Recent trends and advancements in including more diverse and heterogeneous hardware in High-Performance Computing (HPC) are challenging scientific software developers in their pursuit of efficient numerical methods with sustained performance across a diverse set of platforms. As a result, researchers are today forced to re-factor their codes to leverage these powerful new heterogeneous systems. We present our design considerations of Neko—a portable framework for high-fidelity spectral element flow simulations. Unlike prior works, Neko adopts a modern object-oriented Fortran 2008 approach, allowing multi-tier abstractions of the solver stack and facilitating various hardware backends ranging from general-purpose processors, accelerators down to exotic vector processors and Field-Programmable Gate Arrays (FPGAs). Focusing on the performance and portability of Neko, we describe the framework's device abstraction layer managing device memory, data transfer and kernel launches from Fortran, allowing for a solver written in a hardware-neutral yet performant way. Accelerator-specific optimizations are also discussed, with auto-tuning of key kernels and various communication strategies using device-aware MPI. Finally, we present performance measurements on a wide range of computing platforms, including the EuroHPC pre-exascale system LUMI, where Neko achieves excellent parallel efficiency for a large direct numerical simulation (DNS) of turbulent fluid flow using up to 80% of the entire LUMI supercomputer.

The computational power of High-Performance Computing (HPC) systems increases continuously and rapidly. Data-intensive applications are designed to leverage the high computational capacity of HPC resources and typically generate a large amount of data for traditional post-processing data analytics. However, the HPC systems' in-/output (IO) subsystem develops relatively slowly, and the storage capacity is limited. This could lead to limited actual performance and scientific discovery. In-situ techniques are a partial remedy to these problems by reducing or avoiding the data flow through the IO subsystem to/from the storage. However, in current practice, asynchronous in-situ techniques with static resource management often allocate separate computing resources for executing in-situ task(s), which remain idle if no in-situ work is at hand. In the present work, we target improving the efficiency of computing resource usage by launching and releasing necessary additional computing resources for in-situ task(s). Our approach is based on extensions for MPI Sessions that enable the required dynamic resource management. In this paper, we propose a basic and an advanced in-situ techniques with dynamic resource management enabled by MPI Sessions, their implementations on two real-world use cases, and a critical analysis of the experimental results.

An experimental characterization of the turbulent boundary layers developing around a NACA 4412 wing profile is carried out in the Minimum Turbulence Level (MTL) wind tunnel located at KTH Royal Institute of Technology. The campaign included collecting wall-pressure data via built-in pressure taps, capturing velocity signals in the turbulent boundary layers (TBLs) using hot-wire anemometry (HWA), and conducting direct skin-friction measurements with oil-film interferometry (OFI). The research spanned two chord-based Reynolds numbers (Rec=4×105 and 106) and four angles of attack (5°, 8°, 11° and 14°), encompassing a broad spectrum of flow conditions, from mild to strong adverse-pressure gradients (APGs), including scenarios where the TBL detaches from the wing surface. This dataset offers crucial insights into TBL behavior under varied flow conditions, particularly in the context of APGs. Key features include the quasi-independence of the pressure coefficient distributions from Reynolds number, which aids in distinguishing Reynolds-number effects from those due to APG strengths. The study also reveals changes in TBL dynamics as separation approaches, with energy shifting from the inner to the outer region and the eventual transition to a free-shear flow state post-separation. Additionally, the diagnostic scaling in the outer region under spatial-resolution effects is considered, showing further evidence for its applicability for small L+, however with inconsistent results for larger L+. The findings and database resulting from this campaign may be of special relevance for the development and validation of turbulence models, especially in the context of aeronautical applications.

Fully-convolutional neural networks (FCN) were proven to be effective for predicting the instantaneous state of a fully-developed turbulent flow at different wall-normal locations using quantities measured at the wall. In Guastoni et al. (J Fluid Mech 928:A27, 2021. https://doi.org/10.1017/jfm.2021.812), we focused on wall-shear-stress distributions as input, which are difficult to measure in experiments. In order to overcome this limitation, we introduce a model that can take as input the heat-flux field at the wall from a passive scalar. Four different Prandtl numbers Pr=ν/α=(1,2,4,6) are considered (where ν is the kinematic viscosity and α is the thermal diffusivity of the scalar quantity). A turbulent boundary layer is simulated since accurate heat-flux measurements can be performed in experimental settings: first we train the network on aptly-modified DNS data and then we fine-tune it on the experimental data. Finally, we test our network on experimental data sampled in a water tunnel. These predictions represent the first application of transfer learning on experimental data of neural networks trained on simulations. This paves the way for the implementation of a non-intrusive sensing approach for the flow in practical applications.

The statistics obtained from turbulent flow simulations are generally uncertain due to finite time averaging. Most techniques available in the literature to accurately estimate these uncertainties typically only work in an offline mode, that is, they require access to all available samples of a time series at once. In addition to the impossibility of online monitoring of uncertainties during the course of simulations, such an offline approach can lead to input/output (I/O) deficiencies and large storage/memory requirements, which can be problematic for large-scale simulations of turbulent flows. Here, we designed, implemented and tested a framework for estimating time-averaging uncertainties in turbulence statistics in an in-situ (online/streaming/updating) manner. The proposed algorithm relies on a novel low-memory update formula for computing the sample-estimated autocorrelation functions (ACFs). Based on this, smooth modeled ACFs of turbulence quantities can be generated to accurately estimate the time-averaging uncertainties in the corresponding sample mean estimators. The resulting uncertainty estimates are highly robust, accurate, and quantitatively the same as those obtained by standard offline estimators. Moreover, the computational overhead added by the in-situ algorithm is found to be negligible allowing for online estimation of uncertainties for multiple points and quantities. The framework is general and can be used with any flow solver and also integrated into the simulations over conformal and complex meshes created by adopting adaptive mesh refinement techniques. The results of the study are encouraging for the further development of the in-situ framework for other uncertainty quantification and data-driven analyses relevant not only to large-scale turbulent flow simulations, but also to the simulation of other dynamical systems leading to time-varying quantities with autocorrelated samples.

A framework is presented for the spectral-element code Nek5000, which has been, and still is, widely used in the computational fluid dynamics (CFD) community to perform high-fidelity numerical simulations of transitional and high Reynolds number flows. Despite the widespread usage, there is a deficiency in having a comprehensive set of tools specifically designed for conducting simulations using Nek5000. To address this issue, we have created a unique framework that allows, inter alia, to perform stability analysis and compute statistics of a turbulent flow. The framework encapsulates modules that provide tools, run-time parameters and memory structures, defining interfaces and performing different tasks. First, the framework architecture is described, showing its non-intrusive approach. Then, the modules are presented, explaining the main tools that have been implemented and describing some of the test cases. The code is open-source and available online, with proper documentation, to-run instructions and related examples.

Autoregressive models (ARMs) can be powerful tools for quantifying uncertainty in the time averages of turbulent flow quantities. This is because ARMs are efficient estimators of the autocorrelation function (ACF) of statistically stationary turbulence processes. In this study, we demonstrate a method for order selection of ARMs that uses the integral timescale of turbulence. A crucial insight into the operating principles of the ARM in terms of the time span covered by the product of model order and spacing between samples is provided, which enables us to develop computationally efficient implementations of ARM-based uncertainty estimators. This approach facilitates the quantification of uncertainty in downsampled time series and on a series of autocorrelated batch means with minimal loss of accuracy. Furthermore, a method for estimating uncertainties in second-order moments using first-order uncertainties is discussed. These techniques are applied to the time series data of turbulent flow a) through a plane channel and b) over periodic hills. Additionally, we illustrate the potential of ARMs in generating synthetic turbulence time series. Our study presents autoregressive models as intuitive and powerful tools for turbulent flows, paving the way for further applications in the field.

Active flow-control techniques have shown promise for achieving high levels of drag reduction. However, these techniques are often complex and involve multiple tunable parameters, making it challenging to optimize their efficiency. Here, we present a Bayesian optimization (BO) approach based on Gaussian process regression to optimize a wall-normal blowing and suction control scheme for a NACA 4412 wing profile at two angles of attack: 5 and 11∘, corresponding to cruise and high-lift scenarios, respectively. An automated framework is developed by linking the BO code to the CFD solver OpenFOAM. RANS simulations (validated against high-fidelity LES and experimental data) are used in order to evaluate the different flow cases. BO is shown to provide rapid convergence towards a global maximum, even when the complexity of the response function is increased by introducing a model for the cost of the flow control actuation. The importance of considering the actuation cost is highlighted: while some cases yield a net drag reduction (NDR), they may result in an overall power increase. Furthermore, optimizing for NDR or net power reduction (NPR) can lead to significantly different actuation strategies. Finally, by considering losses and efficiencies representative of real-world applications, still a significant NPR is achieved in the 11∘ case, while net power reduction is only marginally positive in the 5∘ case.

Complementing large-eddy simulation (LES) with wall-modelling is, perhaps, the most straight-forward way to enable high-fidelity simulations at high Reynolds numbers. At the same time, high-order methods offer the benefits of high computational efficiency and potentially faster convergence with respect to mesh refinement even outside the asymptotic regime.

The three-dimensional turbulent flow around a Flettner rotor, i.e. an engine-driven rotating cylinder in an atmospheric boundary layer, is studied via direct numerical simulations (DNS) for three different rotation speeds (α). This technology offers a sustainable alternative mainly for marine propulsion, underscoring the critical importance of comprehending the characteristics of such flow. In this study, we evaluate the aerodynamic loads produced by the rotor of height h, with a specific focus on the changes in lift and drag force along the vertical axis of the cylinder. Correspondingly, we observe that vortex shedding is inhibited at the highest α values investigated. However, in the case of intermediate α, vortices continue to be shed in the upper section of the cylinder (y/h>0.3). As the cylinder begins to rotate, a large-scale motion becomes apparent on the high-pressure side, close to the bottom wall. We offer both a qualitative and quantitative description of this motion, outlining its impact on the wake deflection. This finding is significant as it influences the rotor wake to an extent of approximately one hundred diameters downstream. In practical applications, this phenomenon could influence the performance of subsequent boats and have an impact on the cylinder drag, affecting its fuel consumption. This fundamental study, which investigates a limited yet significant (for DNS) Reynolds number and explores various spinning ratios, provides valuable insights into the complex flow around a Flettner rotor. The simulations were performed using a modern GPU-based spectral element method, leveraging the power of modern supercomputers towards fundamental engineering problems.