The numerical simulation of particle-laden turbulent flows has seen significant advances in recent years, yet remains a challenging problem motivated by processes spanning a wide range of scientific and industrial applications. We have developed an in-house code (lbe3d) for DNS (Direct Numerical Simulation) of turbulent flows with finite-size particles. The student will analyze the energy budget of particle-laden homogeneous isotropic turbulence using the DNS data. This involves computing production, dissipation and convective terms from the transport equation for turbulent kinetic energy. This project will investigate how finite size particles affect the energy budget of turbulent flows.
Federico Toschi, Abhineet Gupta
Er valt pas regen uit de wolken als de druppels groot genoeg zijn. De druppels groeien doordat de turbulente wind in de wolken ze laat botsen. Dit wordt onderzocht in een experiment waarin druppels zichtbaar gemaakt worden door ze te laten oplichten door laser-geinduceerde fosforescentie, en ze dan te volgen met snelle camera's. De uitdaging van dit experiment is om met behulp van beeldbewerking de spontane verdichting van de wolk op de allerkleinste schaal (100 μm) te kunnen waarnemen. De vraag is ook welke grootheden gemeten moeten worden. Bij het beantwoorden ervan worden numerieke simulaties uitgevoerd.
Contact:Willem van de Water
Measurements of atmospheric turbulence involve either the positioning of fixed probes on towers or flying probes mounted on airplanes. Here we want to investigate if drones could potentially be used as more flexible instruments to investigate atmospheric turbulence from inside the flow. Drones come in different sizes and, particularly the smaller ones, can be strongly affected by turbulent fluctuations, even at the smaller scales.
The question is whether we can use a small drone as a probe of atmospheric turbulence, for example by studying the signals of its accelerometers. The other question is how to design drone software so that it can cope more effectively with turbulence. We will use the TU/e wind tunnel which has a unique facility to generate "tailored" turbulence using a computer-controlled active grid.
In this project you will learn about the fundamental properties of turbulent flows, boundary layers, experimental techniques for fluid dynamics research, including the programming of active grids, the running of wind tunnels and …. drones flying.
Willem van de Water, Federico Toschi
The goal of Machine Learning (ML) is to get computers learning without being explicitly programmed. Machine Learning, especially its sub-field of deep neural networks and reinforcement learning are constantly in the news nowadays due to the explosive number of successes of ML, e.g. DeepMind beating the world champion at playing Go (March 2016), Tesla teaching their cars to automatically pilot, unsupervised learning able to learn the artistic features of master paintings etc.
In this project we will employ ML for efficient learning of basic turbulence features. Can unsupervised ML techniques like stacked Auto-encoders, Restricted Boltzmann Machines or Deep Belief Networks which have proved to be very successful in other domains, be able to automatically extract the key features of turbulent flows through dimensionality reduction? Can they provide theoretical hints on the numbers of relevant degrees of freedom for 3D homogenous and isotropic turbulence? Can this possibly be smaller than the classical estimate, (L/η)3? In this project you will learn about the most fundamental physical properties of turbulence together with basic knowledge of current state of the art deep learning techniques.
Pinaki Kumar, Federico Toschi