Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 4
Encontrada(s) 1 página(s)

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Identifying regions of importance in wall-bounded turbulence through explainable deep learning

  • Cremades-Botella, Andrés|||0000-0002-7052-4913
  • Hoyas, Sergio|||0000-0002-8458-7288
  • Quintero-Igeño, Pedro-Manuel|||0000-0003-4373-2079
  • Deshpande, Rahul
  • Lellep, Martin
  • Lee, Will Junghoon
  • Monty, Jason P.
  • Hutchins, N
  • Linkmann, M
  • Marusic, I
  • Vinuesa, Ricardo
[EN] Despite its great scientific and technological importance, wall-bounded turbulence is an unresolved problem in classical physics that requires new perspectives to be tackled. One of the key strategies has been to study interactions among the energy-containing coherent structures in the flow. Such interactions are explored in this study using an explainable deep-learning method. The instantaneous velocity field obtained from a turbulent channel flow simulation is used to predict the velocity field in time through a U-net architecture. Based on the predicted flow, we assess the importance of each structure for this prediction using the game-theoretic algorithm of SHapley Additive exPlanations (SHAP). This work provides results in agreement with previous observations in the literature and extends them by revealing that the most important structures in the flow are not necessarily the ones with the highest contribution to the Reynolds shear stress. We also apply the method to an experimental database, where we can identify structures based on their importance score. This framework has the potential to shed light on numerous fundamental phenomena of wall-bounded turbulence, including novel strategies for flow control.; Understanding the role of coherent structures in the dynamics of turbulent flows is of high relevance for fluid dynamics, climate systems, and aerodynamics. The authors propose a deep learning approach to evaluate the importance of various types of coherent structure in the flow, to uncover main mechanisms of wall-bounded turbulence and develop techniques for its control., The deep-learning-model training was enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) at Berzelius (NSC), partially funded by the Swedish Research Council through grant agreement no. 2022-06725. This project has been partially funded by the Spanish Ministry of Science, Innovation, and University through the University Faculty Training (FPU) program with reference FPU19/02201 (AC). The data has been obtained with support of grant PID2021-128676OB-I00 funded by MCIN/AEI/10.13039/ 501100011033 and by "ERDF A way of making Europe", by the European Union (SH). RV acknowledges the financial support from ERC grant no. 2021-CoG-101043998, DEEPCONTROL. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them. RD acknowledges the financial support from the Melbourne Postdoctoral Fellowship of the University of Melbourne. R.D., J.L., J.P.M., N.H., and I.M. are grateful to the ARC (Australian Research Council) for their continuous financial support.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Deep Reinforcement Learning for Flow Control Exploits Different Physics for Increasing Reynolds Number Regimes

  • Varela-Martínez, Pau
  • García-Cuevas González, Luis Miguel|||0000-0001-9340-0617
  • Suárez Morales, Pol
  • Alcántara-Ávila, Francisco
  • Miró, Arnau
  • Rabault, Jean
  • Font, Bernat
  • Lehmkuhl, Oriol
  • Vinuesa, Ricardo
[EN] The increase in emissions associated with aviation requires deeper research into novel sensing and flow-control strategies to obtain improved aerodynamic performances. In this context, data-driven methods are suitable for exploring new approaches to control the flow and develop more efficient strategies. Deep artificial neural networks (ANNs) used together with reinforcement learning, i.e., deep reinforcement learning (DRL), are receiving more attention due to their capabilities of controlling complex problems in multiple areas. In particular, these techniques have been recently used to solve problems related to flow control. In this work, an ANN trained through a DRL agent, coupled with the numerical solver Alya, is used to perform active flow control. The Tensorforce library was used to apply DRL to the simulated flow. Two-dimensional simulations of the flow around a cylinder were conducted and an active control based on two jets located on the walls of the cylinder was considered. By gathering information from the flow surrounding the cylinder, the ANN agent is able to learn through proximal policy optimization (PPO) effective control strategies for the jets, leading to a significant drag reduction. Furthermore, the agent needs to account for the coupled effects of the friction- and pressure-drag components, as well as the interaction between the two boundary layers on both sides of the cylinder and the wake. In the present work, a Reynolds number range beyond those previously considered was studied and compared with results obtained using classical flow-control methods. Significantly different forms of nature in the control strategies were identified by the DRL as the Reynolds number Re increased. On the one hand, for Re¿1000, the classical control strategy based on an opposition control relative to the wake oscillation was obtained. On the other hand, for Re=2000, the new strategy consisted of energization of the boundary layers and the separation area, which modulated the flow separation and reduced the drag in a fashion similar to that of the drag crisis, through a high-frequency actuation. A cross-application of agents was performed for a flow at Re=2000, obtaining similar results in terms of the drag reduction with the agents trained at Re=1000 and 2000. The fact that two different strategies yielded the same performance made us question whether this Reynolds number regime (Re=2000) belongs to a transition towards a nature-different flow, which would only admits a high-frequency actuation strategy to obtain the drag reduction. At the same time, this finding allows for the application of ANNs trained at lower Reynolds numbers, but are comparable in nature, saving computational resources., The authors acknowledge the contribution of Maxence Deferrez to this work. R.V. acknowledges funding from the ERC through grant no. 2021-CoG-101043998, DEEPCONTROL
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Tesis doctoral (DoctoralThesis)

Poiseuille and Couette flows: the effect of the small scales on the large structures

  • Gandía Barberá, Sergio
[ES] Diferentes flujos turbulentos en canales a diversos números de Reynolds de fricción son estudiados para elucidar los cambios en su organización de escalas turbulentas y la transferencia de energía originada por el esfuerzo cortante medio. Los flujos en canales representan el flujo más simple limitado por paredes, proporcionando un modelo simplificado de la dinámica de la turbulencia en geometrías mucho más complejas limitadas por paredes. En este estudio, nos centramos en flujos de Poiseuille, flujos de Couette y configuraciones combinadas. Por un lado, un flujo de canal de Poiseuille es un flujo impulsado por presión en el cual el fluido se mueve entre paredes paralelas debido a un gradiente de presión. Por otro lado, un flujo de canal de Couette es un flujo impulsado por esfuerzo cortante en el cual el fluido se mueve debido al movimiento relativo de una pared respecto a la otra, sin que intervenga un gradiente de presión. Las configuraciones combinadas contienen ambos mecanismos y cubren una transición escalonada desde un flujo puramente de Poiseuille hasta un flujo puramente de Couette. La física de ambos flujos se resuelve utilizando Simulación Numérica Directa (DNS) a números de Reynolds de fricción bajos a moderados, proporcionando información en todas las escalas turbulentas. El estudio analiza y caracteriza remolinos y estructuras coherentes, incluidas streaks y estructuras intensas de esfuerzos de Reynolds. Se calculan y comparan entre las configuraciones de flujo en canal parámetros geométricos, intensidad turbulenta, ubicación espacial y otros parámetros relevantes mediante gráficos probabilísticos y tendencias. En esta tesis, elucidamos que el esfuerzo cortante medio puede crear estructuras de gran escala promediadas en el tiempo, en forma de streaks y rolls, en configuraciones puramente de Couette y combinadas con suficiente contribución de esfuerzo cortante. Las estructuras de gran escala deben estar presentes para producir estadísticas relevantes del canal; por lo tanto, el tamaño mínimo de dominio en la dirección transversal está limitado por ellas. La generación de estas estructuras de gran escala está vinculada a un proceso autosostenido, el cual es sensible a fuerzas externas tales como la estratificación estable. De hecho, explicamos que una estratificación relativamente pequeña es capaz de romper este proceso autosostenido y, en consecuencia, prevenir el crecimiento de estructuras de gran escala. Además, estas estructuras de gran escala producen una reasignación de estructuras coherentes instantáneas a lo largo de la dirección transversal. Curiosamente, estas streaks y bursts instantáneas permanecen autosemejantes en cualquier configuración, independientemente de la presencia de estructuras de gran escala, de modo que sus proporciones dimensionales no difieren. Por último, al observar flujos de Couette térmicos pasivos, aparecen estructuras de gran escala promediadas en el tiempo en forma de regiones frías y calientes de temperatura., [CA] Diferents fluxos turbulents en canals a diversos números de Reynolds de fricció són estudiats per a elucidar els canvis en la seua organització d'escales turbulentes i la transferència d'energia originada per la cizalla mitjana. Els fluxos en canals representen el flux més simple limitat per parets, proporcionant un model simplificat de la dinàmica de la turbulència en geometries molt més complexes limitades per parets. En este estudi, ens centrem en fluxos de Poiseuille, fluxos de Couette i configuracions combinades. D'una banda, un flux de canal de Poiseuille és un flux impulsat per pressió en el qual el fluid es mou entre parets paral·leles a causa d'un gradient de pressió. D'altra banda, un flux de canal de Couette és un flux impulsat per cizalla en el qual el fluid es mou pel moviment relatiu d'una paret respecte a l'altra, sense que intervinga cap gradient de pressió. Les configuracions combinades contenen ambdós mecanismes i cobreixen una transició escalonada des d'un flux purament de Poiseuille fins a un flux purament de Couette. La física d'ambdós fluxos es resol utilitzant Simulació Numèrica Directa (DNS) a números de Reynolds de fricció baixos a moderats, proporcionant informació en totes les escales turbulentes. L'estudi analitza i caracteritza remolins i estructures coherents, incloent-hi streaks i estructures intenses d'esforços de Reynolds. Es calculen i comparen entre les configuracions de flux en canal paràmetres geomètrics, intensitat turbulenta, ubicació espacial i altres paràmetres rellevants mitjançant gràfics probabilístics i tendències. En esta tesi, elucidarem que la cizalla mitjana pot crear estructures de gran escala promediades en el temps, en forma de streaks i rolls, en configuracions purament de Couette i combinades amb suficient contribució de cizalla. Les estructures de gran escala han d'estar presents per a produir estadístiques rellevants del canal; per tant, la mida mínima del domini en la direcció transversal queda limitada per elles. La generació d'estes estructures de gran escala està vinculada a un procés autosostingut, el qual és sensible a forces externes com ara l'estratificació estable. De fet, expliquem que una estratificació relativament xicoteta és capaç de trencar este procés autosostingut i, en conseqüència, prevenir el creixement d'estructures de gran escala. A més, estes estructures de gran escala produeixen una reallocació d'estructures coherents instantànies al llarg de la direcció transversal. Curiosament, estes streaks i bursts instantanis romanen autosemejants en qualsevol configuració, independentment de la presència d'estructures de gran escala, de manera que les seues proporcions dimensionals no diferixen. Finalment, en observar fluxos de Couette tèrmics passius, apareixen estructures de gran escala promediades en el temps en forma de regions fredes i calentes de temperatura., [EN] Different turbulent channel flows at diverse friction Reynolds numbers are studied to elucidate the changes in their turbulent scale organisation and energy transfer originating from the mean shear. Channel flows represent the simplest wall-bounded flow, providing a simplified model of the turbulence dynamics in much more complex wall-bounded geometries. In this study, we focus on Poiseuille flows, Couette flows and combined configurations. On the one hand, a Poiseuille channel flow is a pressure-driven flow in which the fluid moves between parallel walls due to a pressure gradient. On the other hand, a Couette channel flow is a shear-driven flow in which the fluid moves due to the relative movement of one wall to the other with no pressure gradient involved. The combined configurations contain both mechanisms and cover a stepped transition from a pure Poiseuille flow to a pure Couette flow. The physics of both flows is resolved using Direct Numerical Simulation (DNS) at low to moderate Reynolds friction numbers, providing information at all turbulent scales. The study analyses and characterises eddies and coherent structures, including streaks and intense Reynolds stress structures. Geometrical parameters, turbulent intensity, spatial location and other relevant parameters are calculated and compared among channel flow configurations using probabilistic plots and trends. In this thesis, we elucidate that the mean shear can create time-averaged largescale structures, in the form of streaks and rolls, in pure Couette and combined configurations with enough shear contribution. Large-scale structures must be present in order to produce relevant channel statistics; therefore, the minimum domain spanwise size is limited by them. The generation of these large-scale structures is linked to a self-sustaining process, which is sensitive to external forces such as stable stratification. Indeed, we explain that relatively small stratification is able to break this self-sustaining process and consequently prevents the growth of large-scale structures. Furthermore, these large-scale structures produce a reallocation of instantaneous coherent structures along the spanwise direction. Curiously enough, these instantaneous streaks and bursts remain self-similar in any configuration, independently of the presence of large-scale structures, so their dimensional ratios do not differ. Last but not least, when observing passive thermal Couette flows, similar time-averaged large-scale structures appear in the form of cold and hot temperature regions., The present study was developed at the Instituto Universitario de Investigación de Matemática Pura y Aplicada (IUMPA) of the Universitat Politècnica de València (UPV). This work was supported by MINECO/FEDER under grants RTI2018-102256-B-I00 and ENE2015-71333-R, and by MCIN/AEI under grant PID2021-128676OB-I00, co-funded by the European Union through the European Regional Development Fund (ERDF) “A Way of Making Europe”. The computations were made possible thanks to generous grants of computing time from the Barcelona Supercomputing Centre (references FI-2018-1-0037 and AECT-2020-2-0005) and from the supercomputing centre of the Universitat Politècnica de València. The work of M. Oberlack was supported by the German Research Foundation (DFG) under grant OB96/39-1. FAA is partially funded by the GVA/FEDER project ACIF2018. S.H. and M.J.P.Q. acknowledge support from project PID2021-128676OB-I00 by MCIN/AEI/10.13039/501100011033 and ERDF “A Way of Making Europe”. R.V. acknowledges financial support from the ERC grant 2021-CoG-101043998 (DEEPCONTROL).
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Three-dimensional ESRGAN for super-resolution reconstruction of turbulent flows with tricubic interpolation-based transfer learning

  • Yu, Linqui
  • Yousif, Mustafa Z.
  • Zhang, Meng
  • Vinuesa, Ricardo
  • Lim, Hee-Chang
  • Hoyas, Sergio|||0000-0002-8458-7288
[EN] Turbulence is a complicated phenomenon because of its chaotic behavior with multiple spatiotemporal scales. Turbulence also has irregularity and diffusivity, making predicting and reconstructing turbulence more challenging. This study proposes a deep-learning approach to reconstruct three-dimensional (3D) high-resolution turbulent flows from spatially limited data using a 3D enhanced super-resolution generative adversarial networks (3D-ESRGAN). In addition, a novel transfer-learning method based on tricubic interpolation is employed. Turbulent channel flow data at friction Reynolds numbers R e tau = 180 and R e tau = 500 were generated by direct numerical simulation (DNS) and used to estimate the performance of the deep-learning model as well as that of tricubic interpolation-based transfer learning. The results, including instantaneous velocity fields and turbulence statistics, show that the reconstructed high-resolution data agree well with the reference DNS data. The findings also indicate that the proposed 3D-ESRGAN can reconstruct 3D high-resolution turbulent flows even with limited training data., This work was supported by "Human Resources Program in Energy Technology" of the Korea Institute of Energy Technology Evaluation and Planning (KETEP), granted financial resource from the Ministry of Trade, Industry & Energy, Republic of Korea (No. 20214000000140). In addition, this work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (No. 2019R1I1A3A01058576). This work was also supported by the National Supercomputing Center with supercomputing resources including technical support (No. KSC-2022-CRE-0282). R.V. acknowledges the financial support from the ERC Grant No. 2021-CoG-101043998, DEEPCONTROL. S.H. was funded by Contract No. PID2021-128676OB-I00 of Ministerio de Ciencia, innovacion y Universidades/FEDER.