When a gas turbine combustor is manufactured, small geometrical differences between production runs are unavoidable. These geometrical differences potentially influence performances such as service life, lighting, lean blow off, pollutant emissions. To ensure performances, manufacturing tolerances are kept tight enough so as to neglect their impact. However, loosening some key manufacturing tolerances can result in lower manufacturing costs, while maintaining combustion performances at a satisfactory level. The present study aims at challenging some cost-driving manufacturing tolerances of Safran Helicopter Engines new turboshaft combustor, using state-of-the-art 3D numerical tools. The objective is to guarantee robust and long lasting combustor performances at a reasonable manufacturing cost. To this purpose, an optimized Latin Hypercube Sampling of 30 geometrical configurations around the nominal geometry has been simulated with a 3D LES solver. Simulation parameters had to be chosen to keep computation costs manageable and to have a high representability when comparing geometries. These aspects are checked using theoretical considerations and comparisons with both experimental data and previously validated simulations. The temperature field at the combustion chamber exhaust has been analyzed in detail using metamodeling. In particular, the Radial Temperature Distribution Factor (RTDF) at the combustor outlet appears to be robust. Such large designs of numerical experiments (DonE) are computationally expensive but they provide a reliable tool to assess the robustness of a combustor design, and define optimized manufacturing tolerances.
In a gas turbine, the combustor is feeding the turbine with hot gases at a high level of turbulence which in turns strongly enhances the heat transfer in the turbine. It is thus of primary importance to properly characterize the turbulence properties found at the exit of a combustor to design the turbine at its real thermal constraint. This being said, real engine measurements of turbulence are extremely rare if not inexistent because of the harsh environment and difficulty to implement experimental techniques that usually operate at isothermal conditions (e.g. hot wire anemometry). As a counterpart, high fidelity unsteady numerical simulations using Large Eddy Simulations (LES) are now mature enough to simulate combustion processes and turbulence within gas turbine combustors. It is thus proposed here to assess the LES methodology to qualify turbulence within a real helicopter engine combustor operating at take-off conditions. In LES, the development of turbulence is primarily driven by the level of real viscosity in the calculation, which is the sum of three contributions: laminar (temperature linked), turbulent (generated by the sub-grid scale model) and artificial (numerics dependent). In this study, the impact of the two main sources of un-desired viscosity is investigated: the mesh refinement and numerical scheme. To do so, three grids containing 11, 33 and 220 million cells for a periodic sector of the combustor are tested as well as centred second (Lax-Wendroff) and third order (TTGC) in space schemes. The turbulence properties (intensity and integral scales) are evaluated based on highly sampled instantaneous solutions and compared between the available simulations. Results show first that the duration of the simulation is important to properly capture the level of turbulence. If short simulations (a few combustor through-times) may be sufficient to evaluate the turbulence intensity, a bias up to 14% is introduced for the turbulence length scales. In terms of calculation set-up, the mesh refinement is found to have a limited influence on the turbulence properties. The numerical scheme influence on the quantities studied here is small, highlighting that the employed schemes dissipation properties are already sufficient for turbulence characterization. Finally, spatially averaged values of turbulence intensity and lengthscale at the combustor exit are almost identically predicted in all cases. However, significant variations from hub to tip are reported, which questions the pertinence to use 0-D turbulence boundary conditions for turbines. Based on the set of simulations discussed in the paper, guidelines can be derived to adequately set-up (mesh, scheme) and run (duration, acquisition frequency) a LES when turbulence evaluation is concerned. As no experimental counterpart to this study is available, the conclusions mainly aim at knowing the possible numerical bias rather than commenting on the predictivity of the approach.
Nowadays, models predicting soot emissions are neither able to describe correctly fine effects of technological changes on sooting trends nor sufficiently validated at relevant operating conditions to match design office quantification needs. Yet, phenomenological descriptions of soot formation, containing key ingredients for soot modeling exist in the literature, such as the well-known Leung et al. model (Combust Flame 1991). However, when blindly applied to aeronautical combustors for different operating conditions, this model fails to hierarchize operating points compared to experimental measurements. The objective of this work is to propose an extension of the Leung model over a wide range of condition relevant of gas turbines operation. Today, the identification process can hardly be based on laboratory flames since few detailed experimental data are available for heavy-fuels at high pressure. Thus, it is decided to directly target smoke number values measured at the engine exhaust for a variety of combustors and operating conditions from idling to take-off. A large eddy simulation approach is retained for its intrinsic ability to reproduce finely unsteady behavior, mixing, and intermittency. In this framework, The Leung model for soot is coupled to the thickened flame model (TFLES) for combustion. It is shown that pressure-sensitive laws for the modeling constant of the soot surface chemistry are sufficient to reproduce engine emissions. Grid convergence is carried out to verify the robustness of the proposed approach. Several cases are then computed blindly to assess the prediction capabilities of the extended model.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.