Kilian Oberleithner

We investigate the broadband turbulent dynamics of attached and separated flows over a Gaussian bump, focusing on the origin of low-frequency coherent structures. The analysis combines time-resolved experimental measurements with physics-based linear models, using mean fields previously assimilated from the same dataset as base flows. Spectral proper orthogonal decomposition reveals a coherent dynamics in low- and medium-frequency regimes for both flows, with the low-frequency dynamics being substantially stronger in the separated case. In the separated flow, this dynamics is linked to a three-dimensional zero-frequency modal instability that generates large-scale streamwise-elongated structures downstream of the bump. A standing-wave model based on resolvent modes, incorporating finite-span effects, reproduces the experimentally observed spanwise structure of the dynamics and highlights the limitations of simulations with small spanwise extent and periodic boundary conditions. In the attached flow, similar low-frequency structures are identified. These are weaker, do not form a prominent standing-wave pattern and cannot be definitively classified as either modal or non-modal. The three-dimensional zero-frequency instability and finite-span standing-wave dynamics are identified as the main drivers of low-frequency coherent structures in the separated flow. They offer an explanation for persistent discrepancies between simulations and experiments on the Gaussian bump, and provide guidance on spanwise domain size and boundary conditions for future simulations.

The low-frequency unsteadiness of turbulent separation bubbles is manifested in a contraction and expansion of the reverse-flow region. This study shows that, in a pressure-gradient-induced turbulent separation bubble developing in a half-diffuser, the dominant long-time behavior follows nonlinear stochastic dynamics. Planar time-resolved particle image velocimetry reveals a spanwise-organized coherent structure with anticorrelated core and sidewall regions. Its temporal evolution is quantitatively described using a Langevin equation, modeling both the deterministic drift and stochastic forcing of the dynamical system. This model suggests that the low-frequency unsteadiness may arise from stochastic forcing, resulting in transitions between two equilibrium states. The methodology and findings documented in the present paper therefore constitute a significant step toward understanding turbulent separation bubbles as probabilistic dynamical systems.

We give evidence of non-modal amplification mechanisms driven by swirl intensity from a bi-global linear analysis of a cold swirling flow representative of a premixed swirl burner: non-uniform, compressible, turbulent, enclosed and subject to vortex breakdown passed the expansion. The monolithic computational approach embeds a realistic axisymmetric swirler model in the computational domain. The amplification mechanisms are identified by stability and resolvent analysis under variations of the length of the annular duct section and combustion chamber, the swirl intensity and the swirler position. While the spectrum is affected by changes in the length only, the gain of the resolvent strongly depends on the swirl intensity. The results suggest an acoustically dominated amplification in the combustion chamber and a non-modal hydrodynamic-dominated process driven by the swirl intensity. Inertial waves carrying swirl fluctuations play a key role in the latter. The results are complemented by a resolvent sensitivity analysis that identifies the tip of the inner recirculation region and the surrounding shear layer as a wavemaker region that drives at high swirl numbers the non-modal amplification. The sensitivity of that region also enables the transfer of azimuthal momentum perturbations to axial momentum, hence activating a longitudinal acoustic resonance from azimuthal fluctuations.

This study investigates the inference of heat release fields from time-averaged and time-resolved particle image velocimetry data using physics-informed neural networks. The method assimilates density fields using a continuity equation, from which heat release fields are calculated a posteriori using an enthalpy equation. The methodology is applied to data of a laminar, premixed methane V-flame of 1.53 kW thermal power and an equivalence ratio of 0.73 that is forced acoustically in the unsteady cases. Validation for the steady case is provided by comparing the assimilated density fields to fields obtained from the particle image velocimetry seeding concentration and by comparing the inferred heat release distribution to Abel de-convoluted OH* chemiluminescence images. The identified unsteady heat release rate fields are validated against global heat release readings from an OH*-filtered photomultiplier tube. This validation data is also transferred into a flame transfer function. The results indicate that dynamics of laminar, premixed flames can be identified solely from velocity data and detailed insights can be gained without any measurement of the heat release rate.

We investigate the mechanisms underlying the generation of broadband trailing-edge noise from an airfoil, focusing on the role of hydrodynamic structures with nonzero spanwise wavenumbers and the development of a corresponding reduced-order model. A wall-resolved compressible large-eddy simulation (LES) of a NACA0012 airfoil at 3° angle of attack, Reynolds number Re = 200 000, and Mach number M = 0.3 is performed. The simulation replicates the experimental setup of Demange . [AIAA Aviation Forum (2023)], including the tripping elements ensuring turbulent boundary layer development. The numerical domain has a large spanwise extent (43.75% of the chord) to allow for the investigation of large spanwise coherent structures. Validation against experimental data shows good agreement. The simulation results show that acoustic radiation from the trailing edge follows the scattering condition, whereby only hydrodynamic waves with spanwise wavenumbers lower than the acoustic wavenumber can generate propagative sound. Notably, the contribution from nonzero spanwise wavenumber modes becomes significant beyond the cut-on frequency at which the first nonzero spanwise acoustic mode becomes propagative. Through spectral proper orthogonal decomposition (SPOD), the associated flow structures are identified as streamwise-travelling wave packets convecting obliquely across the trailing edge. Reduced-order models based on acoustic-domain extended SPOD accurately reproduce the LES sound pressure levels within 2 dB, using only two modes per frequency–wavenumber pair. This compact representation paves the way for targeted noise-reduction strategies based on the control of the identified coherent structures.

Fluctuations in fluid flows may arise from intrinsic flow instabilities or from the amplification of external disturbances and can severely impact the performance of engineering systems. This study presents a physics-based framework for shape optimization aimed at controlling these coherently fluctuating flow structures. Linear stability and resolvent analyses of the incompressible Navier–Stokes equations are combined with adjoint-based methods to compute first- and second-order sensitivities of eigenvalue growth rates and resolvent gains with respect to geometry. These derivative quantities are integrated into a higher-order Bayesian optimization (HOBO) framework that exploits gradient and curvature information to achieve data-efficient global optimization. The method is demonstrated on two representative configurations: the wake of a hydrofoil, characterized by oscillator-type dynamics, and a backward-facing ramp, exhibiting amplifier-type behavior. In both cases, optimized shapes effectively suppress instability growth or reduce disturbance amplification. Compared to conventional gradient-following algorithms such as BFGS, the proposed HOBO approach shows faster and more robust convergence, particularly for complex, non-convex objective landscapes. The results highlight how combining linear flow analysis with higher-order statistical optimization enables physically interpretable and computationally efficient control of unsteady flow dynamics.