# Professor Nicholas Hills

### Biography

### Biography

Prof. Nick Hills is the Head of Department of Mechanical Engineering Sciences. He holds a Chair in Computational Engineering at the University of Surrey, and is also the Director of the Rolls-Royce University Technology Centre in Thermo-Fluid Systems. He previously held a Royal Academy of Engineering / Rolls-Royce Chair.

His research has focussed on CFD development, particularly for massively parallel simulations (including leading the UK Applied Aerodynamics Consortium with time on the national super-computing facilities), multi-physics models and their application to internal air systems for gas turbine engines. Research funding has been through EPSRC, InnovateUK, EU and direct industrial support.

### My publications

### Publications

?) up to 4.6 ! 10

. Nozzle performance is characterised by the nozzle discharge coefficient (C

), nozzle velocity coefficient (?·) and cooling air delivery temperature. Two commonly used eddy viscosity models are employed for the study, the standard º-µ and Spalart-Allmaras models with wall functions. Both models give very similar results for C

and · and are in reasonable agreement with available experimental data. Effects of nozzle or vane number and sealing flow have been analysed. The cascade vanes perform slightly better than the aerodynamically designed drilled nozzles but the final design choice will depend on other component and manufacturing costs. An elementary model is presented to separate temperature losses due to the nozzle, stator drag and sealing flow. Copyright © 2011 by Rolls-Royce plc.

accurately to predict the interaction of the mainstream flow with the secondary air system cooling

flows using computational fluid dynamics (CFD) with Reynolds-averaged Navier?Stokes

(RANS) turbulence models have proved difficult. In particular, published results from RANS

models have over-predicted the sealing effectiveness of the rim seal, although their use in this

context continues to be common. Previous studies have ascribed this discrepancy to the failure to

model flow structures with a scale greater than the one which can be captured in the small-sector

models typically used. This article presents results from a series of Large-Eddy Simulations (LES)

of a turbine stage including a rim seal and rim cavity for, it is believed by the authors, the first

time. The simulations were run at a rotational Reynolds number Re ¼ 2.2 106 and a main

annulus axial Reynolds number Rex ¼ 1.3 106 and with varying levels of coolant mass flow.

Comparison is made with previously published experimental data and with unsteady RANS simulations.

The LES models are shown to be in closer agreement with the experimental sealing

effectiveness than the unsteady RANS simulations. The result indicates that the previous failure

to predict rim seal effectiveness was due to turbulence model limitations in the turbine rim seal

flow. Consideration is given to the flow structure in this region.

K

be complex and attempts accurately to predict the interaction of

the mainstream flow with the secondary air system cooling flows

using CFD with RANS turbulence models have proved difficult.

In particular, published results from RANS models have overpredicted

the sealing effectiveness of the rim seal, although their

use in this context continues to be common. Previous authors

have ascribed this discrepancy to the failure to model flow structures

with a scale greater than can be captured in the small sector

models typically used. This paper presents results from a series

of Large-Eddy Simulations (LES) of a turbine stage including a

rim seal and rim cavity for, it is believed by the authors, the first

time. The simulations were run at a rotational Reynolds number

Re¸ = 2.2 × 106 and a main annulus axial Reynolds number

Rex = 1.3 × 106 and with varying levels of coolant mass

flow. Comparison is made with previously published experimental

data and with unsteady RANS simulations. The LES models

are shown to be in closer agreement with the experimental sealing

effectiveness than the unsteady RANS simulations. The result

indicates that the previous failure to predict rim seal effectiveness

was due to turbulence model limitations in the turbine rim seal

flow. Consideration is given to the flow structure in this region.

20%?38% reduction of the number of iterations needed for convergence compared to Runge-Kutta coefficients recommended in the literature for comparable schemes and with the same computational cost per iteration.

Initially, the research focussed on modelling of a rotor-stator disc cavity. Steady CFD validations for a plane disc and for a disc with protrusion were carried out and a simplified body force model was developed for including the 3D effects of rotating and stationary bolts into the axisymmetric CFD models. The simplified rotor bolt model was verified and validated by comparing the results with Sussex Windage rig test data and 3D CFD data. The simplified stator bolt model was verified using 3D CFD results. The simplified rotor bolt model was found to predict the drag and windage heat transfer with reasonable accuracy compared to 3D sector CFD results. However, 3D sector CFD under-predicts the high core flow swirl and the adiabatic disc surface temperature inboard of the bolt, compared to experimental data.

In the second part of the study, unsteady Reynolds averaged Navier-Stokes (URANS) calculations of the rotating bolts cases were performed in order to better understand the flow physics. Although the rotor-stator cavity with bolts is geometrically steady in the rotating frame of reference, it was found that the rotor bolts generate unsteadiness which creates time-dependent rotating flow features within the cavity. A systematic parametric study is presented giving insight into the influence of the bolt number and the cavity geometric parameters on the time dependent flow within the cavity. The URANS calculations were extended to a high pressure turbine (HPT) rear cavity to show possible unsteady effects due to rotating bolts in an engine case.

Following this, the body force model was adapted to model the rotating hole velocity changes and flow through honeycomb liners. The honeycomb and hole models were verified by comparing the results with available experimental data and 3D CFD calculations.

In the final part of the study, coupled FE-CFD calculations for a preliminary design whole engine thermo-mechanical (WETM) model for a transient square cycle was performed including the effects of non-axisymmetric features. Six cavities around the HPT disc were modelled using CFD. The coupled approach provides more realistic physical convective heat transfer boundary conditions than the traditional approach. The unvalidated baseline thermo-mechanical model results were verified using the high fidelity coupled FE-CFD solution. It was demonstrated that the FE-CFD coupled calculations with axisymmetric modelling of 3D features can be achieved in a few days time scale suitable for preliminary engine design. The simplified CFD based methods described in this thesis could reduce the computational time of transient coupled FE-CFD calculations several orders of magnitude and may provide results as accurate as 2DFE-3DCFD coupled calculations.

in the lower part of the disc and over-predicted in the outer region by both URANS models, whereas the LES provides a much better agreement with the measurements. The behaviour results primarily from a different flow structure in the source region, which, in the LES, is found to be considerably more extended and show localized buoyancy phenomena that the URANS models investigated do not capture.

swirl ratio predicted by different sub-grid scale models tested is in good agreement with the measurements, although a slight overprediction is observed at lower radii. This has been demonstrated to be caused by an excessive numerical dissipation. Adopting a stable, less dissipative I-LES solution, the swirl ratio matches the data almost perfectly. In the next activity, the prediction from a Large-Eddy Simulation conducted for a rotating cavity with a radial inflow introduced from the shroud and heated on one wall have been compared with experimental data available from the literature, and with those obtained using two URANS eddy-viscosity models. The LES solution has shown a very good agreement especially in the outer part of the cavity, capturing buoyancy effects. The results of two URANS models are considerably worse than the LES.

Since LES is currently limited for application in industry by the high computational demand, Reduced Order Methods (ROM) that use data from LES have been considered in order to construct a model which could result in a computationally efficient method for design purposes. The POD-Galerkin procedure has been validated for the relative simple turbulent shear flow of the plane Couette flow. Then, the low Mach number turbulent flow in a rotor-stator cavity has been modelled. Overall, it is possible to claim that the models studied reasonably well predict the turbulence phenomenon for the rotor-stator flow (LES statistics and experimental measurements have been used as a benchmark).

cavity with a radial inflow introduced from the shroud. The dimensionless mass flow rate of the

radial inflow is C

_{w}= 3500 and the rotational Reynolds number, based on the cavity outer radius, is equal to

*Re*= 1.2 x 10v. The time averaged local Nusselt number on the heated wall is compared

_{¸}with the experimental data available from the literature, and with those derived from the solution

of two Unsteady Reynolds Averaged Navier-Stokes (URANS) eddy viscosity models, namely the

Spalart-Allmaras and the

*k - É SST*model. It is shown that the Nusselt number is under-predicted

in the lower part of the disc and over-predicted in the outer region by both URANS models, whereas

the LES provides a much better agreement with the measurements. The behaviour results primarily

from a different flow structure in the source region, which, in the LES, is found to be considerably

more extended and show localized buoyancy phenomena that the URANS models investigated do

not capture.

URANS and zonal-hybrid URANS-LES methodologies. The prediction is compared with experimental data, obtained via engine testing. The level of seal clearance and coolant

mass-?ow inside the secondary air system cavities of the turbine has a range of uncertainty during testing.

In the ?rst part of this work a small-sector URANS computation with the boundary conditions re?ecting the best understanding of the engine during testing is carried out

and a miss-match with experiments found. As a result, a sensitivity analysis with the seal clearance and coolant mass-?ow is undertaken, in which it is found that the inges

tion prediction is largely a?ected by these two parameters. To assess the e?ect of the computation of turbulence, a zonal-hybrid URANS-LES computation is carried out at

the condition with the best understanding of the test and it is found that the ingestion prediction improves compared to a URANS computation. The similar ?ow structures obtained between the two approaches suggests that the improvement is due to more accurate turbulent mixing being predicted by the zonal-hybrid URANS-LES computation.

Finally, some ingestion models are tuned to the CFD data originating from the sensitivity analysis. An optimization study with the resulting sealing e?ciency characteristic is

then carried out and a condition that minimizes the error to the experimental data found. A new small-sector URANS computation at this condition is unable to match the

experimental data, with a greater miss-match found in the more radially inwards region of the cavity. Due to the better agreement of the CFD solution to the ingestion models

in the part of the cavity closer to the main annulus than in the more radially inwards part of the cavity, it is suggested that predicting the ?ow structure in this region is key to the ingestion prediction. Considering the improvement seen with the zonal-hybrid methodology it is argued that a similar optimization study using this methodology would have improved the alignment to the experimental condition.