Machine Learning for fluid velocimetry and visualization

In the recent past, serval Convolutional Neural Networks (CNNs) have been developed to process Particle Image Velocimetry (PIV) and Background Oriented Schlieren (BOS) recordings. Despite showing promising results, several bottlenecks prevent their spread in the community. For instance, obtaining training data from Direct Numerical Solutions (DNS) for many flow regimes and geometry can be prohibitive, in terms of computational cost. Moreover, the architectures that have been proposed are very large, and thus require powerful GPU hardware and long computational times to process the recordings.

At our lab, we have developed an innovative training strategy which relies on kinematic motion. More precisely, we generate training sets in which particles move according to a random displacement field. This allows us to generate training data with a large variation of spatial scales at little computational cost, thus allowing for a higher generalization of the CNN.


Secondly, we also developed a novel CNN architecture, namely Lightweight Image Matching Architecture (LIMA), that has a considerably lower number of parameters compared to existing networks.  LIMA can process images faster than other ML and classical approaches at comparable or higher accuracy. The network size is small enough that it can be compiled and deployed to embedded GPU devices, paving the way to the development of autonomous measurement systems.

Simultaneous PIV–LIF measurements using RuPhen and a color camera

Particle image velocimetry (PIV) and laser-induced fluorescence (LIF) are currently state-of-the-art, on-intrusive measurement techniques that help in our understanding of heat and momentum transfer in thermal fluid flow applications.


We present a simplified, integrated PIV–LIF system which simultaneously measures velocity and temperature fields in aqueous flows by means of a novel fluorescent dye (RuPhen), a low-cost continuous-wave diode laser, and a single color camera. We demonstrate that RuPhen is well-suited for this approach due to its peak absorption at 450 nm, peak emission at 605 nm, and a strong temperature-dependent emission with a sensitivity coefficient of 4%∕◦C . The large Stokes shift between excitation (which also includes Mie scattering of the flow tracers) and the emission facilitates the handling of the signal components in the RGB channels of the camera. To correct the recordings for laser power fluctuations, we propose a novel method that jointly employs two photoluminescence signals. We provide a spectral characterization of the dye at different temperatures and discuss the choice of each component for our measurement system. We demonstrated the potential of our approach by two experimental test cases that focus on thermally driven and turbulent flow regimes.

Droplet dispersion in airborne pathogen transmission

Airborne pathogen transmission has garnered particular attention in the years following the COVID-19 pandemic. Mathematical models to estimate the risk of infection vary from computationally intensive Direct Numerical Simulations to simplified well-mixed models of indoor air. In the just compromise between accuracy and complexity, we propose a particle-based methodology to study transmissibility of airborne infectious agents.

Upon any exhalation (such as breathing, talking, sneezing, coughing), a turbulent multiphase gas cloud laden with droplets is emitted. How many of these droplets will linger inside the room and how many will fall ballistically to the ground? Which is the set of environmental parameters that favors one outcome against the other? Droplet fate is contingent on a multitude of environmental factors which govern droplet transport, such as relative humidity and turbulence. We develop a stochastic Lagrangian framework to study the dynamics of such respiratory droplets as they travel in air, including a simple model to represent indoor turbulence. How do the timescales of the competing processes such as gravity, evaporation or turbulent dispersion compare, and which is of the highest importance for the aerosol and droplet lifetime?


The virological properties of airborne infectious agents, such as pathogen survival, are coupled to the fluid dynamics to yield the spatiotemporally varying infection risk. We explore qualitative yet informative scenarios to illustrate the relative importance of often-overlooked effects, such as the dependency of the viral load concentration on the size of the emitted droplets. As more data becomes available, this versatile framework can be updated and adapted to offer more insight into this ever-relevant topic at the nexus of virology and fluid dynamics.


The model is developed in Python and is run using CSCS resources.

Experimental characterization of droplet emission by turbulent puffs

The COVID-19 pandemic showed the disrupting potential of viruses, which can rapidly and ubiquitously spread among a vast portion of the population leading to tragic medical, social, and economic consequences. Airborne transmission by contagious droplets can be a primary way of infection. Yet, the risk estimation relies on fluid dynamics models, which are only partially validated.

At our lab, we have developed a Respiratory Droplet Simulator that can precisely and repetitively mimic turbulent puffs and droplet emission from human subjects. The RDS paves the way to analyse the droplet spread at different time and spatial scales, also by means of different measurement techniques. For example, Particle Image Velocimetry (PIV) or 3D Particle Tracing Velocimetry (3D-PTV) can be used to quantify the characteristics of the flow field at for various respiratory activities in indoor environments. Shadowgraphy can be used to quantify droplet sizes and velocity at different locations in the domain to obtain relevant statistical information on droplet spread and evaporation.