Hybrid deep-learning POD-based parametric reduced order model for flow around wind-turbine blade
Tabib, Mandar; Tsiolakis, Vasileios; Pawar, Suraj; Ahmed, Shady E.; Rasheed, Adil; Kvamsdal, Trond; San, Omer
Peer reviewed, Journal article
Published version
View/ Open
Date
2022Metadata
Show full item recordCollections
- Publikasjoner fra CRIStin - SINTEF AS [5969]
- SINTEF Digital [2556]
Original version
Journal of Physics: Conference Series (JPCS). 2022, 2362, 012039. 10.1088/1742-6596/2362/1/012039Abstract
In this study, we present a parametric, non-intrusive reduced order modeling (NIROM) framework as a potential digital-twin enabler for fluid flow around an aerofoil. A wind turbine blade has its basic foundation in the aerofoil shape. A faster way of understanding dynamic flow changes around the aerofoil-shaped blade can help make quick decisions related to wind-turbine operations and lead to optimal aerodynamic performance and power production. In this direction, a case study involving the application of the NIROM methodology for flow prediction around a NACA 0015 aerofoil is considered. The Reynolds number (Re) is the varying parameter, ranging from 320 000 to 1.12 million and high-fidelity CFD simulations are performed to generate the database for developing the NIROM. The aforementioned NIROM framework employs a Grassmann manifold interpolation approach (GI) for obtaining basis functions corresponding to new values of the parameter (Reynolds number), and exploits the time series prediction capabilities of the long short-term memory (LSTM) recurrent neural network for obtaining temporal coefficients associated with the new basis functions. The methodology involves: (a) an offline training phase, where the LSTM model is trained on the modal coefficients extracted from the sampled high-resolution data using the proper orthogonal decomposition (POD), and (b) an online testing phase, where for the new parameter value, the corresponding flow field is obtained using the GI-modulated basis functions for new parameter and the LSTM-predicted temporal coefficients. The NIROM-approximated flow predictions at new parameters have been compared to the high-dimensional full-order model (FOM) solutions for the high-Re aerofoil case and for a low-Re number wake vortex merger case in order to put the performance of NIROM in perspective. The results indicate that the NIROM framework can qualitatively predict the complex flow scenario around the aerofoil for new values of Reynolds number, while it has quantitatively shown that the LSTM predictions improve with the enrichment of the training space. For the low-Re vortex merger case, NIROM works very well. Thus, it can be deduced that there is scope and potential for continued research in NIROMs as digital twin enablers in wind energy applications.