The design of bigger and more powerful wind turbines needs to incorporate complex modelling.
How numerical simulations can optimise wind turbine function.
This article is part of our series on High value manufacturing: dealing with the unknown
Data and modelling techniques can be powerful tools to incorporate when designing high value applications. However, data with different provenance, and models from numerous sources all have different types and sizes of uncertainty which need to be accounted for when trying to make useful estimates of uncertainty in the final application.
A case in point: offshore energy
Wind energy is one of the most utilised sources of renewable energy, and has experienced a tremendous growth in the past few decades. But driving growth means bigger wind turbines and wind farms with higher capacities, creating new challenges for the design and construction process such as higher wind shear and turbulence loads. Highly advanced experimental and computational methods are needed to cope with the challenges.
The process
Numerical simulations have become integral to the design and optimisation of wind turbine farm layouts in recent years. The use of numerical simulation techniques for a-priori design of wind turbine array layouts and control strategies is highly desirable – if the simulation results are reliable.
The challenge
However, these models have their own challenges and uncertainties, such as dealing with flow complexities, dealing with large datasets and representing various scales of motion. And while models for the use of Computational Fluid Dynamics (CFD) in aerospace, automotive and civil engineering have been well validated, this is not the case for wind energy.
In addition, the empirical parameters for turbine wakes have been lifted directly from standard aerospace-scale models without great consideration of the tuning that has been applied – or in some cases the formal bounds of applicability. The industry is now inserting Light Detection and Ranging (LIDAR) and other direct measurement systems into large wind farm arrays to provide in-situ and in-service data feeds, but integration with design simulation models is virtually non-existent.
Uncertainty Quantification and Management considerations
The physical parameters to be included as input to the simulation process include location, layout, type and control laws for the turbines. These parameters are all subject to uncertainty on input. The definition of many of these input uncertainties is not well developed.
The approach
The idea of this use case is to quantify the validity of the numerical approaches for wind turbine array simulations and identify uncertainties in such high fidelity CFD simulations for the use in wind energy industry. Specifically, we aim to analyse wind turbine wakes of a scaled model (as a surrogate for wind farm data) and validate the results using raw experimental data. The rational is that accurate wind farm wake analysis leads to uncertainty reduction and better wind farm layouts and control strategies.
To be able to validate the CFD simulations against LIDAR data, measurements need to be analysed to identify outliers and provide uncertainty bounds on the time series data. To calibrate to the data, a Bayesian optimisation problem was formulated and applied to a surrogate of the computationally expensive simulator.
The learning
“Validating simulation models with raw in-service and experimental data is a key enabler that is starting to attract a lot of attention. We had not considered using Gaussian Process modelling to inform parameter selection in fluid dynamics until the team demonstrated a system using our own data … We are now looking to take this forward as part of an integrated service.” David Standingford, Director Zenotech
For more information on how cross-sector technologies can help drive efficiency in the offshore wind sector, read our news about the Offshore Wind Innovation Exchange (OWiX) Call for Ideas.
Read more use cases in this series
Developing early stage designs for compliance with flight performance requirements
Visualisation of uncertainty to aid the decision making process
Applying inverse mapping across coupled and feedback loop processes
More information about upcoming UQ&M SIG group activities can be found here.