Join on-line via Zoom: https://wse.zoom.us/j/435449376
“Generalizing the Coupled Wake Boundary Layer Model for Wind Farm Power Prediction”
Presented by GENEVIEVE STARKE
(Advisers: Profs. Meneveau & Gayme)
During recent years, wind has continued to grow as an electricity provider in the United States. As interest in wind has grown, a need has arisen to be able to accurately predict the effect of turbines on the wind and how this impacts the overall power of the wind farm. Large-scale simulations can predict this accurately, however, they are too expensive and too slow to be used in real-time applications. To achieve faster results, we are working on developing and applying reduced-order models of wind farms that accurately capture key dynamics in the wind farm. Here we present a model that combines two reduced-order models that depend on different scales to improve the overall prediction of the power of the wind farm. The first model is called a wake model, and represents the individual turbines and their effect on the wind. The second model, called the top-down model, works on a larger scale and represents the wind farm as an impediment to the wind in the atmospheric boundary layer, which describes the interaction between the wind in the atmosphere and the ground. This model gives average quantities over the wind farm that are then matched to those obtained from the wake model to obtain a combined prediction of the power. To determine the area over which the average quantities are calculated, the model uses Voronoi tessellation to establish regions of the flow that belong to each turbine. This enables the models to be applied to each turbine and the area around that turbine individually, which allows application to nonuniform wind farms. As a result of our average treatment of the farm, the boundary layer caused by the wind farm in the atmospheric boundary layer is defined to start at the initial line of freestream turbines, rather than on an individual turbine basis. The coupled wake boundary layer model is validated using data from LES simulations for various wind directions from the National Renewable Energy Laboratory, and found to reproduce trends in overall power produced as a function of the inflow direction of the wind.
“LES of Laminar-Turbulent Transition with Turbulent/Non-Turbulent Classification”
Presented by GHANESH NARASIMHAN
(Advisers: Profs. Meneveau & Zaki)
The laminar-turbulent transition in boundary layers and channel flow is usually classified as either orderly (natural) or bypass transition. Natural transition starts with amplification of Tollmien-Schlichting (TS) instability waves which subsequently undergo secondary instability and ultimately break down to turbulence. Often, however, when background disturbances are present, bypass transition takes place instead. It involves initial linear amplification of streamwise perturbations through lift-up mechanism, followed by non-linear evolution and break down into turbulence spots. Direct Numerical Simulation (DNS) of these transition processes is computationally expensive. Therefore, Large Eddy Simulation (LES) is used as an alternative to simulate the laminar-turbulent transition. Wall-Resolved LES (WRLES) using dynamic Smagorinsky Sub-Grid Scale (SGS) model simulates this transition which is comparable to the DNS. However, resolving the inner layer in WRLES demands more grid points thereby making the WRLES of transition computationally expensive. Hence, LES with wall-modeling which models the inner layer rather than resolving it, is used to further decrease the computational cost. Using Wall-Modeled LES (WMLES) to simulate transition is again difficult as the wall model assumes the flow is fully developed. Since a transitional flow has features from both laminar and turbulent states, LES equations with wall-modeling could still be applied in the turbulent regions. Therefore, it is essential to identify the Turbulent/Non-Turbulent (T/NT) regions in the flow. To this end, self-organizing map (SOM), an unsupervised machine learning algorithm is used for the T/NT classification in a transitional channel flow. Firstly, the natural transition is simulated by the WMLES with SOM of the time evolution of channel flow with 2D TS wave and oblique 3D waves as the initial perturbation. Secondly, the case of bypass transition in channel flow is simulated by performing WMLES with SOM of the time evolution of a three-dimensional localized perturbation. The time evolution of the friction Reynolds number from the LES is compared to the DNS. Both the transition scenarios are well predicted by the LES with significantly less computational resources. Thus, the WMLES with T/NT classification using SOM is proposed as a new technique for modeling transitional flows.