Mitsubishi Heavy Industries Technical Review Vol. 52 No.
1 (March 2015)
1
Computational Fluid Dynamics Technology Applied to
High Performance, Reliable Axial Compressors
for Power Generation Gas Turbines
RYOSUKE MITO*1 XUELIN GAO*2
THOMAS WALKER*2 YASURO SAKAMOTO*3
HIROYUKI HAMANA*4
When developing high-performance, reliable gas turbine compressors, repeated cycles of
experimentation, verification, and analysis are used to eventually determine the optimum design.
Currently, thanks to advances in computational fluid dynamics (CFD) technology, full-stage
analysis results can be obtained within a practical amount of time. This allows for the loading
distribution and internal flow phenomena for all stages to be computed accurately and understood
in a quantitative fashion during the design phase. This is not only for the rated load condition, but
also for the partial load condition meaning that using these results as part of the design
improvement process can reduce the required experimental verification cases. Given that the role
of CFD analysis is expanding, Mitsubishi Heavy Industries, Ltd. (MHI) has made efforts to
improve the analysis accuracy. This paper discusses recent improvements in CFD methods and
illustrates some applications for the improved technology.
|1. Introduction
In recent years, power generation gas turbines are beginning to be required to extend
operability through a larger part load operation range as well as improve efficiency at the rated load
condition. Therefore, high-performance compressor development now requires a more precise,
quantitative understanding of the flow conditions. Specifically, some stages under partial load
conditions have a flow field that is nearly stalled. This implies that the prediction of the
compressor’s operating range requires evaluating the entire compressor using three-dimensional
CFD. So far, however, due to time restrictions, three-dimensional CFD has only been applied to
some of the operating range or some of the compressor stages and thus it could not be used to
accurately predict all of the relevant compressor flow behavior.
Given this, instead of using a conventional CPU (central processing unit) for CFD, through
open innovation MHI has introduced a new ultra-high-speed parallel computing technology which
utilizes a GPU (graphics processing unit) and a new analysis code designed to take advantage of
GPU technology. This advancement in simulation technology was applied to our turbomachinery
well ahead of any other organization in the world. With this technology, coupled with the continued
development of GPUs, CFD calculation time was reduced to one twentieth of that of existing
CPUs. Furthermore, the prediction accuracy was also improved via modifying some of the flow
models used in the CFD code. These modifications where verified by showing that the resulting
flow field was close to the actual internal flow phenomena obtained from detailed experimental
data. As a result, CFD is now being applied to many more aspects of turbomachinery design and a
gas turbine multistage compressor can be evaluated in detail from its inlet through to its outlet.
*1 Chief Staff Manager, Takasago Research & Development Center, Technology & Innovation Headquarters
*2 Takasago Research & Development Center, Technology & Innovation Headquarters
*3 Manager Takasago Research & Development Center, Technology & Innovation Headquarters
*4 Mitsubishi Hitachi Power Systems, Ltd.
Mitsubishi Heavy Industries Technical Review Vol. 52 No. 1 (March 2015)
2
|2. Advanced CFD Technology for Turbomachinery
2.1 Ultra-high-speed parallel computing technology using GPUs
For a given set of commands, GPUs have the ability to process a much larger amount of data
in parallel compared to the conventional CPU. The new parallel computing technology works by
combining the command (and logic) functionality of the CPU with the large amount data
processing capacity of the GPU (Figure 1). Furthermore, MHI has followed the introduction of this
new technology with improvements to the CFD analysis code to be used with the GPUs. This,
combined with the continuing development of GPU technology, has led to reduced calculation time
(about one twentieth of conventional CPU computation). This allows the application of full-stage
CFD analysis at a practical stage in the compressor design process. In addition, it is now possible to
predict the stall point of a multistage compressor within a practical amount of time through
unsteady CFD analysis.
Figure 1 GPU parallel computing technology (20 times faster than conventional CPU
computing technology)
2.2 Detailed simulation of the compressor’s real geometry
Due to the reduction of calculation time it is now possible to run CFD on a large calculation
mesh. This allows for geometries that tended to be simplified in the past, such as cavities and
clearances, to be simulated in detail (Figure 2).
Figure 2 Compressor CFD model(main flow + clearance)
Mitsubishi Heavy Industries Technical Review Vol. 52 No. 1 (March 2015)
3
2.3 Acquisition of detailed measurement data and improvement of CFD methods
The internal flow conditions in a power generation gas turbine compressor change as the
flow moves from the front stages to the rear stages. In the front stages, the flow is transonic,
meaning that prediction of shockwaves generated by the rotor is important when computing
compressor performance characteristics. In the rear stages, the flow is subsonic, meaning that the
secondary flow near the endwall and the rotor tip clearance flow are the most important flow
features. In addition, the pressure ratio of each stage increases or decreases compared to the design
point depending on the power output requirement. Since the compressor is operated at a constant
speed, the power requirement changes based on air temperature, season, and time of day. The
performance characteristic as well as the internal flow of the compressor is interdependent on each
stage and thus, accurate data acquisition for each stage is necessary for the prediction of the
full-stage performance. For this reason, MHI uses several test rigs to reproduce the front stage as
well as the middle/rear stage flow in order to acquire performance characteristic and internal flow
data. This is not only under rated load conditions, but also at high pressure conditions, low speed
conditions, and conditions with enlarged rotor tip clearances. The improved CFD method (shown
in Figure 3) was validated against this data.
Figure 3 Example of improvements to the turbulence model
|3. Verification and improvement of accuracy through comparison
with measured data
3.1 Comparison with rig test data
Figure 4 compares the compressor characteristic obtained with a subsonic four-stage
compressor test rig and the CFD result. Figure 5 compares the total pressure loss coefficient
distribution from a total pressure measurement probe inserted into the flow path with the CFD
result. The improvement of the CFD method increased the prediction accuracy of the characteristic
line as well as the total pressure loss coefficient distribution in the span-wise direction.
Figure 4 Pressure ratio vs efficiency for a four-stage subsonic compressor
(Measurement vs. CFD)
Mitsubishi Heavy Industries Technical Review Vol. 52 No. 1 (March 2015)
4
Figure 5 Total pressure loss coefficient for the 4th rotor (measurement vs. CFD)
3.2 Comparison with gas turbine verification facility (T-point) measurement data
Figure 6 shows a picture of the gas turbine combined cycle power plant verification facility
(T-point), which is a part of Mitsubishi Hitachi Power Systems, Ltd.’s Takasago Works. T-point is
a combined cycle power plant verification facility consisting of a gas turbine, a steam turbine, and a
heat recovery steam generator (HRSG). It is used to acquire various component data through
special measurements taken during trial operation of prototype machines.
Figure 6 Combined cycle power plant verification
facility in Takasago (T-Point)
Figures 7 and 8 show the full-stage CFD results for the actual compressor. Comparing these
to the T-point data, the CFD prediction accuracy for the casing surface pressure distribution and the
total pressure distribution in the flow path can be validated.
Figure 7 Total pressure distribution for the middle
stages of the actual compressor (measurement vs.
CFD)
Mitsubishi Heavy Industries Technical Review Vol. 52 No. 1 (March 2015)
5
Figure 8 Casing pressure distribution for the full-stage compressor before
each blade and OGV exit (measurement vs. CFD)
|4. Application to power generation gas turbine compressor design
4.1 Compressor aerodynamic performance improvements through detailed
understanding of internal flow phenomena
By making use of CFD results, for which the prediction accuracy has been validated against
measurement data acquired through rig testing as well as verification testing on the actual machine,
MHI is working on improving the understanding of internal flow phenomena as well as improving
aerodynamic performance. For example, when the rotor tip clearance is relatively large, such as in
the middle and rear stages, the loss increases near the rotor tip because of the effect of the leakage
jet generated by the flow passing through the tip clearance as shown in Figure 9. Therefore, in
order to increase performance it is important to control the leakage flow from the tip. MHI is trying
to reduce the amount of time required to improve the aerodynamic performance in the actual
compressor by quickly evaluating proposed blade geometry improvements using rig testing and
selecting the best designs before moving to verification tests on the actual machine.
Figure 9 Detailed flow in the rotor tip for the rear stages
Mitsubishi Heavy Industries Technical Review Vol. 52 No. 1 (March 2015)
6
4.2 Prediction of compressor pressure fluctuation during gas turbine startup
Figure 10 shows an example of unsteady CFD of a compressor during gas turbine startup.
When the gas turbine is started, the flow rate is low which in turn causes rotating stall to occur in
the front stages. This is a very unstable flow field; however, advancements in CFD technology now
allows for phenomena such as this to be predicted. The predicted value of the pressure fluctuation
agrees favorably with the measured value at the casing surface. In this analysis, there were four
stall cells, which also agreed with the actual measurement results.
Figure 10 Prediction of pressure fluctuation during start-up of compressor
4.3 Evaluation of the effect of compressor bleed on the main flow
Compressors have bleed pipes, a bleed chamber, and a bleed slot to ensure stability during
startup and to supply the turbine with cooling air. Since a number of pipes, say four, are placed at
discreet circumferential locations, a non-uniform flow develops around the annulus. The effect this
non-uniformity has on the stalling behavior has been evaluated. Figure 11 shows the calculation
model and CFD results for changes in (1) the number of the bleed pipes, (2) the size of the bleed
chamber, and (3) the bleed slot geometry.
Figure 11 Evaluation of the effect of bleed geometry on the main flow
Mitsubishi Heavy Industries Technical Review Vol. 52 No. 1 (March 2015)
7
4.4 Evaluation of the effect of the inlet duct on the compressor front stages
A single suction duct is adopted for the intake of the compressor. The effect of the curved
section of the duct on the upstream flow and the effect on the mounting angle of the inlet guide
vane cause non-uniform flow to form in the circumferential direction. The effect of the
non-uniform flow on compressor aerodynamic performance was evaluated (Figure 12).
Figure 12 Evaluation of the effect of the inlet duct shape on the compressor front stages
|5. Conclusion
This paper presented improvements in our CFD methods and some applications of our new
ultra-high-speed parallel computing GPU technology to detailed calculation models designed to
help improve the performance of power generation gas turbine compressors. The role of CFD
analysis is expanding as computation technology advances and it is expected that high fidelity
analysis will be applied in even more aspects of compressor design in the future. However, to
facilitate this movement toward CFD analysis, reliable and detailed experimental data is necessary
to verify the accuracy and understand the phenomena seen in the CFD results. Although not
mentioned in this paper, MHI has also worked on developing high precision measuring technology
and advanced experimental equipment in order to accurately acquire the necessary data. With a
precise, detailed understanding of the physical phenomena through a combination of these
technologies, as well as efforts put into shortening the new proposal, experimental verification, and
product development cycle, MHI endeavors to deliver superior products to the market faster.
Reference
1. Hada, S. et al., Test Results of the World's First1,600C J-series Gas Turbine, Mitsubishi Heavy Industries
Technical Review Vol. 49 No. 1 (2012) pp.19～24
2. Tobias Brandvik and Graham Pullan, An Accelerated 3D Navier-Stokes Solver for Flows in
Turbomachines, ASME paper, GT2009-60052
3. Ito, E., CFD and Optimization of Gas Turbine Components including Inlet and Exhaust Ducts, Journal
of the Gas Turbine Society of Japan Vol. 40 No. 6 (2012. 11)
