Return Channel Loss Reduction in Multi-Stage ARCHVEsCentrifugal Compressors

by

Anne-Raphaelle AubryBachelor of Science Magna cum Laude,

MASSACHUSETTS INSTITUTEOF TECHNOLOGY

JUL 10 2012

L7Z7RA RIESMlechla~iii~ ~

Engineering, Cornell University, 2010

Submitted to the Department of Aeronautics and Astronauticsin partial fulfillment of the requirements for the degree of

Master of Science in Aeronautics and Astronautics

at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

May 2012

@ Massachusetts Institute of Technology 2012. All rights reserved.

Author .............. ........Department of Aeronautics and Astronautics

May 11, 2012

C ertified by ....... . ...........................Edward M. Greitzer

H. N. Slater Professor of Aeronautics and AstronauticsThesis Supervisor

Accepted by............................ ...............

/ Eytan H. ModianoProfessor of Aeronautics and Astronautics

Chair, Graduate Program Committee

2

Return Channel Loss Reduction in Multi-Stage Centrifugal

Compressors

by

Anne-Raphaelle Aubry

Bachelor of Science Magna cum Laude, Mechanical Engineering,

Cornell University, 2010

Submitted to the Department of Aeronautics and Astronauticson May 11, 2012, in partial fulfillment of the

requirements for the degree ofMaster of Science in Aeronautics and Astronautics

Abstract

This thesis presents concepts for improving the performance of return channels inmulti-stage centrifugal compressors. Geometries have been developed to reduce bothseparation and viscous losses. A number of different features with potential to reduceseparation have also been investigated. The final proposed geometry uses a vanelessdiffuser which narrows on the shroud side at the beginning of the 180' bend, anaxially extended 1800 bend with increasing radius of curvature, and return channelvane leading edge radial position at an increased radius compared to the baseline.Three-dimensional calculations showed a 9% loss reduction compared to previouswork [1], with a cumulative loss reduction of 19% compared to a baseline geometry.

The geometry developed was based on specified inlet conditions. To examine thepotential for increased performance if this constraint was removed, a return channelgeometry was also defined that incorporated the same features but allowed modifiedinlet conditions, specifically radial inlet flow. The design of the impeller required forthis new inlet flow was not considered. An overall loss reduction of 23% comparedto baseline was found from the calculations. Modification of the impeller geometry isthus proposed as future work.

Thesis Supervisor: Edward M. GreitzerTitle: H. N. Slater Professor of Aeronautics and Astronautics

3

4

Acknowledgments

First and foremost, I am grateful to my advisor, Professor Edward Greitzer, for

his guidance over the past two years. I have the utmost admiration for his energy and

dedication, and owe him my deepest gratitude for both encouraging and challenging

me. This thesis, and my learning experience, would not have been the same without

him.

I would like to thank Mitsubishi Heavy Industries for sponsoring this project, and

I am appreciative of the strong and continued support provided by Dr. Eisaku Ito

and Dr. Sumiu Uchida. I especially wish to thank Mr. Akihiro Nakaniwa for his

patience, understanding and helpful comments throughout our regular interactions.

This work would not have been possible without the help of Professor Mick Casey,

whose industry experience, tremendous knowledge, and good humor at all hours were

invaluable from beginning to end.

I am particularly grateful to fellow labmate Jeff Defoe for his mentorship, both

in and out of the lab. I feel very fortunate to have benefited from his knowledge and

advice.

Many thanks to the other members of the GTL-MHI team, Nikola Baltadjiev and

Claudio Lettieri, for their help and shared discussions, technical or otherwise. I will

also never forget Tanya Cruz-Garza's cheerful honesty and friendly support when I

started out at the GTL.

I would not be where I am without Dave Cloud's unwavering support. I am

grateful to him for the possibilities he unlocked, and for his mentorship. I am greatly

indebted to my parents Jean-Pierre and Marie-Cecile Aubry for the opportunities

they gave me.

Finally, to my very best friend George Han, thank you for helping me keep things

in perspective.

5

6

Contents

1 Introduction 11

1.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . 11

1.2 Return Passage Geometry . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.4 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.5 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.6 Thesis Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 Implementation 17

2.1 Computational Approach. . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1.1 FLUENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1.2 MISES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2 Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.3 Inlet Profiles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.4 Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Characteristics of the Baseline Return Passage Flow 21

3.1 Baseline Loss Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 Baseline Flow Features . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2.1 Diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2.2 180' Bend Section . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2.3 Vane Section . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.2.4 90' Bend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

7

4 Return Passage Concept Development 29

4.1 2010 Geom etry ... . .. . . .... ....... . . .. .. . . . .. 29

4.1.1 2010 Geometry Flow Features . . . . . . . . . . . . . . . . . . 29

4.1.2 Lessons Learned from the 2010 Geometry . . . . . . . . . . . . 32

4.2 Axisymmetric Parametric Studies . . . . . . . . . . . . . . . . . . . . 32

4.2.1 Parameterization . . . . . . . . . . . . . . . . . . . . . . . . . 32

4.2.2 Parametric Study Results . . . . . . . . . . . . . . . . . . . . 34

4.3 Proposed New Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.3.1 D iffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.3.2 180' bend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.3.3 Vane Section . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

42

4.3.4 90' bend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.4 O ptiH Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.5 Final Geometry: OptiHb R+10 . . . . . . . . . . . . . . . . . . . . . 48

4.5.1 Geometry Features . . . . . . . . . . . . . . . . . . . . . . . . 48

4.5.2 Loss Reduction Results . . . . . . . . . . . . . . . . . . . . . . 51

5 Assessment of Return Passage with Modified Diffuser Inlet Geome-

try 53

5.1 Previous W ork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.2 Modified Diffuser Inlet . . . . . . . . . . . . . . . . . . . . . . . . . . 54

6 Summary and Proposed Future Work 59

6.1 Sum m ary . .. .. . .. . .. . .... . . ..... . . . . . . . . . .. 59

6.2 Future W ork . . . . . . . . . . .. . . . . .. .. . . . . . . . . . . . . 60

8

List of Figures

1-1 Standard Return Channel Geometry . . . . . . . . . . . . . . . . . . 12

1-2 Geometric Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1-3 Return Channel Geometry Parameters . . . . . . . . . . . . . . . . . 14

2-1 TurboGrid Generated Mesh (Vane Leading Edge) . . . . . . . . . . . 19

2-2 Swirl Angle Definition . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3-1 Baseline Loss Components . . . . . . . . . . . . . . . . . . . . . . . . 21

3-2 Location of Baseline Separation Regions . . . . . . . . . . . . . . . . 22

3-3 Normalized Radial Velocity in the Baseline Vane Section . . . . . . . 25

3-4 Baseline 900 Bend: Normalized Axial Velocity Contours mid-passage. 27

4-1 Performance of 2010 Geometry . . . . . . . . . . . . . . . . . . . . .. 30

4-2 Normalized Velocity Contours for Baseline (left) and 2010 Geometry

with Swept Diffuser (right), Axisymmetric Calculations . . . . . . . . 31

4-3 2010 Geometry: Normalized Radial Velocity . . . . . . . . . . . . . . 31

4-4 Return Channel Sample Bezier Parameterization . . . . . . . . . . . . 33

4-5 2010 Design: Study Parameters . . . . . . . . . . . . . . . . . . . . . 34

4-6 Proposed design parameters near (local) minimum loss . . . . . . . . 35

4-7 Baseline: Study Parameters . . . . . . . . . . . . . . . . . . . . . . . 35

4-8 Diffuser Geometry Comparison . . . . . . . . . . . . . . . . . . . . . 37

4-9 Normalized LSVD Perfomance Characteristics . . . . . . . . . . . . . 39

4-10 Width Increase for Different Geometries . . . . . . . . . . . . . . . . 40

9

4-11 Effect of Vane Number Reduction on Performance (MISES Computa-

tio n s) . . . . . . . . . . . . . . . . . . . . . . . . . . .

Different Vane Loading Distributions . . . . . . . . .

Leading Edge Flow Angles for OptiH geometries . . .

OptiH Series b6 Values . . . . . . . . . . . . . . . . .

OptiH series geometries . . . . . . . . . . . . . . . . .

Total Loss as a Function of b6 for OptiH Geometries .

OptiH-a: Normalized Radial Velocity . . . . . . . . .

OptiH-c: Normalized Radial Velocity . . . . . . . . .

OptiH-d: Normalized Radial Velocity . . . . . . . . .

OptiH-b: Normalized Radial Velocity . . . . . . . . .

Performance of OptiHb R+10 . . . . . . . . . . . . .

4-22 Off-Design Performance Comparison: Baseline,

tiH c R + 10 . . . . . . . . . . . . . . . . . . . .

OptiHb R+10 and Op-

Baseline and FD Diffuser Inlet Geometries . . . . . . . . . . . . .

OptiHb R+10 and Modified Diffuser Inlet Geometries . . . . . . .

Modified Diffuser Inlet Geometry . . . . . . . . . . . . . . . . . .

Tangential and Axial Velocity Levels Comparison at Diffuser Top

Tangential and Axial Velocity Levels Comparison at Vane LE . .

10

4-12

4-13

4-14

4-15

4-16

4-17

4-18

4-19

4-20

4-21

. . . . . . . 42

. . . . . . . 43

. . . . . . . 44

. . . . . . . 45

. . . . . . . 46

. . . . . . . 46

. . . . . . . 47

. . . . . . . 48

. . . . . . . 49

. . . . . . . 50

. . . . . . . 50

5-1

5-2

5-3

5-4

5-5

51

54

55

55

56

56

Chapter 1

Introduction

1.1 Background and Motivation

Multi-stage centrifugal compressors are used for a variety of industrial applications,

for example gas injection, ethylene plants, or oil refining. They are robust, with

less moving parts than their alternatives. To reduce floor space and cost there is an

interest in shrinking the compressor size and developing radially and axially compact

machines; multi-stage machines with a large axial extent can encounter shaft stiffness

issues, and vibrations. There is also a desire to increase machine efficiency. For

conventional designs however, reducing the radial and axial extent of the machine

can lead to an undesirable efficiency loss and a reduction in range.

With the high impeller efficiencies of machines now in operation, return channel

losses become critical to the overall stage performance. Development of return channel

geometries without reductions in range or efficiency, while meeting size constraints, is

thus an important problem. Overall guidelines for return channel design are described

in this thesis.

1.2 Return Passage Geometry

Figure 1-1 shows a centrifugal compressor stage composed of an impeller and a return

channel. The return channel itself includes four sections: diffuser, 180' return bend,

11

d

ShroudSide

BladedRegion

ShroudSide

Entry Hub Bend ExitFlow Side Flow

----- ----- ----- ----- -- ..-------------- L

Figure 1-1: Standard Return Channel Geometry

return vane, and 900 bend. In the cases examined all sections of the return channel

except the return vane passage are vaneless.

The role of the diffuser is to convert kinetic energy (from the impeller exit) into

static pressure rise, while the role of the vane channel is to remove the swirl velocity

component, guiding the flow axially into the next impeller. Flow exiting the return

vane channel should ideally be free of swirl for input into the next stage impeller.

The return channel geometry modifications described in subsequent chapters fol-

low the geometrical constraints indicated in Figure 1-21. Specifically, the inlet flow

path and impeller, the maximum diameter (at the top of the 1800 bend), and the

axial stage length, measured from impeller hub to return vane shroud, are all defined

as maximum values.

'Personal communication from A. Nakaniwa, 2011

12

Figure 1-2: Geometric Constraints on Multi-stage Compressor Design

1.3 Nomenclature

The return channel geometry will be described in terms of a set of geometric param-

eters as in Figure 1-32, which shows those varied in the present study. Parameter

b2 defines the impeller outlet width. b3 defines the width at the end of the diffuser

pinch, and b4 defines the diffuser width at its outlet. The first half of the 1800 bend

is defined by the radial location at which it begins, r4, and its radius 14. Similarly,

the second half of the 180' bend is defined by its location r6 and radius 16. The 180'

bend width is defined by b5, width at the top of the bend, and b6, width at the outlet

of the bend. The width at the outlet of the return channel vane is defined as b7.

With respect to geometric constraints, the casing dimensions limit r5, the max-

imum radius, and the axial length constrains 12, the distance between impeller hub

and return vane shroud. Finally, the next stage geometry fixes outlet parameters r7

and b7.2Courtesy of Professor M. Casey, 2011

13

b2 r2 ( b

Figure 1-3: Return Channel Geometry Parameters

1.4 Previous Work

A number of computations and experiments have been reported on the subject of flow

behavior in return channels and information for design, a recent short review can be

found in [2].

Veress and Braembussche [3] developed an analytical design procedure for three-

dimensional vanes using a prescribed load distribution, with the vane leading edge

extended upstream into the 1800 bend as far as the vaneless diffuser exit. Within

the 1800 bend, local flow accelerations and decelerations can lead to separation. The

meridional velocity increases in regions of convex curvature (ie: hub side of the bend

inlet, shroud side of the bend exit), and decreases in regions of concave curvature (ie:

shroud side of bend inlet, hub side of bend exit). By adjusting the vane thickness a

smoother Mach number distribution was obtained on the vane, leading to performance

improvement with a reduction by 3.4% of total pressure loss coefficient.

Aalburg et al [4] sought to reduce the diffuser radius ratio (outer radius/inner

14

radius) from 1.45 to 1.19 without losses in performance. Despite the increase in flow

turning and higher vane loading, optimization of the return channel vane and endwalls

gave an increase of up to one point in stage efficiency across the operating range. They

extended the return channel vane upstream, with its leading edge at the diffuser exit,

introducing flow guidance earlier in the flow path, to reduce both the swirl component

and magnitude of the velocity. Although the bend losses increased as a consequence

of additional wall surface area and increased blockage, the overall effect was beneficial,

as the reduced velocities had a large impact on stagnation pressure losses, achieving

0.5-1 point gain in stage polytropic efficiency. This work demonstrated potential for

improvement with an upstream vane extension for centrifugal compressors, although

the diffuser radius ratio was less (1.45) than those of interest here.

As part of a design development strategy to reduce diffuser radius ratio without

penalizing either efficiency or operating range, Simpson [5] evaluated the benefit of

using steady injection to reduce return channel losses. The injection was intended

for use at off-design conditions, at mass flows lower than at design, where separation

regions are present along a significant portion of the vane chords. The goal was to

reduce losses by increasing the momentum in the vane boundary layers. Because

of the proprietary nature of the information, no exact numbers were provided. For

this method to be viable, the reduction in loss coefficient achieved must more than

compensate for the penalty from using recirculated flow as the injected fluid, and the

study did conclude that for certain configurations steady injection flow control was

useful for decreasing overall losses.

The work described in this thesis follows that of Glass [1] from 2008 to 2010. In

the course of his work loss mechanisms in the baseline return passage were identified,

the effects on losses of changes to the hub and shroud wall geometry were determined,

and a geometry in which return passage losses were reduced was developed.

In Glass's work, flow in the baseline passage was analyzed to identify those loss

sources with potential for reduction. Separation regions were observed on the shroud

near the 1800 bend inlet and on the hub at the exit of the 180' bend. The former is due

to a sharp curvature in the return bend, with large meridional velocity deceleration

15

and aggressive diffusion in the bend. The latter separation results from the sharp

hub-side curvature at entry to the return vane section, which propagates downstream.

Separated flow in the region over the vane suction side can also result from a vane

incidence mismatch caused by large flow variations between hub and shroud.

The final loss reduction obtained in the computations by Glass gave a loss coeffi-

cient reduction of 10% compared to the baseline.

1.5 Research Questions

In this thesis a new concept for multi-stage centrifugal compressor return channel

geometry is developed. The design space is constrained by stage overall dimensions,

and ease of manufacturing is kept in mind. The thesis addresses the following research

questions:

(1) What are the effects of meridional geometry features on return channel perfor-

mance, and what changes can reduce losses ?

(2) What mechanisms other than meridional geometry changes can reduce losses ?

(3) What is the effect of impeller exit flow on performance ?

1.6 Thesis Contributions

(1) Loss reduction mechanisms for compact return channels in multi-stage centrifugal

compressors are characterized. The effect of these mechanisms are quantitatively

evaluated, using three-dimensional calculations, for a range of design parameters.

(2) Based on the computational results a new concept for improved performance

channels is defined.

16

Chapter 2

Implementation

2.1 Computational Approach

2.1.1 FLUENT

Numerical simulations were performed with the commercially available ANSYS FLU-

ENT code (V12.0), which uses a viscous, compressible, steady, RANS solver. All

calculations were done with stationary reference frames. Three-dimensional calcu-

lations were for a single passage with periodic boundary conditions. Second order

discretization was used for all calculations.

The turbulence model used was the k-omega SST, as developed by Menter [6]

and as also used by Glass [1]. This model transitions between near wall turbulence

equations and that for the free stream. The k-omega SST turbulence model has been

suggested for swirling flows in centrifugal machines' and is indicated as more accurate

and reliable than the simple k-omega model for adverse pressure gradient flows 2 . A

low-Reynolds formulation ensures boundary layer resolution, for wall distance y+ ~ 1

(within the viscous sublayer).

'Personal communication between B. Glass and M. Casey, 20082 FLUENT user guide, section 12.5.2

17

2.1.2 MISES

To assess return vane performance, Drela's MISES software [7] was employed. MISES

(Multiple blade Interacting Streamtube Euler Solver) is a quasi-2D solver dividing the

flow in an inviscid region, and a viscous boundary layer. Losses include an inviscid

shock-loss component, and a viscous loss component. The loss absolute values differ

from those obtained with FLUENT calculations, and only the trends were used to

provide preliminary guidelines concerning the potential for improvement.

2.2 Mesh Generation

All grids used were created using the BladeGen and TurboGrid capabilities in ANSYS

CFX. Meridional outlines and vane characteristics were defined in BladeGen, then

imported into TurboGrid where the grid was automatically generated. Figure 2-1

displays a typical leading edge mesh created using the automated gridding function

in TurboGrid. The grid generation was supplemented by the use of Pointwise to grid

straight vaneless segments, and all meshes were merged in FLUENT using interfaces

with periodic repeats.

2.3 Inlet Profiles

Spanwise profiles providing a circumferentially 'mixed out' average of the flow were

applied at the inlet to the diffuser and static pressure was specified at the outlet of

the return channel. Glass [1] had defined the impeller flow field, computing the flow

at a selected mixing plane location and the mixed-out average. The mixing plane's

location, downstream from the impeller exit by 20% of the impeller exit width was

set to ensure that the mixed out flow field did not include reverse flow. Previous

work showed that the main source of losses in the diffuser originated from spanwise

variations at the inlet, which are captured by the mixing plane model. See [1] for a

full description of the mixing plane model.

18

x

Figure 2-1: TurboGrid Generated Mesh (Vane Leading Edge)

2.4 Performance Metrics

The loss coefficient metric used to evaluate return channel performance is defined as:

Pt,upstream - Pt,downstream (2.1)(0.5Pinietuiniet)

where (0.5 pinleUnlet) is the dynamic pressure at diffuser inlet, and upstream and

downstream refer to diffuser inlet and 90' bend exit. For each component, quantities

are mass-weighteda at the corresponding upstream and downstream stations.

The stage operating condition is determined by the flow coefficient, which we

define [8] as:

D Q (2.2)id 2

3 As a note on the effect of averaging, for the geometries of interest defining the loss in terms ofarea-averaged quantities, instead of mass-averaged, leads to an approximately 5% increase in losscoefficient.

19

UUmeridionalI

tan gential

Figure 2-2: Swirl Angle Definition

with d2 the mean impeller tip diameter, U2 the impeller tip speed, and Qo the inlet

stagnation volume pressure flow rate, calculated as:

r= (2.3)

Pt,o

Calculations were performed both at the specific design point, and off-design at

flows ranging from <D/ 4 d = 0.89 to <D/ 4 d =1.17.

Figure 2-2 provides a schematic of the swirl angle definition. The swirl velocity

is defined as positive in the direction of the impeller rotation, with the swirl angle a

defined as:

a = tan-' ( (2.4)\s axiali+ ueadial s

Velocities in all subsequent figures are normalized by the impeller tip speed.

20

Chapter 3

Characteristics of the Baseline

Return Passage Flow

3.1 Baseline Loss Mechanisms

Figure 3-1 lists the losses for the different components in the baseline geometry. The

figure shows that the vane section contributes the most (39%) to the overall losses,

followed by the diffuser (29%). The 180' bend and 90' bend had lower loss levels at

18% and 14% respectively.

Both viscous dissipation and separation contribute to losses in the baseline return

channel. Losses due to wall friction are approximately proportional to the cube of

Component (% Baseline Total Loss)

Diffuser 29%

Bend 18%

Vane 39%

900 14%

CUMULATIVE 100%

Figure 3-1: Baseline Loss Components

21

Figure 3-2: Location of Baseline Separation Regions

the velocity and the fluid path length [9]. For flows with a large swirl component

the majority of stagnation pressure losses are thus due to wall friction. Separation

regions are low velocity magnitude regions associated with high entropy generation.

Figure 3-2 gives a schematic of the three main separation regions in the return channel.

" In region A, separation occurs if the diffuser convex curvature is too sharp.

" In region B, separation is the result of the curvature and aggressive diffusion in

the 180' bend.

* In region C, separation originating from region B propagates downstream through

the return channel.

22

3.2 Baseline Flow Features

3.2.1 Diffuser

A typical diffuser should have a radius ratio of at least rtop/rniet > 1.5 to satisfy the

pressure recovery requirement', and to appropriately reduce velocities coming into

the 180' bend. A vaneless diffuser enables a large operating range.

Losses in the diffuser are largely a result of losses from dissipation in the boundary

layers. Velocities (and kinetic energy) are highest at the impeller outlet, and the

diffuser has the overall highest velocities of any of the return channel components. To

decrease velocity levels and reduce losses the diffuser width ratio could be increased,

although this can worsen the separation region on the outbound region of the diffuser

shroud. Reducing velocities without increasing the diffuser width ratio could also be

achieved by extending the diffuser's radial extent. However, there are both constraints

on the machine geometry, that limit the radial extent, and efficiency considerations2

Potential for improvement is limited.

The presence of a separation region on the diffuser shroud (sketched as section A

in Figure 3-2) can be remedied either by using a swept back diffuser and relaxing the

shroud's convex curvature, or by pinching the top of the diffuser to increase velocities.

3.2.2 1800 Bend Section

The kinetic energy in the 180' bend is lower than that in the diffuser. This section

accounts for 18% of the baseline losses, with potential for improvement.

Flow separation in the 180' bend (region B in Figure 3-2) is a result of aggressive

meridional curvature, ie: trying to turn the flow over a short axial distance. The

sharpness of the bend curvature can be roughly quantified as e = 2f/x where fc

is the mean radius of curvature at the top of the 180' bend, and x is the axial

extent from mid-span at diffuser top to mid-span at vane channel inlet. In effect

'Personal communication from Professor M. Casey, 20112The diffuser radius cannot be overly reduced either since this can lead to a decrease in efficiency.

As example efficiency detriments of 3 to 5 % have been seen for a diffuser ratio reduction of 18 %[10].

23

e = 2fc/x captures the bend's ellipticity. An increase in axial length can be beneficial

in reducing the bend curvature, although, again, too large an axial extent can lead

to unacceptable viscous losses.

Flow separation on the hub at the 180' bend exit (Region C in Figure 3-2) results

from an adverse pressure gradient. The blockage induced by the separation region

increases the freestream velocity and the losses. Further, the bend curvature estab-

lishes a spanwise pressure gradient at bend exit, with higher velocities on the hub

and lower velocities on the shroud. As the flow transitions to the vane section it

decelerates at the hub, with a consequent adverse pressure gradient that can cause

separation. The presence and extent of separation at the exit of the 180' bend is

linked to the sharpness of this curvature. Flow distortion at the 1800 bend exit and

the large turning required of the vanes can enhance the tendency for separation in

the vane section.

3.2.3 Vane Section

The role of the return vane is to remove swirl, and the flow angles at vane passage

inlet are typically 450 or more. There is a pressure rise through the vane section even

though the radius and meridional area are reduced.

The baseline displayed an incidence mismatch at the vane leading edge with a

large (> 30') incidence angle at the hub and the computations showing the presence

of separation. Lower but positive incidence angles are present over half the span,

giving additional potential for loss reduction [11].

Figure 3-3 represents normalized radial velocity contours in the baseline vane

section. The figure shows the separation that originates on the hub at the exit of the

180' bend (the inlet of the vane section) and propagates downstream, with reverse flow

on the pressure side of the vane. The flow reattaches near mid-chord on the pressure

side, but the reduced velocities on the suction side create blockage and increase losses

in the vane section.

In the baseline case, the non-uniform nature of the flow angle at the leading

edge created a need for a three-dimensional vane, with metal angle ranging from 50'

24

V" 0.1

0

-0.1

-0.2

-0.3

Y -

-0.4

Figure 3-3: Normalized Radial Velocity in the Baseline Vane Section

25

on the hub to 70' on the shroud. A re-design of the channel allowed for a better

incidence match, reducing separation. (Axisymmetric calculations can be used to

define the incident flow angle on the vane leading edge, and, if desired the baseline

three-dimensional vane could be adjusted with a linear variation in angle to reduce the

incidence. However, a two-dimensional vane is preferred for ease of manufacturing and

may be appropriate if the flow into the return channel can be made more uniform.)

In a multi-stage machine, the performance of a stage is dependent on the previous

stage's exit flow characteristics. The outlet flow from the previous stage should be

axial. For the baseline the mass averaged flow angle at outlet had 0.10 of overturning,

although local angles ranged from ±15' from hub to shroud. No vane geometry

overturning was included in the baseline.

3.2.4 900 Bend

The curvature of the 900 bend led to a spanwise pressure gradient, with higher static

pressure on the hub, and higher velocities on the shroud. Figure 3-4 shows contours

of normalized axial velocity mid-passage. A region of reduced axial velocities is visible

on the shroud of the 900 bend, indicative of incipient separation.

26

VUIVheel 0.4

0.3

0.2

0.1

Y

x--z0

Figure 3-4: Baseline 900 Bend: Normalized Axial Velocity Contours mid-passage

27

28

Chapter 4

Return Passage Concept

Development

The approach taken to develop a new return passage concept consisted of an ini-

tial meridional geometry re-design using both the results from Glass [1] and a new

parametric study, followed by the investigation of additional flow features that were

potentially useful in providing loss reduction.

4.1 2010 Geometry

Figure 4-1 lists changes in losses between Glass's 2010 geometry and the baseline.

The 2010 geometry [1] achieved a 10% calculated overall loss reduction. This occurred

through reducing losses in the 1800 bend by 26% and in the vane section by 20%,

compared to baseline. However, the diffuser section of the return passage, the largest

loss contributor to the overall channel performance, displayed an increase in losses of

6%.

4.1.1 2010 Geometry Flow Features

The 2010 geometry used an oblique diffuser which removed separation on the diffuser

shroud (region A in Figure 3-2). The consequent reduction in blockage also reduced

29

% ImprovementComponent (Compared to

Baseline)

Diffuser t6%

Bend 126%

Vane 120%

900 t18%

CUMULATIVE 110%

Figure 4-1: Performance of 2010 Geometry

the freestream velocity, and hence the boundary layer dissipation, the primary loss

mechanism in the diffuser and 1800 bend. Figure 4-2 displays contours of normalized

velocity magnitude for both baseline and 2010 geometries, on the left hand and right

hand sides of the figure respectively. These show the reduced separation in the 180'

bend for the 2010 geometry.

Compared to the baseline, the 2010 geometry exhibited reduced separation in the

vane section, at the vane leading edge. Figure 4-3 shows contours of normalized radial

velocity at radial stations through the vane section, where positive values indicate

reverse flow (separation). The figure shows that reverse flow in the 2010 geometry

was localized near the vane leading edge, on the hub side, and was no longer present

past the 20% chord, leading to a performance improvement.

Non-zero leading edge incidence angles cause increased acceleration and losses

around the leading edge with positive incidence angles leading to higher losses than

identical negative incidence angles [11]. In the 2010 geometry, the incidence angle

was reduced compared to baseline, and the presence of positive incidence limited to

10% of the span at inlet. The use of a three-dimensional vane was effective for better

incidence matching, but it was desired to eliminate this complexity.

30

VUWheel

0.6

10.3Figure 4-2: Normalized Velocity Contours for Baseline (left) and 2010 Geometry withSwept Diffuser (right), Axisymmetric Calculations

0.1

0

-0.1

-0.2

-0.3

Y

-0.4

Figure 4-3: 2010 Geometry: Normalized Radial Velocity

31

V,,U //U Wh,,e

I'I

4.1.2 Lessons Learned from the 2010 Geometry

The 2010 geometry reduced overall losses using a gradually increasing radius of cur-

vature in the 1800 bend. However, the diffuser performance penalty was undesirable.

In subsequent sections we describe the use of a radial diffuser to remove this efficiency

loss.

4.2 Axisymmetric Parametric Studies

We first assessed the potential for improvement of the 2010 geometry, before investi-

gating additional areas of the design space. The initial step was a parametric study

of the meridional geometry.

There were two objectives of the parametric study. One was to obtain an improved

description of the meridional geometry so trends for varying each parameter could

be established. The second was to develop physical understanding of the impact of

each parameter on the flow, so future designs could be built on integrating individual

trends. The major challenge in the interpretation of the parametric study was the

integration, or interdependency, between the geometrical features.

4.2.1 Parameterization

The meridional geometry can be represented as a series of Bezier patches, in which

the channel walls are defined based on geometrical, slope and curvature parameters

[12]. Figure 4-4 demonstrates how the return channel geometry can be defined using

(as an example) five Bezier patches. Parameters were coded into a Matlab script

which produces hub and shroud outlines for import into a meshing software.

For patches to have four points each, a third order Bernstein polynomial is required

for each patch [13]. The Bernstein polynomials are defined below, with n the order

of the Bernstein polynomial, B, and k an index varying between 0 and n.

B n"(t) = ! k(1 - t-k (.k k!(n - k)!t

32

Figure 4-4: Return Channel Sample Bezier Parameterization

The points on the Bezier curve are defined by the following coordinates:

n

XBezier Z k B (4.2)k=O

n

YBezier Yk Bk (4.3)k=0

Once the geometry was captured using the above parameterization for both hub

and shroud contours, three parameters were selected to serve in establishing loss

coefficient trends over a range of 50-150% of each parameter's baseline value

" Bend axial extent (14 + 16)

" Bend width ratio (b6/b 4)

" Diffuser top width/inlet width (b4/b3 )

These parameters were chosen because they determined the geometry of the 1800

bend, where separation is induced.

33

Figure 4-5: 2010 Design: Study Parameters

4.2.2 Parametric Study Results

Figure 4-5 is a sketch of the 2010 geometry meridional flow path and the parameters

varied, using the same nomenclature as in section 1.3. Results from the parametric

study are shown in Figure 4-6, which displays change in computed loss coefficient ob-

tained from individually varying each of the three parameters listed in section 4.2.1.

The parametric study demonstrated that the 2010 geometry led to a local loss coef-

ficient minimum but that there was little room for further improvement. The main

message was thus that to obtain further loss reduction, there was a need to explore

alternative designs in another area of the design space.

A parametric study was also conducted for the baseline to provide physical in-

sights on trade-offs. Figure 4-7 illustrates the baseline meridional flow path with the

parameters used.

" Bend axial extent 14 + 16. Too short an axial extent leads to separation in the

bend, while too long an axial extent leads to high viscous losses. Stage axial

extent requirements (12) may limit the allowable extent.

" Bend width ratio (b6/b4 ). Too small a bend width ratio leads to separation in

the vane section, while too large a bend ratio also causes separation, but in the

180' bend.

" Diffuser top width/inlet width (b4 /b3). Too small a ratio causes high diffuser

34

A

'-a

- U--

2010 Geometry

70

50

0 20 40 60 80 100 120 140 160 180

Metric Value (100% is Reference Value)

Figure 4-6: Proposed design parameters near (local) minimum loss

Figure 4-7: Baseline: Study Parameters

35

250

* Bend Axial Extent

* Bend Width Ratio

A DiffTop/Inlet Width Ratio

230 1

210U

C

0)190

0

U 170

S130

0 110

90

A

A

viscous losses, while too large a ratio causes separation at the inlet to the 1800

bend.

4.3 Proposed New Geometry

Based on the 2010 geometry flow features and the results of the parametric studies, a

series of axisymmetric calculations were used for preliminary investigation of design

options. These iterations led to a return channel geometry with the following fea-

tures: a radial diffuser, with a pinch at the top, an increased bend axial extent with a

gradual increase in bend curvature, and a swept back vane section. The axisymmet-

ric calculations also highlighted the importance of the b6 (vane channel inlet width)

parameter, as a driver of the trade-off between increased velocities and reduced sep-

aration. The resulting geometry, referred to as 'OptiH' series, is detailed in section

4.4 below.

4.3.1 Diffuser

Meridional Features

The baseline vaneless diffuser has a high radius ratio providing good pressure recov-

ery and a large operating range. To avoid a performance penalty (as in the 2010

geometry), and increased losses where velocity levels are highest (at impeller outlet)

a radial diffuser appeared to be an appropriate candidate. With the given impeller,

however, there is streamline curvature present in the baseline between impeller exit

and approximately 30% of the diffuser height (outer radius - inner radius), yielding

the baseline configuration. This baseline diffuser was selected for the new design, with

the addition of a pinch at diffuser top on the shroud side, to turn the flow earlier and

to reduce separation into the 180' bend. Figure 4-8 shows the 2010 diffuser geometry,

and the proposed new design diffuser geometry described above.

36

2010 NewDesign Design

Figure 4-8: Diffuser Geometry Comparison

Low Solidity Vaned Diffuser (LSVD) as a Potential Design Feature

Vaneless diffusers (VLD) are typically used in applications that target good efficiencies

and a large operating range. Conventional Vaned Diffusers (CVD) offer better static

pressure recovery, and better efficiency, than VLDs, but the former have a reduced

flow range compared to the latter. Further, at off-design conditions there can be

separation from the vanes, resulting in an efficiency drop. Thus, CVDs have high

peak efficiency but small available operating range.

Senoo (quoted in [14]) introduced the concept of a low solidity vaned diffuser

(LSVD). This geometry can be defined by a solidity, chord/spacing < 0.9. LSVD

geometries offer higher efficiency levels than a VLD while maintaining flow range

close to that of the VLD. The LSVD flow rate is typically controlled by the impeller

choking rather than the diffuser, improving the choke margin compared to a CVD.

LSVDs have received limited coverage in the open literature, and no firm design

criteria have been established. Some rough guidelines exist for initial LSVD de-

sign screening, although the effect of varying geometric parameters depends on what

quantities are kept constant so that iterations are required to determine the optimum

parameter values for both the geometry and operating conditions.

To show the overall features of LSVDs Figure 4-9 presents normalized published

data from Hohlweg [15], Amineni [16] and Engeda [17]. The figure portrays the rela-

37

tion between range and efficiency changes for LSVDs compared to vaneless diffusers.

The solidity was o- = 0.7 for all cases, while the number of vanes, the incidence angle

and the inlet Mach number vary between geometries.

As shown in Figure 4-9, depending on different operating conditions and geomet-

rical set-ups, LSVDs are able to provide gains in efficiency (compared to a vaneless

diffuser) varying between 2-6%, if it is allowable to accept reductions in range. Care

needs to be taken to optimize both diffuser and return channel for losses, because the

optimum return channel geometry for an LSVD might differ from that for a vaneless

diffuser. Furthermore, as mentioned, efficiency gains obtained using an LSVD will

have to be traded with a reduction in range compared to a vaneless diffuser. The

latter was important for this particular machine and return channel, in that use of

an LSVD was not recommended because of the associated range decrease.

4.3.2 180' bend

The 2010 geometry increased the 1800 bend axial extent, and used an increasing

radius of curvature throughout the 180' bend to help turn the flow. These features are

retained in the new geometry. Increasing the axial length reduces the bend curvature,

and reduces the likelihood of separation.

The width of the 180' bend impacts the spanwise velocity variations at inlet to

the vane section. A narrower span leads to increased velocity levels, but reduced

spanwise velocity variations, while a larger span leads to lower velocity levels but

large spanwise velocity variations. Determination of the exit width of the 180' bend,

b6 is described in section 4.4.

4.3.3 Vane Section

Meridional Geometry

The three geometries baseline, 2010 and new are given in Figure 4-10. The figure

shows changes in both channel width and bend axial extent, and also gives contours

of the normalized velocity magnitude. In the 2010 geometry the 180' bend axial

38

1 2 3 4

Z=16

5 6

i=-1.9*

A

Z=14O =0.7

Z =# of blades

i = incidence angle

ZZ=16

Z=14

Range =choke =

0

-5L

U1

-10C

>-15~0

m -20

E0U -25toC

CU~-30

-35

-40

4.10

Mu= Tip speed Mach Number

% Efficiency Change Compared to Vaneless Diffuser* Mu=0.53 - Mu=0.7 A Mu=1.38 E Mu=1.02 I Mu=0.69 - Mu=0.88 A Mu=1.02

93-GT-98 (Hohiweg et al) Engeda (2001) 96-GT-155 (Amineni et al)

Figure 4-9: Normalized LSVD Perfomance Characteristics

39

70

i=0.3*

II I

v 0.6

0Baseline 2010 New

Figure 4-10: Width Increase for Different Geometries

length was increased relative to the baseline, meeting stage axial extent requirements

through use of a swept diffuser. In the new geometry, using the baseline radial diffuser

(with a pinch) and increasing the 180' bend axial length requires the vane section to

be swept. The vane section width increases between leading edge and trailing edge

to meet the axial length requirement.

The vane geometry was adapted to provide incidence matching.

Investigation of Potential Benefits from Additional Vane Rows

Two geometries with additional vane rows were investigated to explore potential for

loss reduction. The first was a vane row in the second half of the 1800 bend to take

swirl out earlier in the flow path. The second was a vane row added in the 900 bend

to unload the main return channel vane. In both cases it was found that the velocities

were reduced, but there remained a trade-off between lower velocities and the added

blockage/wall surface area.

The addition of a vane row in the 180' bend was intended to introduce flow

guidance, reducing swirl and velocity magnitude. While benefits were not expected

in the vane (due to the reduced velocity/increased wall friction trade-off), there was

the potential for cumulative benefits [4]. Increased vane loading was observed and no

clear benefits were noted in the return vane section. It can be noted that upstream

extension of the vane has greater potential for loss reductions in geometries with a

40

small diffuser radius ratio such as used by Aalburg et al [4], rather than in geometries

with a higher diffuser radius ratio, as is the case for the geometries of interest in this

project.

The additional vane row in the 900 bend displayed low loading, was not effective

in loss reduction, and was not pursued. Positive and negative lean' options were also

examined for this additional vane row but neither had a significant impact.

In summary, the use of additional vanes did not show benefits and was not pursued

further.

Investigation of Potential Benefits from Reduced Number of Vanes

The large number of vanes in the baseline return channel was set to ensure that

there would be no difficulties in the downstream stages. Given the existing trade-

off between skin friction and vane loading, however, we also examined reducing the

number of vanes to obtain better efficiency.

This vane number reduction was investigated using the MISES blade-to-blade

software. Flow inlet angle and Mach number were kept constant, while the number

of vanes was varied; without vane length adjustments varying pitch is equivalent to

varying solidity.

Calculations with different vane numbers were carried out for both the baseline,

and the OptiH (proposed) geometry, for different flow inlet angles, Mach inlet numbers

and pressure ratios.

Figure 4-11 shows loss coefficients, between the vane leading edge and vane trailing

edge radial locations, as computed in MISES, versus the percent number of vanes

compared to the baseline. As number of vanes is increased, the solidity and wall area

is increased. For lower vane numbers there are increased aerodynamic losses from the

larger velocity. While a small improvement was noted for the baseline configuration,

there was no appreciable improvement observed for the OptiH geometry.

Calculations carried out with MISES were intended as a tool to determine which

'Lean is the angle between the vane and the hub or shroud surface and impacts the hub-shroudpressure gradient, hence redistributing the flow [3].

41

150

u 140 - Baseline -- New Design(U

130

120

110Ma

100

4J90 . -80030 -- -- - ------ ...---ta

0 70U

VO 600

50

60 65 70 75 80 85 90 95 100 105 110

Number of Vanes (% Baseline)

Figure 4-11: Effect of Vane Number Reduction on Performance (MISES Computa-tions)

design to test in FLUENT, and the absolute value are less important than overall

trends. Three-dimensional FLUENT calculations were subsequently performed on

the OptiH geometry with varying vane number. These also showed no significant

improvement in performance (< 3%), consistent with the MISES prediction.

Investigation of Potential Benefits from Vane Loading2

For a typical return channel the major influence on the vane shape is the vane inlet

and outlet angle. The vane inlet angle is approximately 50' from the meridional

direction, with spanwise variations of roughly 15', and the desired vane outlet angle

is 00. To investigate the effect of vane loading distribution the vane angles of the

camber line were defined as a Bezier spline where m is the fraction of the meridional

length (m=0 at the leading edge, m-1 at the trailing edge). Figure 4-12 shows the

angle values for the Bezier points at 0%, 25%, 50%, 75% and 100% of the meridional

length. The vane thickness distribution was maintained at the original value, and the

inlet and outlet angles of the vanes were kept at 65' and 0' respectively.

2Work carried out by Professor M. Casey

42

70

-+-Front Loaded(g 60

+ New Geometry

U s0 +Rear Loaded

C a)40

a)C 30

a)

20

a)-oE

U 10

00 20 40 60 80 100 120

% Meridional Length

Figure 4-12: Different Vane Loading Distributions

The rear loaded case was found to have the highest velocities at the passage throat.

It also had large deviations in outlet angle compared to the front loaded vane. As

such it was not suitable.

The front loaded cases reduce velocities early in the passage, with the highest

velocities near the leading edge. The study showed that losses and deviation are

not sensitive to angle distribution provided the vane is not heavily rear loaded. The

front loaded configurations were found to give better performance, consistent with

common practice [2]; they provide higher static pressure recovery, and better flow

angle uniformity. The front loaded configuration indicated as 'new geometry' in

Figure 4-12 was thus retained.

Increase in Vane Leading Edge Radius Location

Due to wall curvature, at the 180' bend exit the meridional hub velocity is greater

than the meridional shroud velocity. Provided there is no separation at the vane

leading edge, since the swirl is approximately the same near both hub and shroud at

43

At Vane Leading Edge (r varies)90

*===Baseline

85 - - -OptiH-b- OptiHb R+10

----- OptiHc80 -.- - ---- OptiHc R+10

75--7 0 -- - -. ..- . . .. ... . .-. ... . .-. .-.-.-. .-.- -

0-

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1A xial coordinate (Hub to Shroud)

Figure 4-13: Leading Edge Flow Angles for OptiH geometries

a given radius, the variation in meridional velocity results in a lower hub flow angle

than that on the shroud.

As the leading edge radial location is increased, the flow angle at the hub decreases.

Figure 4-13 shows this effect occurring at the leading edge, to different extents, in

the OptiHb and OptiHc series. The different names for the OptiHb and OptiHc

correspond to the vane section inlet width be. Configurations with an increased radial

location are indicated by 'R+'. OptiHb has b6 63% of vane trailing edge width and

OptiHc has b6 = 76% of vane trailing edge width (see section 4.4 for a description

of the OptiH geometry series). When the vane is extended upstream, it needs to be

adapted to the hub and shroud contours, and the vane angle on the hub is lower than

the vane angle on the shroud, better matching the flow angle distribution.

In sum, moving the vane upstream was found to provide a better incidence match.

Further, moving the leading edge radially outwards reduces peak flow angle and

incidence. This is most advantageous for geometries with a non-uniform profile at

the inlet to the return vane, ie, geometries with a large be value. For geometries with

44

Figure 4-14: OptiH Series b6 Values

smaller b6 , and more uniform profiles, there is less performance gain from moving the

leading edge radius outward.

4.3.4 900 bend

Some performance may be gained by following the same guidelines as for the 180'

bend, but since the 900 bend is the smallest contributor to the overall losses, the gain

was viewed as small and no design changes were made to the 900 bend.

4.4 OptiH Series

For the OptiH series three-dimensional calculations were used to select the b6 width

values. A series of OptiH type geometries with varying b6 values were assessed.

The different b6 values investigated are given in Figure 4-14. Figure 4-15 shows the

corresponding meridional flow paths for OptiHa, OptiHb and OptiHc. In the study

only the second half of the 1800 bend was modified.

Figure 4-16 displays return channel losses for the OptiH geometry series, compared

to the baseline, as a function of the b6 width value. For b6 values close to that of

the baseline, the OptiH series provides a substantial reduction in losses. The b6 value

with the largest reduction was found at 62.5% of the baseline b6 .

Differences in performance within the OptiH series can be explained by examining

velocity levels in the vane passage. For geometries with small b6 such as OptiHa, the

velocities in the vane channel are high and there is almost no separation. Figure 4-173This was suggested by A. Nakaniwa, personal communication, 2012

45

OptiHz OptiHa OptiHb OptiHc Baseline OptiHd

b6 44.9 50.9 62.5 75.9 76.8 87.5(% TE width)

OptiHa OptiHb OptiHc

Figure 4-15: OptiH series geometries

110OptiH-z

0

OptiH-d

OptiH-a

OptiH-b OptiH-c

0-OptiH Series

* Baseline

40 45 50 55 60 65 70 75 80 85 90

b6 (% Trailing Edge Width)

Figure 4-16: Total Loss as a Function of b6 for OptiH Geometries

46

105

1000

~95

90

85

80

0 75

4-JP 70

65

60

Vrad U 0.1IUWheel

0

-0.1

-0.2

-0.3Y _

-0.4

Figure 4-17: OptiH-a: Normalized Radial Velocity

shows the normalized radial velocity contours in the OptiHa geometry. There are

separation regions (positive velocity values) present on the suction side downstream

of the 60% chord and a small separation region close to the leading edge. The major

part of the losses however, are a result of the high velocities and friction loss.

Figure 4-18 shows normalized radial velocity contours for OptiHc, which has an

increased b6 . There is again separation evident at the vane leading edge, most notice-

ably on the hub, and there is downstream propagation. However, the velocity levels

are lower than in OptiHa.

The effect of b6 is even more pronounced in the OptiHd results given in Figure 4-19.

The normalized radial velocity contours show the spanwise extent of the separation

at the leading edge has increased, as did the separation propagating downstream up

to the trailing edge.

The amount of flow distortion at the exit of the 1800 bend is related to the area

ratio at the exit of the bend. The non-uniformity in the flow profile into the return

47

V/rad U/U Wheel

0.1

0

-0.1

-0.2

-0.3

Y

-0.4

Figure 4-18: OptiH-c: Normalized Radial Velocity

channel increases as the b6 value increases (OptiHc or OptiHd), leading to a higher

risk of separation. For geometries with lower b6 values, for example OptiHb, there

is lower likelihood of separation. However, the velocity levels (and viscous losses)

are increased compared to the wider channels. This trade-off between high velocities

and potential for separation is an essential part of the iterations to define the most

appropriate geometry. Figure 4-20 displays radial velocity contours for the OptiHb

geometry, which is seen as an effective compromise between the high velocities of

OptiHa and the separation present in OptiHd.

4.5 Final Geometry: OptiHb R+10

4.5.1 Geometry Features

The final geometry selected, denoted as OptiHb R+10, incorporates the following

features. First, the baseline diffuser geometry with an additional pinch at the exit

48

I

V,ad /0.1IUWheel

0

-0.1

-0.2

-0.3

-0.4

Figure 4-19: OptiH-d: Normalized Radial Velocity

is used. The diffuser thus becomes narrower on the shroud side at the beginning of

the 180' bend. Pinching on the shroud side is beneficial to the flow in the 180' bend.

Second, the 180' bend has the maximum axial length allowed by the geometrical

constraints, with a progressively increasing radius of curvature. The width of the

bend at the top is 80% of the diffuser initial width.

The proposed geometry uses a b6 value which is 62.5% of the trailing edge width.

This b6 value reduces viscous losses in the bend and vane section as well as separation

at the onset of the return vane section. The return vane section walls are sloped

towards the stage inlet to meet the geometrical constraint on axial length. The

increase in the vane leading edge radius location also allows for reduced vane incidence,

as described in section 4.3.3.

49

V, Wheel

Y-0.4

Figure 4-20: OptiH-b: Normalized Radial Velocity

A( (% Change Compared toComponent Baseline)

Diffuser 18%

Bend t22%

Vane 160%

900 119%CUMULATIVE 119%

Figure 4-21: Performance of OptiHb R+10

50

0.1

0

-0.1

-0.2

-0.3

120

0jj 110

90

0

U

080

-- Baseline

70e OptiH lb R+10

0 OptiHc R+10

60

85 90 95 100 105 110 115

% Design Flow Coefficient

Figure 4-22: Off-Design Performance Comparison: Baseline, OptiHb R+10 and Op-tiHc R+10

4.5.2 Loss Reduction Results

The 2010 geometry increased diffuser losses by 6%, reduced bend and vane losses by

26% and 20%, and increased 90' bend losses by 18%. Figure 4-21 shows the results for

the proposed new geometry (OptiHb R+10). The losses in the diffuser were reduced

by 8%. There are increased losses in the 180' bend and 90' bend by 22% and 19%

respectively, but there are reduced losses in the vane passage by 60%. The result is

an overall 19% loss reduction from the baseline, compared to the 2010 geometry's

10%.

All the loss coefficient comparisons quoted above are at design, but off-design

computations have also been carried out to assess the OptiHb R+10 and OptiH-

R+10 configurations. Figure 4-22 presents the performance of both OptiHb R+10

and OptiHe R+10, as a percent of baseline loss coefficient at design flow, over the

flow coefficient range of interest. The figure shows that for flow coefficients at 90 % of

design, and up to 110% of design, OptiHb R+10 performed consistently better than

51

OptiHe R+10 with an additional 2% of loss reduction. At flow coefficient greater than

110% design OptiHc R+10 performed better than OptiHb R+10, although worse than

the baseline. Given that the machine's range of operation is largely below 110% flow

coefficient, the consistent advantage obtained from OptiHb R+10 was deemed to be

sufficient, and this geometry was selected.

It is planned that the proposed concept be assessed experimentally at Mitsubishi

Heavy Industries.

52

Chapter 5

Assessment of Return Passage with

Modified Diffuser Inlet Geometry

5.1 Previous Work

Glass described the investigation of a return channel configuration where the diffuser

inlet geometry and flow conditions differed from that currently imposed [1]. Figure 5-

1 is a sketch of the diffuser inlet. For that geometry, referred to as FD, the diffuser

inlet was rotated by 35' from the baseline. The computations showed a roughly 20%

decrease in loss, almost twice the improvement obtained from the changes that were

made in the return channel.

Modifying the diffuser inlet flow to reduce return channel losses affects the im-

peller losses. To examine this in depth a full impeller and return channel calculation

should be carried out. The intent here, however, is to determine the potential for loss

reduction in the diffuser from impeller design changes.

The FD geometry defined by Glass had the straight, leftward swept walls shown

in Figure 5-1. The meridional component of the flow was parallel to the diffuser walls.

To obtain the inlet conditions for the return passage, the inlet profile was modified.

Mass flow, stagnation enthalpy flux, angular momentum flux, and radial momentum

flux were matched with that of the baseline, resulting in the axial extent of the FD

diffuser inlet increasing by 9% compared to baseline [1].

53

FD

Baseline

A/

Figure 5-1: Baseline and FD Diffuser Inlet Geometries

5.2 Modified Diffuser Inlet

The effect of inlet modifications in the present study were assessed using the OptiHb

R+10 return channel, which had yielded the 19% overall loss improvement with the

existing inlet condition. Figure 5-2 indicates the diffuser portion of the geometry

which was replaced by a modified inlet section with radial diffuser walls. The new

inlet location has a radial location at the mean radius of the original inlet, and an

axial location based on the axial shift needed to maintain the location (and pinch) of

the diffuser top. Mass flow, stagnation enthalpy flux, angular momentum flux, and

radial momentum flux were matched with that of baseline, with the axial extent of

the inlet increased by 7.5%.

The use of the radial diffuser led to an additional 4% reduction in loss coefficient

from OptiHb R+10, an overall reduction of 23% compared to baseline. The loss

breakdown is shown in Figure 5-3.

The flow features of the geometry with modified diffuser inlet are qualitatively

similar to those of the OptiHb R+10 geometry. However, there is now no axial

component of the velocity at inlet, and as shown in Figure 5-4 and Figure 5-5 the

54

SModifiedDiffuserInlet

Figure 5-2: OptiHb R+10 and Modified Diffuser Inlet Geometries

A( (% Change Compared toComponent Baseline)

Diffuser 113%

Bend t 6%

Vane 162%

900 123%CUMULATIVE 123%

Figure 5-3: Modified Diffuser Inlet Geometry

55

Vtan, Diffuser Top

-0.4'0 0.2 0.4 0.6 0.8 1

Axial Length (Shroud to Hub)

Vax,Diffuser TopA -

0.035-

0.03

( 0.025

0.02

S0.015E

z 0.01

0.005

0 0.2 0.4 0.6 0.8Axial Length (Shroud to Hub)

Figure 5-4: Tangential and Axial Velocity Levels Comparison at Diffuser Top

Vtan, Vane LE VVane LE

0.2 0.4 0.6 0.8Axial Length (Shroud to Hub)

E0z

0.2 0.4 0.6 0.8Axial Length (Shroud to Hub)

Figure 5-5: Tangential and Axial Velocity Levels Comparison at Vane LE

56

-0.15

-0.2

-0.25a>C

a)-0.3

z -0.35-- pibR1 Raia Inle

-- -.-..-.... Op.~ R+10 ..

OptiHb R+10 Radial Inlet|

Ca)

C-a

a)

E0z

-

-

-

1

velocity levels differ at the top of the diffuser and at the vane leading edge. The

lower velocities throughout the diffuser with the radial inlet geometry account for the

observed 4% reduction in losses.

The results from this modified diffuser are encouraging. Using a radial inlet,

however, will require changes to the impeller geometry, and it is hoped that changes

of this type can be explored in the future.

57

58

Chapter 6

Summary and Proposed Future

Work

6.1 Summary

This thesis has assessed the potential for efficiency improvement in multi-stage cen-

trifugal compressor return channels. Calculations for different channel geometries

were carried out, and an overall 19% improvement compared to baseline was achieved,

an additional 9% improvement in performance compared to previous work. The pro-

posed geometry was found to perform better than the baseline from 90% to 110% of

the design flow coefficient.

The geometric features were an essentially radial diffuser, pinched at the top, a

180' bend with a progressively increasing radius of curvature, a tailored bend exit

width to reduce viscous losses and separation, and a swept back vane channel. An

additional loss reduction mechanism was the adjustment of the vane leading edge

radial location.

To assess the potential for further loss reduction the proposed geometry was as-

sessed for operation with a radial inlet, leading to an overall 23% loss reduction from

baseline. Such a modification implies a change in impeller as well and results suggest

that the impeller should be a target for future proposed work.

59

6.2 Future Work

The use of a radial diffuser inlet shows potential for loss reduction, but it is not

known how this change will affect the losses in the impeller. The impeller and return

channel thus need to be considered as a whole, and the impeller redesigned such that

the overall losses would be reduced. This is proposed as a main target for future

work.

The work in this thesis showed a substantial reduction in the return channel

losses, but it was done by trial and error. As such no guarantee exists that the

selected geometry yields an absolute loss minimum. Parts of the design space have

been explored based on lessons learned, and the observation of flow properties, but

an automated optimization process would provide a desirably much more thorough

investigation of the design space. Adjoint methods may be able to help with this

type of optimization, as they allow the optimization of one output (overall loss) with

respect to many inputs (shape parameters, flow properties). The use of a RANS solver

with adjoint method capabilities would allow the channel shape to be optimized with

respect to an objective function (ie: total pressure loss). This is another area proposed

for future work.

60

Bibliography

[1] Glass B. "Improved Return Passages for Multistage Centrifugal Compressors".

Master's thesis, MIT, 2010.

[2] Reutter 0., Hildebrandt A., Raitor R. Jakiel C., and Voss C. "Automated

Aerodynamic Optimization of a Return Channel Vane of a Multistage Radial

Compressor". European Turbomachinery Conference 9, 45, 2011.

[3] Veress A. and Van Den Braembussche R. "Inverse Design and Optimization of

a Return Channel for a Multistage Centrifugal Compressor". Journal of Fluids

Engineering, 126, 2004.

[4] Aalburg C., Simpson A., Carretero J., Nguyen T., and Michelassi V. "Extension

of the Stator Vane Upstream Across the 180 Degree Bend for a Multistage Radial

Compressor Stage". Proceedings of ASME Turbo Expo 2009: Power for Land,

Sea and Air, 2009.

[5] Simpson A., Aalburg C., Schmitz M., Pannekeet R., Larisch F., and Michelassi

V. "Application of Flow Control in a Novel Sector Test Rig". Proceedings of

ASME Turbo Expo 2009: Power for Land, Sea and Air, 2009.

[6] Menter F. "Improved Two-Equation K-Omega Turbulence Models for Aerody-

namic Flows". Tech. Rep. TM 103975 NASA, 1992.

[7] Drela M. and Youngren H. "A User's Guide to MISES 2.63". As of, February

2008.

61

[8] Masutani J., Koga J., and Kawashima Y. "Development of High-Performance,

High-Speed, Compact Centrifugal Compressors Stage". Mitsubishi Heavy Indus-

tries, Ltd. Technical Review, 36, 1999.

[9] Denton J.D. "Loss Mechanisms in Turbomachines". IGTI Scholar Lecture, 1993.

[10] Lindner P. "Aerodynamic Tests on Centrifugal Process Compressors - Influence

of Diffusor Diameter Ratio, Axial Stage Pitch, and Impeller Cutback". ASME,

105, 1983.

[11] NASA. "Turbine Design and Application". NASA-SP-290, N95-22341, 1972.

[12] Casey M. "A Computational Geometry for the Blades and Internal Flow Chan-

nels of Centrifugal Compressors". Transactions of the ASME, 105, 1983.

[13] Zorin. D. "Bezier Curves and B-splines, Blossoming". Lecture, 2002.

[14] Osborne C. and Sorokes JM. "The Application of Low Solidity Diffusers in

Centrifugal Compressors". ASME FED, Flows in Non-Rotating Turbomachinery

Components, 69, 1988.

[15] Hohlweg W., Direnzi G., and Aungier R. "Comparison of Conventional and Low

Solidity Vaned Diffusers". International Gas Turbine and Aeroengine Congress

and Exposition, 93-GT-98, 1993.

[16] Amineni N., Engeda A., Hohlweg W., and Direnzi G. "Performance of Low

Solidity and Conventional Diffuser Ssystems for Centrifugal Compressors". In-

ternational Gas Turbine and Aeroenginc Congress and Exposition, 96-GT-155,

1996.

[17] Engeda A. "The Design and Performance Results of Simple Flat Plate Low

Solidity Vaned Diffusers". Institution of Mechanical Engineers, Part A: Journal

of Power and Energy, 215:109, 2001.

62

[18] Ohsaki H., Hiraishi E., Hashizume K., Noda S., and Masutani J. "Mitsubishi

Centrifugal Compressors and Steam Turbines for Mega Ethylene Plants". Mit-

subishi Heavy Industries Ltd, Technical Review, 41, 2004.

[19] Aalburg C., Simpson A., and Schmitz M. "Design and Testing of Multistage

Centrifugal Compressors with Small Diffusion Ratios". Proceedings of ASME

Turbo Expo 2008: Power for Land, Sea and Air, 2008.

[20] Kmecl T. and Dalbert P. "Optimization of a Vaned Diffuser Geometry for Radial

Compressors, Part 1: Investigation of the Influence of Geometry Parameters on

Performance of a Diffuser". ASME, 437, 1999.

[21] Dawes W. N. "A Simulation of the Unsteady Interaction of a Centrifugal Impeller

with its Vaned Diffuser: Flow Analysis". ASME Journal of Turbomachinery, 117,

pp213-222, April 1995.

[22] Denton J.D. and Xu L. "The Effects of Lean and Sweep on Transonic Fan

Performance". Proceedings of ASME Turbo Expo, GT-2002-30327, 2002.

[23] Denton J.D. "Some Limitations of Turbomachinery CFD". Proceedings of ASME

Turbo Expo, GT-2010, 2010.

[24] Casey M. and Gersbach F. "An Optimization Technique for Radial Compressor

Impellers". Proceedings of ASME Turbo Expo, GT2008-50561, 2008.

[25] Casey M. "Radial Compressor Stages for Low Flow Coefficients". Institution of

Mechanical Engineers, C403/004, 1990.

63