Refine
Document Type
- Doctoral thesis (2)
- Scientific article (1)
Has Fulltext
- yes (3)
Is part of the Bibliography
- no (3)
Language
- English (3)
Keywords
- Large Eddy Simulation (3) (remove)
Physics-informed neural networks (PINN) are machine-learning methods that have been proved to be very successful and effective for solving governing equations of fluid flow. In this work we develop a robust and efficient model within this framework and apply it to a series of two-dimensional three-component (2D3C) stereo particle-image velocimetry datasets, to reconstruct the mean velocity field and correct measurements errors in the data. Within this framework, the PINNsbased model solves the Reynolds-averaged-Navier-Stokes (RANS) equations for zeropressure-gradient turbulent boundary layer (ZPGTBL) without a prior assumption and only taking the data at the PIV domain boundaries. The TBL data has different flow conditions upstream of the measurement location due to the effect of an applied flow control via uniform blowing. The developed PINN model is very robust, adaptable and independent of the upstream flow conditions due to different rates of wall-normal blowing while predicting the mean velocity quantities simultaneously. Hence, this approach enables improving the mean-flow quantities by reducing errors in the PIV data. For comparison, a similar analysis has been applied to numerical data obtained from a spatially-developing ZPGTBL and an adverse-pressure-gradient (APG) TBL over a NACA4412 airfoil geometry. The PINNs-predicted results have less than 1% error in the streamwise velocity and are in excellent agreement with the reference data. This shows that PINNs has potential applicability to shear-driven turbulent flows with different flow histories, which includes experiments and numerical simulations for predicting high-fidelity data.
A front-tracking algorithm for large-eddy simulation (LES) is developed to untangle the numerical and physical contributions to entrainment in stratocumulus-topped boundary layers. The front-tracking algorithm is based on the level set method. Instead of resolving the cloud-top inversion, it is represented as a discontinuous interface separating the boundary layer from the free atmosphere. The location of the interface is represented as an isosurface of an evolving marker function the evolution of which is governed by an additional transport equation. The algorithm has been implemented in an existing LES code based on the anelastic approximation of the Navier-Stokes equations.
The original LES algorithm is verified against direct-numerical simulation (DNS) data of an idealized two-dimensional cloud-top mixing layer. For this, the subgrid-scale model of the LES code was replaced by a constant molecular viscosity in order to focus on numerical errors only. A grid convergence study confirmed the anticipated global second-order rate of convergence and the convergence to the DNS solution. The slower convergence of the LES code as compared to the higher-order DNS yielded leading-order errors in the mixing layer growth at the coarsest resolutions, which were finer still than typical LES resolutions.
The front-tracking algorithm is verified by LESs of two different convective atmospheric boundary layers: the smoke cloud, a solely radiatively driven boundary layer, and a stratocumulus-topped boundary layer based on data from the DYCOMS II field study. Specifying zero entrainment, it was shown that entrainment in LES can be controlled effectively by the front-tracking algorithm. The algorithm drastically reduces entrainment errors and reduces dependencies of the solution to numerical parameters such as the choice of flux-limiters and grid resolution.
The environmental emergency has led to the development of new combustion technologies. In this context, flameless combustion (FC in this manuscript) offers the prospect of a less polluting and more efficient technology. In FC, combustion is strongly diluted with recirculated burnt gases. Consequently the oxygen content is reduced and temperature peaks are smoothed, yielding reduced heat release. These conditions dramatically reduce the conditions of NO pollutant formation and increase the efficiency of the combustion process. Being FC a relatively new technology, it still needs optimization and R&D, which can be expensive and time consuming. Potentially, CFD can reduce both the financial costs as well as the R&D projects length. The context in which this thesis is inserted is exactly the numerical modeling of FC, by using Large Eddy Smulations for its better prediction of the turbulent ternary mixing (fuel - burnt gases -air), compared to RANS. This work has been divided into two main parts. In the first, combustion in FC has been investigated by means of a new tabulated combustion model initially written in the context of the EC-KIAI project and developed and adapted to FC in this thesis. The model uses diluted homogeneous reactors DHR to simulate FC and it was developed to account for under adiabatic enthalpy losses and the ternary mixing typical of FC. The model was firstly validated on a non-premixed flame academical configuration called Flame D and subsequently on a real FC combustor from the work of Verissimo et al. The results obtained for these configurations are quite correct although some discrepancies in CO prediction are observed. In the second part of the thesis, the NO pollutant modeling in FC is investigated. With this aim, the Diffusion Flame - NO relaxation approach DF-NORA was developed. It consists in tabulating the NO relaxation towards equilibrium of the NO source term in a flamelet structure. As done in the first part, the model was first validated on Flame D and then employed in a real FC configuration. Results are quite satisfactory in both config- urations. The encouraging results obtained in this work open the possibility of applying the proposed developments to real industrial configurations in the future.