Uncertainty quantification of leakages in a multistage

simulation and comparison with experiments

C. M. Mazzoni 1 *, R. Ahlfeld2, B. Rosic1, F. Montomoli2

Abstract
The present paper presents a numerical study of the impact of tip gap uncertainties in a multistage
turbine. It is well known that the rotor gap can change the gas turbine efficiency but the impact of the
random variation of the clearance height has not been investigated before.
In this paper the radial seals clearance of a datum shroud geometry, representative of steam turbine
industrial practice, was systematically varied and numerically tested. By using a Non-Intrusive
Uncertainty Quantification simulation based on a Sparse Arbitrary Moment Based Approach, it is
ISROMAC 2016 possible to predict the radial distribution of uncertainty in stagnation pressure and yaw angle at the exit
International of the turbine blades.
Symposium on This work shows that the impact of gap uncertainties propagates radially from the tip towards the hub of
Transport the turbine and the complete span is affected by a variation of the rotor tip gap. This amplification of the
Phenomena and uncertainty is mainly due to the low aspect ratio of the turbine and a similar behavior is expected in high
Dynamics of pressure turbines.
Rotating Machinery
.
Hawaii, Honolulu Keywords
April 10-15, 2016 Uncertainty Quantification — Tip Leakage — CFD
1
Department of Engineering Science, University of Oxford, United Kingdom
2
Department of Aeronautics, Imperial College of London, United Kingdom
*Corresponding author: cosimo.mazzoni@eng.ox.ac.uk

INTRODUCTION by turbine designers as guidelines for their particular shroud

geometry rather than to create universal design rules.
In order to accommodate rotor to casing relative Finally, blade tip clearance changes during operation due to
movement during turbine operation, circumferential inlet and rotor unbalances, dynamic deformations and blade vibrations
exit shroud cavities must be formed. These necessary may have a huge impact on the safety, the lifetime and the
alterations in the overall flow path geometry significantly energy efficiency of turbo machines. Thus, the increasing need
modify the main blade flow field. All reported works (i.e. for reliability in gas turbine design, both for power generation
Wallis [1], Pfau et al. [2], Rushton [3], Giboni et al. [4], and aero-propulsion (Fadlun et al. [16]), requires tools that are
Metzger and Rued [5-6], Anto et al. [7]) show the highly three- able to include the strong variability due to manufacturing
dimensional nature of the flow in the cavities and the complex tolerances, assembly process and in service operations in the
interaction between the leakage flow and the main passage design system [17]. The question arising in recent years is how
flow. The cavities are fed by mainstream fluid and the flow to take into account these features with their associated aleatory
inside the cavities is influenced by the blade-to-blade pressure distribution. Bunker [18] applied a Monte Carlo simulation to
field from the upstream and the downstream blade row and study the impact of geometrical variations on HP components.
secondary flow structures from the upstream blade row The author showed that there is a wide range of parameters that
(passage vortex, wakes, etc.). The presence of the shroud can vary and their relative impact on the component life is
cavities, the mixing process when the leakage flow re-enters the different. Manufacturing variations can reduce the nozzle life
mainstream, and the influence of leakage flow on the by 33%. A stochastic study on statistical variations introduced
downstream blade row flow field all cause additional losses. by in service degradation of the rotor tip was studied by [19].
Shroud leakage loss mechanisms have been investigated by The authors proved that the tip heat transfer is strongly
many authors. Traupel [8], Denton and Johnson [9], Denton influenced by the stochastic distribution of fillet radius and tip
[10], Gier et al. [11], Rosic and Denton [12], Rosic et al. [13], gap. The study has shown that is crucial to use an uncertainty
all tried to quantify these losses. However, the loss-generating quantification approach in hot streak migration predictions. A
mechanisms are highly dependent on the particular shroud and summary of the current state of UQ for aircraft engines can be
stage geometry, therefore it is difficult to develop universal found in [20].
design rules and to correlate different geometrical and Most of the uncertainty studies in literature have been
operational turbine parameters to generic shroud design. A map obtained in simplified geometries. The study of Bunker [18],
of the possible turbine efficiency changes caused by different for example, was based on a flat plate assumption. More
shroud modifications was obtained by Rosic et al. [14-15] who recently some studies have considered the impact or real
investigated the influence of each geometric parameter on the geometries [19, 21-27]. However it is not clear how uncertainty
mainstream aerodynamics. The intention was to summarize is propagating in a flow. Carnevale et al. [21] have shown that
these effects and to highlight efficiency trends that can be used the impact of uncertainty is modulated in a channel assuming a
“transitional” like behavior. For this reason in this work we Table 1. Turbine Geometrical and Operational Parameters
decided to simulate a multistage HP steam turbine to evaluate Design operating point data
how the impact of uncertainty on seals clearance is affecting the Flow Coefficient φ 0.384
downstream stages and is redistributing in the span-wise Stage Loading Ψ=∆h0/U2 1.0
Design Speed [rpm] 830
direction.
Inlet Mach Number 0.045
Inlet/Outlet Total Pressure [Pa] 100465.4 75545.5
1. THE MODEL TURBINE AND EXPERIMENTAL Inlet/Outlet Total Temperature [K] 300 279
METHODS Number of Stages 3
Inter Blade Row Spacing [mm] 25
The experiments were carried out by Rosic [28] using a Blade Height h [mm] 75
low speed multistage air turbine that was designed to represent Mid-span Radius [mm] 462.5
Hub to Tip Ratio 0.85
the first few stages of a high pressure steam turbine (Fig.1).
Atmospheric air enters the inlet section radially and passes Blade parameters Stator Rotor
through a flow straightener before entering the turbine. Air is Blade Number 40 38
drawn through three turbine stages by a fan downstream of the True Chord l [mm] 89.73 105
turbine. Both stator and rotor blades were shrouded. The Aspect Ratio h/l 0.836 0.714
operating point, determined by the flow coefficient and Pitch to Chord Ratio 0.81 0.728
Reynolds Number (1) 2.53·105 2.97·105
rotational speed, is set by the fan power and dynamometer (1) based on true chord and exit velocity
brake. The flow coefficient vx/U is calculated using the axial
velocity at turbine inlet, which is measured indirectly using a
mass flow measurement in the exhaust section and a density
measurement at inlet.

Fig. 1. Schematic of model turbine Fig. 2. Shroud sealing arrangement

Experiments were performed on 50% reaction blading at traverses over one blade pitch at an axial distance of 6 mm and
the design operation condition ( φ = 0.384 ). The key rig 17 mm downstream of the stator and rotor trailing edge tip
geometrical and operational parameters are presented in Tab.1. respectively (Fig. 1). The traverse grid was 37 x 37 uniformly
The blade parameters are taken at blade mid-height. The spaced points in both radial and pitch-wise directions. A
blading was designed to represent typical high pressure steam detailed description of the experimental setup is reported in
turbine conditions, and all parameters satisfy that condition. [28].
Only the Reynolds number, based on exit velocity and true
chord, is significantly lower than that in a real turbine. 2. TBLOCK DESCRIPTION
The rotor shroud cavities geometry and sealing
arrangement under scrutiny is described in Figure 2. This TBLOCK is a multi-block structured grid solver developed
configuration is representative of commercial turbine design as by John Denton [29]. This is a steady or unsteady RANS solver
it allows for the axial movement of the rotor shaft relative to and uses the finite volume method, with the explicit SCREE
the casing. The shroud forms a radial clearance of 3mm with scheme [30] for steady simulations and dual-time stepping
the casing. Using two approximately 0.75 mm clearance radial method for unsteady calculations. Turbulence is modelled using
seals resulted in an over-shroud leakage flow of approximately a mixing length approach. Laminar to turbulent boundary layer
2%. A representation of the three stages experimental turbine is transition can be modelled by assuming a laminar boundary
layer up to a specified point on each blade and end-wall surface
reported in Fig. 3.
To investigate the time-mean flow of the main passage, and a turbulent one downstream. The mixing plane approach
traverses were conducted using a five hole pneumatic probe. allows reversed flow across the mixing plane, which may occur
The properties were measured by performing full-span area near the end-walls close to the leakage cavities. To accelerate
convergence multigrid and spatially varied time steps are used.
 Fig. 3. Representation of the experimental turbine

Fig. 4. TBLOCK computational domain

TBLOCK can be run in parallel with domain decomposition grid points. The coarser mesh adopted 99 grid points in the
done on a block basis. axial direction and 91 points in pitch-wise and radial directions,
The code is aimed at predicting both main blade path and for a total of 18.92 million computational nodes.
secondary gas path flows in turbomachinery although it can be
used to calculate other types of flow. The solver overcomes
most of the limitations of the simple shroud leakage model in
calculating complex geometries, and was used for the
calculations in this paper including the shroud leakage flow
path. The flow field is divided into as many ‘cuboid’ blocks as
are necessary to represent the geometry and each block is
solved separately. Information passes between blocks using
different types of interfaces treated as ‘patches’ (inlet and exit
boundary, solid boundary, mixing plane, periodic boundary).
The method is tolerant of extreme levels of grid distortion, and
when necessary can interpolate the flow data from one patch
onto another, which may lie on a non-contiguous face and may
have a different number of grid points.
Four configurations with different radial seals clearance
(0.25, 0.50, 0.75, 1.00 mm) on the shroud cavities for all the
three stages were modelled using TLOCK. Simulations were
completed modelling all three stages single passage domain,
including rotor tip shroud and stator hub leakage paths. Figure
4 presents the flow domain decomposition.
The domain was divided into 72 blocks. In particular, the
leakage flow paths were divided into the necessary number of
rectangular blocks using structured H meshes. Two different
levels of mesh refinement were analysed. For the finer mesh,
the main blade passage was modelled using 124 grid points in
the axial, and 114 grid points in both pitch-wise and radial
directions (Fig. 5). The whole domain consists of 36.89 million Fig. 5. TBLOCK computational grid
 The detailed grid and block structure used to model each
rotor blade passage is presented in figure 6. The rotor tip
shroud for all the configurations under scrutiny was modelled
with different radial clearance gaps reproduced for every stage.

3. NUMERICAL RESULTS
3.1 Comparison with experiments
In comparing calculated and measured flow patterns it
must be remembered that the real clearance is not known with
great accuracy and that it varies significantly around the
circumference. In the experimental configuration, the shroud
forms a radial clearance of 3mm with the casing and the
average radial gap for seals is assumed to be 0.75 mm.
Predicted distributions of total pressure coefficient
downstream of the second and third stators and their
comparison with experimental results are shown in Fig. 6. Both
pitch-wise averaged profiles (left) and 2-D contour map (right)
are presented. The measurement plane covers one pitch from
3% to 95% span. As it can be observed on the pitch-wise
Fig. 6. TBLOCK – rotor and tip shroud grid structure averaged profiles, numerical simulations on both coarse and
fine mesh well reproduce the pressure levels and the main trend
The simulation was run in parallel using 72 processors. A fully along the span. A comparison between 2-D contours show that
converged unsteady solution initialised from previous steady TBLOCK calculations captured reasonably well the main flow
computations was obtained for 10 blade-to-vane passes. features, such as the locations and intensity of the main loss
Averaging of the flow variables was then conducted for a core and of the hub end-wall secondary flow, and the migration
further 10 blade-to-vane passes. The computational time towards the mid-span of the concentrated low momentum flow
required to complete one blade-to-vane pass for the finer mesh associated with casing end-wall boundary layers and leakage.
was about 7 hours.

Fig. 7. Measured and predicted total pressure coefficient downstream stator 2 & 3
 Article Title — 5
span and end-walls loss cores, although their levels were
slightly overestimated.
Velocity field is analysed in Fig. 9 and 10, which show the
span-wise distribution of pitch-wise averaged values of axial
velocity (Fig. 9) and yaw angle (Fig. 10) at different
downstream stator/rotor locations. Numerical simulations were
in good agreement with experimental results in predicting
magnitude and directions of the flow downstream both vanes
and blades passages. In particular, in the region close to the
casing and the shroud exit cavity of the third rotor, TBLOCK
was able to reproduce the overturning trend in the main
passage flow angle caused by mixing of the leakage flow.

Fig. 8. Measured and predicted pitch-wise averaged total

pressure coefficient downstream rotor 2 & 3

Fig. 9. Measured and predicted pitch-wise averaged axial

velocity downstream stator 3 & rotor 3

Fig. 10. Measured and predicted pitch-wise averaged yaw

angle downstream stator 2 & rotor 3

Comparisons between experimental and numerical pitch-

wise averaged relative total pressure coefficients downstream
the second and third rotor, are reported in Fig. 8. Numerical Fig. 11. Predicted pitch-wise averaged total pressure
simulations are able to predict the radial locations of the mid- coefficient profiles for different seals clearances
 Article Title — 6
influence of the different shroud radial gap on the main blade
path. This was done in a numerical study using the same
numerical flow domain as in the first part of this section
varying the seal clearance over both radial fins simultaneously
while the datum shroud geometry was kept unchanged.
The variations of pitch-wise averaged total pressure
coefficient and yaw angle distributions with clearance gap are
reported in Figures 11 and 12, respectively. Absolute and
relative values are shown at stators exits (the left sides) and
downstream the rotors (right side), respectively.
It is worthwhile noting how the flow field in the exit rotor
shroud cavity is greatly affected by the leakage jet generated in
the upstream seal gap and re-entering the main passage. The
leakage flow fraction is enhanced with increasing clearances,
causing higher pressure losses near the casing (Fig. 9, right
side). Also, as the clearance increases, the leakage jet increases
its momentum (Fig. 11, right side), forming a region of higher
relative yaw angle in the near casing of the main annulus (Fig.
12, right side), because of mechanism explained in [14].
The negative influence of the leakage flow on the flow
field in the downstream blade row can be observed on the left
side of Fig. 12, by analysing the yaw angle distributions
downstream the stators exits. The leakage flow caused a strong
underturning of the flow at the mid-span, and also the ‘S’
shape angle distribution close to the casing, associated with the
casing secondary flows, was enhanced.

Fig. 12. Predicted pitch-wise averaged yaw angle profiles for

different seals clearances Fig. 13. Change in turbine efficiency with leakage fraction
3.2 Different seals clearance gaps Finally, the influence of leakage flow on the mainstream
According to steam turbine industrial practice, due to flow in the three stage turbine analysed in this paper can be
manufacturing tolerances and assembly process, real shroud summarized in the diagram of Fig. 13, which shows the
geometry presents some mismatches with their theoretical changes in turbine efficiency with leakage fraction. He
design and it is also very hard to have a reliable estimation on resulting trend is in agreement with the classical shroud
rotor shroud gap. Furthermore, during long-term turbine leakage theory which assumes that the loss in efficiency
operation the sealing effectiveness deteriorates due to rubbing associated with leakage flow is directly proportional to the
and worn sealing elements. Because of this, leakage fractions leakage flow rate
for shrouded blades can vary from 1% to 4 % within HP
turbines [14]. Therefore it is important to investigate the
 Article Title — 7
4. UNCERTAINTY QUANTIFICATION where N P is a truncation to limit the infinite expansion series.

A basic method to obtain the stochastic distributions of For the one dimensional case N P can simply be chosen as the

output variables like the stagnation pressure or the yaw angle maximum polynomial order of the chosen expansion plus one.
if the clearance gap varies randomly is the Monte Carlo Moreover, ξ is a Gaussian random variable representing the
method. First, a probability distribution must be assumed for
clearance gap, and y is representative for either the stagnation
the clearance gap. Then, a suitable random number generator
is used to create the individual input random samples to pressure ot the yaw angle.
follow the prescribed distribution. Finally, the simulation only For a Gaussian random variable, the optimal orthogonal
has to be run repeatedly for every random sample drawn from

()
polynomials ψ i ξ are the Hermite polynomials [33]. The
the PDF. Convergence is achieved by increasing the number choice of the polynomials is always a consequence of the
of the simulations. Unfortunately, the convergence is slow: probability distribution of the inputs variables. Thus, only the
the sample size has to be quadrupled every time to half the
error. For a cost intensive computational model, as it is the
()
coefficients α i x are unknown. In order to find these, a linear
case here, it is therefore prohibitive to use a Monte Carlo system of Np equations can be solved in the optimal Gaussian
method. collocation points ξ i :
In the given simulation, only one random variable is
present, the clearance gap, and it can be assumed to be
Gaussian. This assumption is made based on the available
Eq.2
literature on manufacturing uncertainties, which tend to
display Gaussian behaviour [26]. Both assumptions together
can be used to achieve an immense cost reduction by using a
Non-Intrusive Polynomial Chaos Method: the Probabilistic Here, the clearance gap is chosen as Gaussian random with
Collocation Method (PCM) as described in [31]. ( )
distribution N 0.75,0.03356 . This results in approximate gap

The underlying idea of all Polynomial Chaos methods is widths collocation points 0.25,0.5,0.74,1.00 and 1.25 as used
that the functional form of the stochastic outputs of a model before and the PDF as shown in Fig. 14.
can be found with significantly lower computational effort, if
they are approximated through a basis of optimal orthogonal
polynomials defined through the input random variables. The
additional concept of probabilistic collocation is that since the
statistical moments are integrals of the output PDFs, they can
be cheaply and accurately integrated by using optimal
Gaussian quadrature on the found polynomial expansion. In
consequence, the model only has to be run a few times using
the Gaussian collocation points of the clearance gap to
calculate accurate numerical approximations for all output
distributions.
The number of the Gaussian collocation points is
naturally connected to the polynomial order of the expansion.
They are chosen as the roots of the next higher order
polynomials. For example, 5 points are needed for a 4th order
expansion. Moreover, the number is also dependent on the Fig. 14. Probability distribution used as input PDF including
number of input variables used. Since for multiple input optimal Gaussian collocation points.
variables the collocation points have to be combined through
tensor products, Polynomial Chaos methods become quickly The mean µ y and variance σ 2y of y can then be found

using the coefficients
by calculating
inefficient for increasing numbers of input random variables.
NP −1
For only one input variable, however, five collocation points
can achieve an accuracy level comparable a million Monte µ y = α0 σ 2y = ∑α 2
i
ψ i2 Eq.3
Carlo runs [31]. i=1

The PDF of y can be illustrated by Monte Carlo sampling the

found expansion in eq. 1 using random samples with PDF
4.1 Probabilistic Collocation Method (PCM)

( )
N 0.75,0.03356 . Unlike direct sampling of the CFD model,
The basic concept of Polynomial Chaos expansions can this sampling process needs negligible computation time.
be traced back to Wiener [32], who found that the stochastic Convergence of the polynomial expansion can be found by
( )
output y x ,ξ of a function can be decomposed into a series

increasing the order of the expansion. Here, orders from 3 to 5
were tested. Comparing 4th and 5th order a four decimal
with deterministic coefficients αi x

() and stochastic accuracy was found so that no higher orders were computed.

()
orthogonal polynomials ψ i ξ , such that:
4.2 Results
N P −1

( ) () ()
y x ,ξ ≈ ∑ α i x ψ i ξ Eq.1 Figure 15 shows the pressure coefficient distribution at the
i=0 exit of each row. The dashed red line is the standard deviation.
The uncertainty bars represent one standard deviation
superimposed on the mean pressure distribution.
 Article Title — 8
quantification study therefore gives suggest a reason of this
disagreement between CFD and experiments: it can be
associated to a small variation of the tip gap in the machine
that is different from the CFD model.

Fig. 15. Stagnation pressure coefficient distribution and

uncertainty

In the first stage we can observe that the impact of

uncertainty in the rotor tip is affecting only the higher part of
the rotor. Only about 10% of the span (100-90%) is
influenced by the rotor tip random variation. Similarly large Fig. 16. Yaw angle variation and standard deviation
random variation in this region can be observed for rotor 2 superimposed
and rotor 3 as well. For the second stator, the peak of
uncertainty is surprisingly at 50% and the region between
40% and 60% of the span is mostly affected. For the second REFERENCES
rotor, the entire span is affected. In the last stage, the standard 1. Wallis, A. M., “Secondary and Leakage Flows in a
deviation for the stator peaks at 78%, with a region of Multistage Turbine”, PhD Thesis, Cambridge University
increased variation higher between 65 and 80%. The last rotor
Engineering Department, 1997.
is most strongly affected by the uncertainty with a strong
variation from 20 to 80%. 2. Pfau, A., Treiber, M., Sell, M., Gyarmathy, G, “Flow
There are different mechanisms involved. The changes in Interaction From the Exit Cavity of an Axial Turbine Blade
rotor tip gaps modifies the mass flow distribution and this Row Labyrinth Seal”, Transaction of the ASME, Journal of
alter the radial uncertainty. At the same time, the impact of Turbomachinery., 2001, Vol. 123, pp. 342-352.
the leakage flows modifies the work carried out by the rotor
due to the variation of the inlet (and exit) angle. This is 3. Rushton, G. J., “High-Pressure Turbine Shroud Leakage,
clearly shown in figure 16 where the exit angles indicates the PhD Thesis”, Cambridge University Engineering
variation in load associated to each row. Department, 2003.
It is important to notice that the found regions with high
4. Giboni, A., Menter., J. R., Peters, P., Wolter, K., Pfost, H.,
standard deviation agree with the regions in which the CFD
results differed largely from the experimental in the original Breisig, V., “Interaction of Labyrinth Seal Leakage Flow
work of Rosic et al [12]. The performed uncertainty and Main Flow in an Axial Turbine”, ASME Paper GT2003-
 Article Title — 9
38722, 2003. 19. D'Ammaro A, Montomoli F.: "Uncertainty Quantification
and Film Cooling", Journal of Computer and Fluids, 2013,
5. Metzger, D. E., Rued., K., “The Influence of Turbine
Vol:71, pp.320-326.
Clearance Gap Leakage on Passage Velocity and Heat
Transfer Near Blade Tips: Part I—Sink Flow Effects on 20. Montomoli, F., Carnevale, M., D’Ammaro, A., Massini, M.,
Blade Pressure Side”, Journal of Turbomachinery, 1989, & Salvadori, S. (2015). Uncertainty Quantification in
Vol. 111, pp. 284-292. Computational Fluid Dynamics and Aircraft Engines.
Springer.
6. Metzger, D. E., Rued., K., “The Influence of Turbine
Clearance Gap Leakage on Passage Velocity and Heat 21. Carnevale M., Montomoli F., D’Ammaro A., Salvadori, F
Transfer Near Blade Tips: Part II—Source Flow Effects on Martelli, “Uncertainty Quantification: A Stochastic
Blade Suction Sides”, Journal of Turbomachinery, 1989, Method for Heat Transfer Prediction Using LES”, Journal
Vol. 111, pp. 293-300. of Turbomachinery, 2013, doi: 10.1115/1.4007836.

7. Anto, K., Xue, S., Ng, W. F., Zhang, L. J., Moon, H. K., 22. Montomoli F., D’Ammaro A, Uchida S.: “Uncertainty
“Effects of Tip Clearance Gap and Exit Mach Number on Quantification and Conjugate Heat Transfer: a Stochastic
Turbine Blade Tip and Near-Tip Heat Transfer”, ASME Analysis”, Journal of Turbomachinery, 2013.
Paper No. GT2013-94345, 2013. 23. Montomoli F, M. Massini, S. Salvadori: “Geometrical
8. Traupel, W., “Erbenisse von Turbinenversuchen”, From: Uncertainty in Turbomachinery”: International Journal of
Flow Research on Blading, Elsvier, 1970. Computer and Fluids, Elsevier, 2010.

9. Denton, J. D., Johnson, C. G., “The Tip Leakage Loss of 24. Büche, D., Beetz, M., Ribi, B., & Turbo, M. A. N. (2010).
Turbine Blades – A Review and a Theory for Shrouded “Uncertainty Analysis for large-scale industrial radial
Blades”, CEGB Report No R/M/N627, Marchwood compressors”, ASME paper GT2010-22918, 2010.
Engineering Laboratories, 1972. 25. De Maesschalck, C., Lacor, C., Paniagua, G., “Performance
10. Denton, J. D., “Loss Mechanisms in Turbomachines”, Robustness of Turbine Squealer Tip Designs due to
ASME Paper 93-GT-435, 1993. Manufacturing and Engine Operation”. In ISABE
conference paper, 2015.
11. Gier, J., Stubert, B., Brouillet, B., de Vito, L., “Interaction
of Shroud Leakage Flow and Main Flow in a Three-Stage 26. Panizza, A., Bonini, A., Innocenti, L., “Uncertainty
LP Turbine”, ASME Paper GT2003-38025, 2003. Quantification of Hot Gas Ingestion for a Gas Turbine”,
ASME paper GT2015-42679, 2015.
12. Rosic, B., Denton, J. D., Pullan, G., “The Importance of
Shroud Leakage Modelling in Multistage Turbine Flow 27. Wunsch, D., Hirsch, C., “Quantification of Combined
Calculations”, ASME Paper GT2005-68459, 2005. Operational and Geometrical Uncertainties in Turbo-
Machinery Design”, ASME paper GT2015-43399, 2015.
13. Rosic, B., Denton, J. D., “The Control of Shroud Leakage
Loss by Reducing Circumferential Mixing”, ASME Paper 28. Rosic, B., “The Control of Shroud Leakage Flows in Low
GT2006-90946, 2006. Aspect Ratio Multistage Turbines”, PhD Thesis, Cambridge
University Engineering Department, 2006.
14. Rosic, B., Denton, J. D., Curtis, E. M., “The Influence of
Shroud and Cavity Geometry on Turbine Performance – 29. Denton, J. D., “TBLOCK Manual”, 2002.
An Experimental and Computational Study, Part 1: 30. Denton, J. D., “The Calculation of Three Dimensional
Shroud Geometry”, ASME Paper GT2007-27769, 2007. Viscous Flow Through Multistage Turbomachines”, ASME
15. Rosic, B., Denton, J. D., Curtis, E. M., Peterson, A. T., Paper 90-GT-19, 1990.
“The Influence of Shroud and Cavity Geometry on 31. Hosder, S., Walters, R. W., “Non-intrusive polynomial
Turbine Performance – An Experimental and chaos methods for uncertainty quantification in fluid
Computational Study, Part 2: Exit Cavity Geometry”, dynamics”, 48th AIAA Aerospace Sciences Meeting,
ASME Paper GT2007-27770, 2007. January 2010.
16. Fadlun E. A., Michelizzi I. De Iaco M., “Measurement http://arc.aiaa.org/doi/pdf/10.2514/6.2010-129
error influence on gas turbine operability for condition- 32. Wiener, N., "The Homogeneous Chaos". American Journal
based maintenance and reliability/availability of Mathematics, 1938, Vol. 60, pp. 897–936
improvement”, ASME Paper GT2008-50749, 2008.
33. Xiu, D., & Karniadakis, G. E., “Modeling uncertainty in flow
17. Williams, D. T., Smout, P., Bianchi, M., Joinson, M. B., simulations via generalized polynomial chaos”, Journal of
“The Use of Probabilistic Methods in Determining Computational Physics, 2003, Vol. 187, pp. 137–167.
Turbine Disc Cyclic Life Uncertainty”, ASME Paper
GT2013-94972, 2013.
18. Bunker R.S. “The Effect of Manufacturing Tolerances on
Gas Turbine Cooling”, Journal of Turbomachinery, 2009,
vol. 131, pp. 41018-41018-11.