Computational Analysis of Stall and Separation Control in Centrifugal Compressors

1
Computational Analysis of Stall and Separation
Control in Centrifugal Compressors
Presented By Alexander Stein School of
Aerospace Engineering Georgia Institute of
Technology Supported by the U.S. Army Research
Office Under the Multidisciplinary University
Research Initiative (MURI) on Intelligent Turbine
Engines

2
Outline of Presentation

• Research objectives and motivation
• Background of compressor control
• Introduction of numerical tools
• Configurations and validation results
• DLR high-speed centrifugal compressor (DLRCC)
• NASA Glenn low-speed centrifugal compressor
  (LSCC)
• Off-design results without control
• Surge analysis
• Off-design results with air injection control
• Steady jets
• Pulsed jets
• Conclusions and recommendations

3
Motivation and Objectives

• Use CFD to explore and understand compressor
  stall and surge
• Develop and test control strategies (air
  injection) for centrifugal compressors
• Apply CFD to compare low-speed and high-speed
  configurations

4
Motivation and Objectives
Compressor instabilities can cause fatigue and
damage to entire engine
5
Greitzers Phenomenological Model

• Assumptions
• Compressor modeled as actuator disk
• Fluid inertia contained in pipes
• Spring-like properties confined to plenum

Helmholtz-Resonator Model
Non-dimensional B-Parameter (Greitzer)
B lt Bcritical gt Rotating Stall B gt Bcritical gt
Surge
6
What is Surge?
Mild Surge
Deep Surge
7
How to Alleviate Surge

• Diffuser Bleed Valves
• Pinsley, Greitzer, Epstein (MIT)
• Prasad, Neumeier, Haddad (GT)
• Movable Plenum Wall
• Gysling, Greitzer, Epstein (MIT)
• Guide Vanes
• Dussourd (Ingersoll-Rand Research Inc.)
• Air Injection
• Murray (CalTech)
• Weigl, Paduano, Bright (NASA Glenn)
• Fleeter, Lawless (Purdue)

8
Numerical Formulation (Flow Solver)
Reynolds-averaged Navier-Stokes equations in
finite volume representation

• q is the state vector.
• E, F, and G are the inviscid fluxes (3rd order
  accurate). R, S, and T are the viscous fluxes
  (2nd order accurate).
• A one-equation Spalart-Allmaras model is used.
• Code can handle multiple computational blocks and
  rotor-stator-interaction.

9
Boundary Conditions (Flow Solver)
Periodic boundary at clearance gap
Solid wall boundary at impeller blades
Solid wall boundary at compressor casing
Inflow boundary
Periodic boundary at diffuser
Solid wall boundary at compressor hub
Outflow boundary (coupling with plenum)
Periodic boundary at compressor inlet
10
Outflow Boundary Condition (Flow Solver)
Conservation of mass and isentropic expression
for speed of sound
11
NASA Low-Speed Centrifugal Compressor

• Designed and tested at NASA Glenn
• Mild pressure ratio
• Ideal CFD test case

12
NASA Low-Speed Centrifugal Compressor

• 20 Main blades
• 55? Backsweep
• Grid 129 x 61 x 49 (400,000 nodes)
• A grid sensitivity study was done with up to 3.2
  Million nodes.
• Design Conditions
• 1,862 RPM
• Mass flow 30 kg/s
• Total pressure ratio 1.19
• Adiab. efficiency 92.2
• Tip speed 492 m/s
• Inlet Mrel 0.31

13
Validation Results (Low-Speed)Blade Pressure
Computations vs. Measurements
5 Span
49 Span
79 Span
p/p
Meridional Chord
14
Validation Results (Low-Speed)Blade Pressure
Computations vs. Measurements
5 Span
49 Span
79 Span
p/p
Meridional Chord
15
DLR High-Speed Centrifugal Compressor

• Designed and tested by DLR
• High pressure ratio
• AGARD test case

16
DLR High-Speed Centrifugal Compressor

• 24 Main blades
• 30? Backsweep
• Grid 141 x 49 x 33 (230,000 nodes)
• A grid sensitivity study was done with up to 1.8
  Million nodes.
• Design Conditions
• 22,360 RPM
• Mass flow 4.0 kg/s
• Total pressure ratio 4.7
• Adiab. efficiency 83
• Exit tip speed 468 m/s
• Inlet Mrel 0.92

17
Validation Results (High-Speed) Static Pressure
Along Shroud
Excellent agreement between CFD and experiment
Local Static Pressure, p/pstd
18
Validation Results (High-Speed)
Same momentum deficit was observed experimentally
in other configurations.
Near suction side
Mid-passage
Near pressure side
19
Off-Design Results (High-Speed) Performance
Characteristic Map
Computational and experimental data are within 5
Fluctuations at 3.2 kg/sec are 23 times larger
than at 4.6 kg/sec
20
Off-Design Results (High-Speed) Performance
Characteristic Map
Large limit cycle oscillations develop
Oscillations remain bound gt mild surge
21
Off-Design Results (High-Speed) Mass Flow
Fluctuations
Mild surge cycles develop
Surge amplitude grows to 60 of mean flow rate
Surge frequency 90 Hz (1/100 of blade passing
frequency)
22
Off-Design Results (High-Speed)
Velocity vectors colored by Mrel at mid-passage
Flowfield vectors show a large separation zone
near the leading edge
23
Off-Design Results (High-Speed) Stagnation
Pressure Contours

• Vortex shedding causes reversed flow
• Origin of separation occurs at leading edge
  pressure side

24
Off-Design Results (Low-Speed) Velocity Vectors
at Design Speed
Flowfield stalls but no surge occurs This is in
accordance with Greitzers B-Criterion
25
Off-Design Results (Low-Speed)
Velocity vectors at 200 design speed at
mid-passage
26
Off-Design Results (Low-Speed) Performance
Characteristic Map
Unsteady fluctuations are denoted by size of
circles
Surge fluctuations at 200 design speed are 7
times larger than at 100 design speed
27
Off-Design Results (Low-Speed)Comparison of
Different Shaft Speed

• Conclusions
• Compressibility effects are fundamental for surge
• For surge to occur B gt Bcritical

28
Air Injection Setup
Systematic study injection rate and yaw angle
were identified as the most sensitive
parameters. Related work Rolls Royce, Cal Tech,
NASA Glenn /MIT,
29
Air Injection Results (High-Speed) Different Yaw
Angles, 3 Injected Mass Flow Rate
Yaw angle directly affects incidence angle gt
Maximum control for designer
30
Air Injection Results (High-Speed) Different Yaw
Angles, 3 Injected Mass Flow Rate
Optimum yaw angle of 7.5deg. yields best result
Mass Flow (kg/sec)
Rotor Revolutions, wt/2p
Reduction in Surge Amplitude ()
Positive yaw angle is measured in opposite
direction of impeller rotation
Yaw Angle (Degree)
31
Air Injection Results (High-Speed)
Velocity vectors colored by Mrel at mid-passage
Leading edge separation is suppressed by injection
32
Air Injection Results (High-Speed)
Leading edge reversed flow regions has vanished
33
Air Injection Results (Parametric Studies)
High-Speed Compressor
Low-Speed Compressor
Nondim. Surge Amplitude ()
Nondim. Surge Amplitude ()
Yaw Angle (Deg.)
Yaw Angle (Deg.)
Injection Rate ()
Injection Rate ()

• An optimum yaw angle exists for both compressors.
• A reasonable amount (3 to 5) of injected air is
  sufficient in both configurations to suppress
  surge.

34
Air Injection Results (Neural Network Model)

• A Neural Network can be trained to model the
  injection maps
• Include more parameters (shaft speed, throttle
  settings, etc.)
• Use NN-model as a controller in a real engine
• Training of such a controller by CFD is much
  cheaper than by experiments

35
Air Injection Results (Neural Network Model)
High-Speed Compressor
Low-Speed Compressor
Nondim. Surge Amplitude ()
Nondim. Surge Amplitude ()
Yaw Angle (Deg.)
Yaw Angle (Deg.)
Injection Rate ()
Injection Rate ()
Reasonable agreement between CFD injection
performance maps and NN models is observed.
36
Air Injection Results (Pulsed Jets)
Surge fluctuations decrease as long as the
injection phase was lagged 180 deg. relative to
the flow gt suggests feedback control
Nondim. Surge Fluctuations ()
Rotor Revolutions, wt/2p
37
Air Injection Results (Pulsed Jets)
Amplitude of pulsed jets has a stronger impact
than mean injection rate gt reduction in
external air requirements by 50
Nondim. Surge Fluctuations ()
Rotor Revolutions, wt/2p
38
Air Injection Results (Pulsed Jets)
A short boost from the injected air is sufficient
to suppress surge onset
39
Air Injection Results (Pulsed Jets)
No separation occurs
40
Air Injection Results (Pulsed Jets)

• Jets pulsed at higher frequencies are more
  effective than low-frequency jets (increased
  mixing, higher turbulent intensity)
• There is a practical limitation on the highest
  possible frequency

Nondim. Surge Fluctuations ()
Rotor Revolutions, wt/2p
41
Air Injection Results (Pulsed Jets)
1.5 injected mass is sufficient to suppress surge
Nondim. Surge Fluctuations ()
Rotor Revolutions, wt/2p
42
Conclusions

• A Viscous flow solver has been developed to
• obtain a detailed understanding of surge in
  centrifugal compressors.
• determine fluid dynamic factors that lead to
  stall onset.
• The non-dimensional B-Parameter is a useful way
  to determine a priori which configuration will
  surge.
• Steady jets are effective means of controlling
  surge
• Alter local incidence angles and suppress
  boundary layer separation.
• Yawed jets are more effective than parallel jets.
• An optimum yaw angle exists for each
  configuration.
• Air injection can be modeled by a multi-parameter
  neural network.
• Pulsed jets yield additional performance
  enhancements
• Lead to a reduction in external air requirements.
• Jets pulsed at higher frequencies perform better
  than low-frequency jets.

43
Recommendations

• Perform studies that link air injection rates to
  surge amplitude via a feedback control law.
• Use flow solver to analyze and optimize other
  control strategies, e.g. inlet guide vanes,
  synthetic jets, casing treatments.
• Employ multi-passage flow simulations to study
  rotating stall and appropriate control
  strategies.
• Study inflow distortion and its effects on stall
  inception.
• Improve turbulence modeling of current generation
  turbomachinery solvers. Analyze the feasibility
  of LES methods.

44
How to Control Surge (Active Control)
45
Literature Survey on Air Injection

• Rolls Royce (Day et al., 1997)
• Injection into Tip Region is More Effective than
  Injection into the Core Flow
• Cal Tech (Murray et al., 1997)
• Steady Air Injection Reduces Bandwidth
  Requirements for Bleed Valves
• NASA/MIT (Bright et al., 2000)
• Effectiveness of Air Injectors is Independent of
• 1.) Azimuthal Jet Arrangement
• 2.) Number of Jets

46
Numerical Formulation (Flow Solver)
A Four Point Stencil is Used to Compute the
Inviscid Flux Terms at the Cell Faces According
to Roes Flux Splitting Scheme
Third-Order Accurate in Space

• Turbulence is Modeled by One-Equation
  Spalart-Allmaras Model
• Code Can Handle Multiple Computational Blocks and
  Rotor-Stator-Interaction

47
Overview of Configurations
48
The Present Approach
The Tools
The Results
49
Validation Results (Low-Speed) Velocity Vectors
in Meridional Planes
Clearance Gap Flow Produces Velocity Deficit
Trailing Edge
Leading Edge
Same Phenomenon was Observed Experimentally
Wake-like momentum deficit
97 away from Pressure Side
4 away from Pressure Side
50 away from Pressure Side
50
Validation Results (High-Speed)
51
Validation Results (High-Speed)
52
Eigenmode Analysis (GTSYS3D)

• Calculates eigenvalues/-vectors of the
  compression system matrix
• Based on small perturbation Euler model
• q q0 dq
• The resulting form is
• d/dt(dq) Adq
• where - dq is the state vector of small
  perturbations
• - A is the system matrix of size
• 5N1N2N3 x 5N1N2N3

53
Off-Design Results (High-Speed)System
eigenvalues at stable condition (4.6 kg/sec)

• Mostly acoustic modes with Re lt 0 (damping,
  stable)
• Complex conjugate pairs are oscillatory
• Simple poles (Im 0) near origin are unstable

54
Off-Design Results (High-Speed) System
eigenvalues during surge cycle
At beginning of surge cycle
After 25 of surge cycle

• Most acoustic (damping) modes have vanished
• Simple pole at origin destabilizes system

After 50 of surge cycle
After 75 of surge cycle