) an information entropy correlating the subdomains, and (5) a multi-variate probability function relating sensor input and targeted velocity estimates. This framework is demonstrated using drone flight paths through a three-building cluster as a simple example. We anticipate adaptations and applications for estimating complete cities and incorporating weather input.

cs.LG↗

arXiv2025

An efficient offline sensor placement method for flow estimation

We present an efficient method to optimize sensor placement for flow estimation using sensors with time-delay embedding in advection-dominated flows. Our solution allows identifying promising candidates for sensor positions using solely preliminary flow field measurements with non-time-resolved Particle Image Velocimetry (PIV), without introducing physical probes in the flow. Data-driven estimation in advection-dominated flows often exploits time-delay embedding to enrich the sensor information for the reconstruction, i.e. it uses the information embedded in probe time series to provide a more accurate estimation. Optimizing the probe position is the key to improving the accuracy of such estimation. Unfortunately, the cost of performing an online combinatorial search to identify the optimal sensor placement in experiments is often prohibitive. We leverage the principle that, in advection-dominated flows, rows of vectors from PIV fields embed similar information to that of probe time series located at the downstream end of the domain. We propose thus to optimize the sensor placement using the row data from non-time-resolved PIV measurements as a surrogate of the data a real probe would actually capture in time. This optimization is run offline and requires only one preliminary experiment with standard PIV. Once the optimal positions are identified, the probes can be installed and operated simultaneously with the PIV to perform the time-resolved field estimation. We show that the proposed method outperforms equidistant positioning or greedy optimization techniques available in the literature.

physics.flu-dyn↗

arXiv2025

Meshless Super-Resolution of Scattered Data via constrained RBFs and KNN-Driven Densification

We propose a novel meshless method to achieve super resolution from scattered data obtained from sparse, randomly positioned sensors such as the particle tracers of particle tracking velocimetry. The method combines K Nearest Neighbor Particle Tracking Velocimetry (KNN PTV, Tirelli et al. 2023) with meshless Proper Orthogonal Decomposition (meshless POD, Tirelli et al. 2025) and constrained Radial Basis Function regression (c RBFs, Sperotto et al. 2022). The main idea is to enhance the spatial resolution of flow fields by blending data from locally similar flow regions available in the time series. This similarity is assessed in terms of statistical coherency with leading features identified by meshless POD applied directly to scattered data, without interpolation onto a grid and relying instead on RBFs to compute the relevant inner products. The denser scattered distributions are then used within a constrained RBF framework to derive an analytical representation of the flow fields that incorporates physical constraints. The approach is fully meshless and does not require a grid at any stage, offering flexibility in complex geometries. An ablation study highlights the role of penalties and physical constraints in regularizing the regression and ensuring physically consistent reconstructions. The method is validated using three dimensional measurements of a jet flow in air. The assessment focuses on statistics, spectra, and modal analysis. Performance is evaluated against standard Particle Image Velocimetry, KNN PTV, and c RBFs. The results show improved accuracy, with an average error of about ten percent compared to twelve to thirteen percent for the other methods, nearly halved errors in reduced order reconstructions, and a higher frequency cutoff based on the noise floor.

physics.flu-dyn↗

arXiv2025

Model-based time super-sampling of turbulent flow field sequences

We propose a novel method for model-based time super-sampling of turbulent flow fields. The key enabler is the identification of an empirical Galerkin model from the projection of the Navier-Stokes equations on a data-tailored basis. The basis is obtained from a Proper Orthogonal Decomposition (POD) of the measured fields. Time super-sampling is thus achieved by a time-marching integration of the identified dynamical system, taking the original snapshots as initial conditions. Temporal continuity of the reconstructed velocity fields is achieved through a forward-backwards integration between consecutive measured Particle Image Velocimetry measurements of a turbulent jet flow. The results are compared with the interpolation of the POD temporal coefficients and the low-order reconstruction of data measured at a higher sampling rate. In both cases, the results obtained show the ability of the method to reconstruct the dynamics of the flow with small errors during several flow characteristic times.

physics.flu-dyn↗

arXiv2025

Instantaneous convective heat transfer at the wall: a depiction of turbulent boundary layer structures

We demonstrate the ability to experimentally measure fluctuations of the convective heat transfer coefficient at the wall in a turbulent boundary layer. For this, we measure two-dimensional fields of wall-temperature fluctuations beneath a zero-pressure-gradient turbulent boundary layer, at two moderate friction Reynolds numbers ($Re_\tau \approx 990$ and $Re_\tau \approx 1800$). Spatiotemporal data of wall-temperature are acquired by means of a heated-thin-foil sensor as sensing hardware, and an infrared camera as temperature detector. At low $Re_\tau$ conditions, the fields of the Nusselt number fluctuations are populated by elongated structures comprising streamwise and spanwise length scales comparable to those of near-wall streaks. At higher $Re_\tau$ conditions, the effective width and length of the coherent $Nu$ fluctuations increases. These findings are based on two-point correlations, as well as streamwise-spanwise energy spectra of $Nu$ fluctuations. The convective velocities of the $Nu$ fluctuations are also computed with the available time resolution from the measurements. This allows for resolving the multi-scale nature of convective footprints of wall-bounded turbulence: our experimental data reflect that larger streaks in the footprint convect at velocities in the order of the free-stream velocity, while the more energetic smaller-scale features move at velocities in the order of $10u_\tau$. Measurements of the kind presented here offer a promising method for sensing, as they can be used as input to flow control systems.

physics.flu-dyn↗

arXiv2025

Full-domain POD modes from PIV asynchronous patches

A method is proposed to obtain full-domain spatial modes based on Proper Orthogonal Decomposition (POD) of Particle Image Velocimetry (PIV) measurements performed at different (overlapping) spatial locations. This situation occurs when large domains are covered by multiple non-simultaneous measurements and yet the large-scale flow field organization is to be captured. The proposed methodology leverages the definition of POD spatial modes as eigenvectors of the spatial correlation matrix, where local measurements, even when not obtained simultaneously, provide each a portion of the latter, which is then analyzed to synthesize the full-domain spatial modes. The measurement domain coverage is found to require regions overlapping by 50-75% to yield a smooth distribution of the modes. The procedure identifies structures twice as large as each measurement patch. The technique, referred to as Patch POD, is applied to planar PIV data of a submerged jet flow where the effect of patching is simulated by splitting the original PIV data. Patch POD is then extended to 3D robotic measurement around a wall-mounted cube. The results show that the patching technique enables global modal analysis over a domain covered with a multitude of non-simultaneous measurements.

physics.flu-dyn↗

arXiv2025

Advection-based multiframe iterative correction for pressure estimation from velocity fields

A novel method to improve the accuracy of pressure field estimation from time-resolved Particle Image Velocimetry data is proposed. This method generates several new time-series of velocity field by propagating in time the original one using an advection-based model, which assumes that small-scale turbulence is advected by large-scale motions. Then smoothing is performed at the corresponding positions across all the generated time-series. The process is repeated through an iterative scheme. The proposed technique smears out spatial noise by exploiting time information. Simultaneously, temporal jitter is repaired using spatial information, enhancing the accuracy of pressure computation via the Navier-Stokes equations. We provide a proof of concept of the method with synthetic datasets based on a channel flow and the wake of a 2D wing. Different noise models are tested, including Gaussian white noise and errors with some degree of spatial coherence. Additionally, the filter is evaluated on an experimental test case of the wake of an airfoil, where pressure field ground truth is not available. The result shows the proposed method performs better than conventional filters in velocity and pressure field estimation, especially when spatially coherent errors are present. The method is of direct application in advection-dominated flows, although its extension with more advanced models is straightforward.

physics.flu-dyn↗

arXiv2024

An assessment of event-based imaging velocimetry for efficient estimation of low-dimensional coordinates in turbulent flows

This study explores the potential of neuromorphic Event-Based Vision (EBV) cameras for data-efficient representation of low-order model coordinates in turbulent flows. Unlike conventional imaging systems, EBV cameras asynchronously capture changes in temporal contrast at each pixel, delivering high-frequency output with reduced data bandwidth and enhanced sensitivity, particularly in low-light conditions. Pulsed Event-Based Imaging Velocimetry (EBIV) is assessed against traditional Particle Image Velocimetry (PIV) through two synchronized experiments: a submerged water jet and airflow around a square rib in a channel. The assessment includes a detailed comparison of flow statistics and spectral content, alongside an evaluation of reduced-order modeling capabilities using Proper Orthogonal Decomposition (POD). The event stream from the EBV camera is converted into pseudo-snapshots, from which velocity fields are computed using standard PIV processing techniques. These fields are then compared after interpolation onto a common grid. Modal analysis demonstrates that EBIV can successfully identify dominant flow structures, along with their energy and dynamics, accurately discerning singular values, spatial modes, and temporal modes. While noise contamination primarily affects higher modes - less critical for flow control applications - overall performance remains robust. Additionally, comparisons of Low-Order Reconstruction (LOR) validate EBIV's capability to provide reliable reduced-order models of turbulent flows, essential for flow control purposes. These findings position EBV sensors as a promising technology for real-time, imaging-based closed-loop flow control systems.

physics.flu-dyn↗

arXiv2024

Measuring time-resolved heat transfer fluctuations on a heated-thin foil in a turbulent channel airflow

We present an experimental setup to perform time-resolved convective heat transfer measurements in a turbulent channel flow with air as the working fluid. We employ a heated thin foil coupled with high-speed infrared thermography. The measurement technique is challenged by the thermal inertia of the foil, the high frequency of turbulent fluctuations, and the measurement noise of the infrared camera. We discuss in detail the advantages and drawbacks of all the design choices that were made, thereby providing a successful implementation strategy to obtain high-quality data. This experimental approach could be useful for experimental studies employing wall-based measurements of turbulence, such as flow control applications in wall-bounded turbulence.

physics.flu-dyn↗