The Project

SPANDRELS (Sparse and parsimonious event-based flow sensing) is a 5-years ERC Consolidator Grant project (2,0M€ budget) being carried out at the Experimental Aerodynamics and Propulsion Lab , at Universidad Carlos III de Madrid.

Motivation

The closed-loop control of unsteady turbulent flows requires efficient strategies to sense the flow state. Despite the challenge posed by the non-linearities and the large range of scales of turbulent flows, their ubiquitous nature motivates unabated research efforts.
Over the last years, we have developed linear and non-linear flow estimation tools, with relevant laboratory applications. Nevertheless, the state of the art requires an intractable number of sensors, making the data acquisition and analysis unfeasible in a practical scenario. Moreover, the current paradigm of flow control requires continuous sensing and action in time, leading to very large data rates.
Strangely, this seems at odds with what nature does: insects estimate the flow surrounding them with a few event-based sensors embedded in their wings.
Algorithms for event-based signal processing avoid aliasing without the need for high-frequency periodic sampling, reducing the amount of data needed to estimate complex temporal series: this could enable flow estimation with easy-to-handle and cheap-to-compute data. Furthermore, our recent findings show that many complex flows can be represented on low-dimensional manifolds. The availability of a reduced set of coordinates for state representation is a key enabler for the choice of a sparse set of sensors in space.
This project will develop a novel framework for the estimation of turbulent and unsteady flows coupling manifold learning and event-based sensors. Tackling selected relevant laboratory problems, with and without control, we will reduce problem dimensionality and represent turbulent unsteady flows on low-dimensional manifolds, identify parsimonious methods for sensor choice and location in complex flows, and define a theoretical framework for turbulent-flow measurements from event sensors. Such a framework will be a key enabler for flow control and will open a novel research path in fluid mechanics.

Objectives and roadmap

The main objective is the development of sparse and parsimonious event-based sensing for the feedback-loop control of turbulent and unsteady flows. To achieve this, we will first have to resolve the following issues:
1.- How can we discover low-dimensional manifolds for actuated turbulent flows?
2.- Where is the best position and which is the minimum number for efficient and parsimonious placement of sparse sensors?
3.- How do we design neural-inspired filters that are specific for certain flow events?
4.- How do we estimate the full state of a flow from a sparse set of event signals?
This objective will be reached through the following intermediate stepping stones:
• Identify low-dimensional manifolds of flows with control.
• Develop methods for sparse sensor placement.
• Explore analogue filters for event detection within a flow.
• Identify a theoretical framework for event-based flow estimation.

Workplan

To reach the proposed objectives ferstly we will employ numerical simulations to obtain preliminary datasets or to enable a better design of the experiments, but the project will focus on experiments with actuated turbulent flows. To produce datasets, we will leverage our expertise in flow diagnostics, including advanced optical methods such as PIV and IR thermography. We will focus on three test cases with increasing levels of complexity (increased expected manifold rank):

1.-The first and simplest case considers a 1:20 scaled ground transportation system (truck) model tested in the UC3M wind tunnel. The idealized model (without wheels or moving ground) will operate at Reynolds numbers above 10⁵, producing a wide range of turbulent wake scales. Flow control will be applied through momentum injection, and sensing will be carried out using a force balance and multiple high-frequency pressure sensors embedded in the model.
2.-A greater level of complexity will be achieved with a turbulent jet, studied in the UC3M anechoic chamber. The jet flows, studied at a bulk Reynolds number greater than 104, based on jet diameter and velocity, will exhibit a wide range of turbulent scales. Actuation, to control mixing and/or noise will be performed analogously to recent flow control studies (Zhou et al., 2020) employing both nozzle modification and momentum injection, while flow sensing will be implemented using a microphone-array acoustic camera (Merino-Martínez et al., 2020).
3.- The most complex configuration focuses on a controlled turbulent boundary layer in the UC3M water tunnel at a friction Reynolds number of 1000. Flow measurements will use high-speed infrared thermography coupled with a heated thin-foil heat transfer sensor. Actuation will be achieved through pulsed momentum injection.