The SKARV project held its fourth meeting in Trondheim on 1 October. A year ago, the project showed that shifting blade position by about 20 degrees at the moment a bird crosses the rotor plane could prevent a collision. As the project continues, the questions are becoming sharper: exactly when should the turbine react, and what is the cost of the manoeuvre?
A narrow window
Radar data analysed by project partner NINA show that a bird’s flight path can be predicted a few seconds ahead, but not much further. Beyond about ten seconds, the gap between prediction and actual position is so large that it makes the prediction of little value for avoidance purposes.
At the other end, a large rotor needs time to respond. Achieving the 20-degree shift that avoids a collision takes 4 seconds longer on a 22 MW turbine than on a 3.4 MW one, because of the rotor’s inertia. The avoidance manoeuvre has to fit between those two limits.
Deciding when to act
William Jussiau, postdoctoral fellow at NTNU’s Department of Engineering Cybernetics, presented the simulation architecture that handles this decision. SKARV is designed as a layer on top of the turbine’s existing control system rather than a replacement for it.
A supervisor continuously estimates the risk that an approaching bird will hit a blade if the turbine keeps operating normally. The estimate combines two things: the probability that the bird will cross the rotor plane at all, and, given all uncertainties, how close to a blade it is likely to pass. The idea is that the turbine reacts when the risk exceeds a set threshold.
Shortly before a potential collision comes a commit point: once a manoeuvre is engaged, it runs to completion. The project is testing several control strategies of increased sophistication, from a fixed speed reduction to one that also accounts for wind speed and load on the tower and blades.
The project is now running a Monte Carlo campaign, which will continue over the coming months. It consists of simulating a large number of encounters in which bird type, wind conditions, turbulence and sensor noise are varied randomly from one turn to the next, so that the system is tested against the full range of situations it could meet in reality. Running the same encounters with and without the avoidance system shows where the concept works and where it does not. The results will also help set the parameters of the supervisor, such as the risk threshold at which the turbine reacts.
What the manoeuvre costs
Each avoidance event is a sudden change in the loads on the turbine. Valentin Chabaud, of SINTEF Energy Research, pointed out that fatigue damage is driven disproportionately by large load amplitudes, so a short extreme event can consume as much of a turbine’s lifetime as a long period of normal operation.
A first simulation suggests that the damage can fall mainly on the tower and drivetrain rather than the blades, and is of the same order of magnitude as the damage that can be caused by curtailment events that wind farms already perform for grid support. The estimate is preliminary and based on a single load case, but it is enough to make turbine lifetime an explicit design criterion for SKARV. The comparison that matters is with the alternative: an avoidance manoeuvre is less damaging than altogether stopping the turbine.
From design to demonstration
SKARV uses two simulation setups. Both chain together the same layers: a model of the bird’s flight path, the collision avoidance control, the turbine’s own controller and a model of the turbine itself.
The first setup is used to design the SKARV control. It pairs an in-house turbine controller with a mid-fidelity turbine model. This is where the work on deciding when to act was done. The second setup is the one that will be used for testing and demonstration: a typical industry controller, represented by the ROSCO controller, and a high-fidelity turbine model. The demonstrations will use the software Ashes from Simis, an NTNU spin-off company.
Project lead Paula Bastos Garcia Rosa presented work that sits between the two. The team ran the in-house controller, which follows the basic structure of ROSCO, on the high-fidelity turbine model to check its performance and see whether it needs adapting to better reflect the dynamics of industry controllers.
So far, the control design has used a simplified model of bird flight paths. Over the coming months, this will be replaced by NINA’s bird prediction model, and real bird tracking data will be added.
"The project is on schedule. We have a simulation set-up for designing the control system, and we are now checking what it takes to bring it closer to the controllers used in industry. Over the coming months, we will bring in NINA's prediction model for bird flight paths and continue the simulations that will help us tune the alert system," says Paula Bastos Garcia Rosa.
