You are on page 1of 12

GE Global Research

______________________________________________________________

2D/3D CFD Design Optimization


Using the Federated Intelligent Product
Environment (FIPER) Technology

R. Sampath, R.M. Kolonay and C. Kuhne

2002GRC202, August 2002


Class 1

Technical Information Series


Copyright © 2002 General Electric Company. All rights reserved.
GE Global Research

Technical Report Abstract Page

Title 2D/3D CFD Design Optimization Using the Federated Intelligent Product Environment
(FIPER) Technology

Author(s) R. Sampath Phone (518)387-6401


R.M. Kolonay 8*833-6401
C. Kuhne*

Component Advanced Mechanical Technologies, Niskayuna

Report
Number 2002GRC202 Date August 2002

Number
of Pages 8 Class 1

Key Words MDA/MDO, FIPER technology, Turbine Aerodynamics, Response-Surface


Methodology

This paper describes the application and implementation of an aircraft engine turbine blade
aerodynamic design optimization problem using a response-surface based methodology within
the Federated Intelligent Product EnviRonment (FIPER) framework. The design problem
considered is a shape optimization problem, in which, important blade shape parameters such as
the stagger angle, trailingedge thickness, leading-edge radius, over-turning and wedge angle are
chosen such that an optimal design is achieved. In this work, the criterion for optimality is taken
as the blade-row efficiency. The approach used in this work to solve this optimization problem is
to start with a baseline design, perturb the variables within a given range of validity, conduct a
design-ofexperiments (DOE) using an automated, distributed analysis DOE tool developed and
generate a response-surface. The resulting response surface is used along with a gradient-based
optimizer to calculate the optimal solution. The core of this paper describes in detail the FIPER
architecture and addresses specific implementation issues in the context of solving the above
shape optimization problem.

Manuscript received Augusst 2, 2002

*GE Aircraft Engines


AIAA-2002-5479

2D/3D CFD DESIGN OPTIMIZATION USING THE FEDERATED INTELLIGENT

PRODUCT ENVIRONMENT (FIPER) TECHNOLOGY

Rajiv Sampath, Raymond M. Kolonay+


GE Global Research Center
Schenectady, NY 12301
Craig M. Kuhne
GE Aircraft Engines
Cincinnati, OH 45215
ABSTRACT competitive market. The need of the day is a truly
digitized design process that can provide true
This paper describes the application and
concurrency between design and manufacturing. In
implementation of an aircraft engine turbine blade
order to address these challenges, GE has teamed up
aerodynamic design optimization problem using a with Goodrich, Parker Hannifin, Engineous Software,
response-surface based methodology within the Ohio University, Stanford University and OAI to
Federated Intelligent Product EnviRonment (FIPER) develop a concurrent engineering design system as a
framework. The design problem considered is a shape part of a four-year advanced technology program
optimization problem, in which, important blade supported by the National Institute of Standards and
shape parameters such as the stagger angle, trailing- Technology (NIST). This system referred to as the
edge thickness, leading-edge radius, over-turning and “Federated Intelligent Product EnviRonment”
wedge angle are chosen such that an optimal design is (FIPER) strives to drastically reduce design cycle
achieved. In this work, the criterion for optimality is time and time-to-market by intelligently automating
taken as the blade-row efficiency. The approach used elements of the design process in a linked, associative
in this work to solve this optimization problem is to environment, thereby providing true concurrency in
start with a baseline design, perturb the variables the design process. Such an approach allows for
within a given range of validity, conduct a design-of- distributed design of robust and optimized products
experiments (DOE) using an automated, distributed by employing an open, network-centric, web-based
analysis DOE tool developed and generate a environment1 (Fig. 1).
response-surface. The resulting response surface is
used along with a gradient-based optimizer to As an application of the FIPER system, here, a novel
implementation of a turbomachinery blade design
calculate the optimal solution. The core of this paper
system is discussed. The design problem being
describes in detail the FIPER architecture and
considered is a shape-optimization problem in which
addresses specific implementation issues in the
the task is to optimize the blade cross-section by
context of solving the above shape optimization
varying shape parameters such as the stagger angle,
problem.
trailing-edge thickness, leading-edge radius, over-
INTRODUCTION turning and wedge angle in order to maximize the
blade-row efficiency. In the main part of this paper,
Gas turbine engine development is a highly coupled
this design problem is discussed along with the
multidisciplinary process involving several
techniques used to solve this problem within the
disciplines such as aerodynamics, solid mechanics
FIPER framework.
and heat transfer. In a market with ever-increasing
demands in terms of reduced design cycle time, BACKGROUND
reduced life cycle costs and performance
FIPER draws extensively on several prior GE
improvements, turbine engineers can no longer
initiatives, primarily the Common Geometry Initiative
depend on the manual serial design process to
and the System Oriented Layout and Integrated
produce efficient or cost-effective designs in today’s
Design (SOLID) project. Concepts developed in these

+Senior Member AIAA


Copyright © 2002 General Electric Company.
Published by American Institute of Aeronautics and
Astronautics, Inc., with permission.

1
American Institute of Aeronautics and Astronautics
AIAA-2002-5479

Figure 1: The FIPER Project


projects like the top-down Product Control Structure the design and manufacturing process, so that at all
(PCS), Fig. 2, and the Linked Model Environment times participants in the design and manufacturing
(LME), Fig. 3, form the backbone of the geometry process have access to the same valid geometry
aspects of FIPER. Since these projects have been information. Taking advantage of the
described in previous publications2,3, only key UNIGRAPHICS® WAVE top-down geometry
linking capability, a PCS for turbine engines was
developed, specifying interfaces between the different
engine subsystems and giving control of these
interfaces to the groups responsible for system-level
design. Each subsystem group owns its own interfaces
between individual components.
Understanding that different disciplinary engineering
design and analysis tools require geometry at
different levels of detail, the concept of a "context
model" was introduced. The context model represents
a disciplinary context-specific, yet fully associative,
"view" of the master model geometry. Feature
suppression is extensively used in context models.
For example, a bolt hole, which is important for the
stress analyst, may not be required for a thermal or
CFD analysis and therefore would be suppressed in
the thermal/fluids context model. These context
models are then linked to the respective disciplinary
analysis tools, e.g. FEA, CFD, cost, producibility,
etc, in the LME , see Figure 3.
FIPER extends the concept of the geometric Master
Model with the introduction of the “Intelligent Master
Model” (IMM). Intelligence is added through
extensive use of Knowledge-Based Engineering
(KBE) systems in the design process. Initially, rules
for geometry generation were captured separately
from the geometry itself. Lately, with the addition of
the Knowledge Fusion product into the Unigraphics
Figure 2: Product Control Structure CAD system (UG/Knowledge Fusion), this has
become the tool of choice at GE Aircraft Engines.
(UG/Knowledge Fusion is derived from Engineering
concepts are addressed here as necessary to
Intent’s (formerly Heide Corp.) Intent product.)
understand this work. The primary objective of the
Common Geometry initiative was the ability to use a
common (evolving) geometry master model through

2
American Institute of Aeronautics and Astronautics
AIAA-2002-5479

FIPER is composed of various service providers; any


of these can come and go and the system can respond
to changes in its environment in a reliable way
(network centricity). The services connected to
FIPER discover each other and cooperate in a
distributed environment (service centricity). Users
can request to use multiple services and check the
status of their submissions in different locations
through HTTP portal with thin web clients (web
centricity).
The three neutralities FIPER deploys are location
neutrality, protocol neutrality, and implementation
neutrality, Figure 4. Services need not be co-located;
they are discovered and then join the federation,
Figure 3: Linked Model Environment
which simplifies management of the entire software
environment (location neutrality). In addition, the
FIPER ARCHITECTURE way clients communicate with a service provider is
Traditionally MDA and MDO efforts have led to not essential. A service proxy can use any protocol,
systems that are monolithic, hard-wired systems, that for example, Remote Method Invocation (RMI), IIOP
tend to provide only limited capabilities for or even a plain socket communication. Clients are not
distributed computing and collaboration of various aware of what protocols are used and where the
designers located at spatially disjoint facilities5. implementations reside (protocol neutrality).
Moreover, they rarely support a plug-and-play type Furthermore, the clients who use the FIPER services
architecture where one analysis or simulation tool can do not need to know what languages are used and
easily be substituted by another essentially fulfilling how a service is implemented (implementation
the same function, with either a faster newer method, neutrality). In all, FIPER provides accessibility
or potentially at a higher level of fidelity. Today, the through a web centric architecture, self-manageability
web offers the vehicle for efficient data transfer using federated services, scalability via network
across the globe, but its capabilities are so far only centricity, and adaptability with the power of plug-
beginning to be utilized by the engineering and design and-play capability.
community. This is the case even though much of
engineering development in a project these days
occurs at spatially different locations. FIPER FIPER
addresses this deficiency by providing a network- Federation of Services
centric infrastructure supporting distributed Proxy
(location
engineering services in a Peer-to-Peer paradigm. neutrality)
FIPER federates processes, tools, methods, discover register
documents, knowledge bases, and data into a and join and publish
dynamic, distributed environment with its underlying
services. Some services are generic (for example,
optimization algorithms or knowledge-based
protocol Service
systems), and thus, are not associated with a Client Proxy Proxy Provider
particular IMM context but are globally available (protocol
within FIPER. Members of a federation agree on neutrality) (implementation
basic notions of administration, identification, and neutrality)
policy. The resulting federation provides the
Figure 4: FIPER's Three Neutralities
simplicity of access, ease of administration and
support for sharing services provided by a large FIPER's federated architecture is based on Java and
monolithic system, while retaining the flexibility, and Sun's emerging JiniTM software system (Figure 5).
control provided by a plug-and-play environment. The overall goal is to turn the network into a flexible,
FIPER supports three centricities and deploys three easily administered tool on which resources can be
neutralities. FIPER's three centricities are network found by humans or computational clients. The JiniTM
centricity, service centricity, and web centricity. system consists of:

3
American Institute of Aeronautics and Astronautics
AIAA-2002-5479

A set of components that provides the FIPER. For the purpose of the NIST project FIPER
infrastructure for federating services in a focuses on the services necessary for the design and
distributed environment manufacture of a product. Specifically, the domains
of Design for Six Sigma/ Multidisciplinary
A programming model that supports the
Optimization (DFSS/MDO), CAD/KBE, Engineering
production of reliable distributed
Analysis & Sensitivities, Pre/Post processing, and
environment
Data Repositories are addressed. Services are
The functionality to register services and provided by a service provider (which is typically a
resolve service requests computer program, for example, an engineering
analysis package, that has been wrapped to operate
Java and the emerging JiniTM technology are at the
within the FIPER environment). A service provider
heart of this system. Services are found and resolved
can offer multiple services, and the same service can
through a "lookup" service (Figure 5). New services
be offered by multiple service providers. The FIPER
are added to the look-up service by a process called
task submitted by the client is matched against the
"discover and join". When plugged into the
services offered by the various service providers, and
environment, the service first uses a discovery
the proper service is selected based on service
protocol to locate an appropriate lookup service and
attributes.
then joins, or registers, with the lookup service.
Services can communicate with any other generic Several engineering service providers are currently
service in the entire federated product space. In the implemented in the FIPER environment such as
case of FIPER, this is achieved by an IMM context, UNIGRAPHICS® for a variety of CAD and geometry
user, or service posting a need which is resolved by a services, PATRAN® and ICEM® primarily for finite
lookup service. The lookup service connects the element and CFD meshing services, ANSYS® for
requesting entity to an entity that has the functionality finite element meshing, boundary condition, analysis,
to supply the service. Figure 5 illustrates this in a and post processing services. Several in-house GE
given space with four services; CAD, KBE, proprietary codes are used as service providers for
Optimization and Robust Design, and the Simulation addressing the deterministic as well as probabilistic
Engine. Each service provider must be Java wrapped design of turbomachinery components.
in order to join the federation, but it can have its own
Members of a federation agree on the basic notions of
framework of execution. A service could be based on
administration, identification, and policy. The
RMITM, CORBATM, Java Native Interface (JNITM),
resulting federation provides the simplicity of access,
Microsoft COMTM/DCOMTM, or even simple socket
ease of administration and support for sharing
connection.
services provided by a large monolithic system, while
retaining the flexibility, and control provided by a
plug-and-play architecture.
Clients define and submit their jobs via web
browsers. A FIPER service manager then dispatches
each job into tasks. These tasks (or exertions) can be
executed sequentially, in parallel, or combination of
both in the FIPER environment, depending on their
input/output data dependency. If a parallel strategy is
chosen, tasks are dropped into spaces (by using
JavaSpacesTM, for example) for distributed
computation. Each service provider agent, if present,
picks up appropriate tasks and generates results and
returns them back to the space. On the other hand,
Figure 5: Web-Based FIPER Architecture FIPER provides a service catalog for direct task
execution. The catalog discovers all FIPER services
The basic premise of FIPER is that everything is on
and maintains a list of currently active ones.
the network and everything on the network is viewed
Appropriate registered service providers will then be
as a service. With this in mind, FIPER can contain
selected to perform tasks. Finally, a service manager
any "service" needed to support a product throughout
collects all the outputs and informs the FIPER
its life cycle. For example, services for customer
notification manager about the outcome. The results
requirements, design, manufacture, sales, distribution,
are presented to the clients when they request. FIPER
maintenance, and disposal can all be supported by

4
American Institute of Aeronautics and Astronautics
AIAA-2002-5479

also allows for real-time monitoring and editing of blade is considered (and this implies that 3D losses
jobs submitted through the system. This allows the are accounted for in the aerodynamic analysis). The
user the flexibility to modify a submitted job during process map used in the 2D/3D CFD based design of
any stage of its execution. This is particularly the blade is illustrated in Fig. 7. As described, the
important in multidisciplinary optimization problems, first step in the design process is to start with a scaled
wherein, there is greater probability for user errors or airfoil design. This design data is copied to the
service failures. If such a situation arises, the FIPER optimization workspace to be modified by the design
job monitor would be very effective in process. This is followed by an update process in
reviewing/editing the failed task and resuming the which important blade shape parameters are modified
execution of the optimization problem. to enforce blade shape change. This is followed by
the execution of a blade generation/stacking program
TURBINE BLADE AERODYNAMIC DESIGN
in order to generate the appropriate blade shape.
The design of a new aircraft engine from scratch is an Once the blade geometry is defined, the next step is
extremely rare occurrence in today’s competitive the generation of a 2D/3D grid. The 2D/3D grid is
environment. “New” designs are most often derived used in a corresponding CFD analysis to evaluate the
from existing engines. Derivative engine design is performance of the blade design. This entire loop is
accomplished by first scaling from an existing design, executed for each DOE run. The outputs from this
then optimizing this derivative geometry on the basis DOE are used to generate a response surface to
of a list of design criteria. Modifications from this optimize the design. In this work, due to the non-
scaled geometry are made as required to meet the linear nature of the transfer function, a 2nd order
design detailed criteria, such as component life and central-composite design6 augmented with a one-
aerodynamic performance. factor-at-a-time (OFAT) DOE design close to the
optimal point is employed. The upper bounds, lower
As an application of the FIPER system, a turbine
bounds, variable limits are chosen in order to ensure
blade aerodynamic design optimization problem is
an approximate transfer function that was within the
presented. Blade shape parameters (see Fig 6.) such
required accuracy bounds.
as the stagger angle, trailing-edge thickness, unguided
turning, over-turning and wedge angle are chosen Figure 8 shows the internals of the 2D/3D CFD
such that the blade-row efficiency is maximized. design tool developed within the FIPER system. As
Blade row efficiency defines the overall aerodynamic described in this figure, five major service providers
performance of the turbine blade-row and is an were developed to automate the 2D/3D CFD based
indicator of the performance of the engine as a whole. optimization process map shown in Fig. 7. A database
manager service to manage data transfer between the
aero database containing the scaled airfoil data and
the FIPER system. A morphing service to generate
the morphed or mapped geometry for each DOE run.
This service updates the airfoil shape parameters
(such as stagger, wedge angle, etc), regenerates the
airfoil design sections and then stacks the blade
sections to generate the 3D blade. A gridding or
meshing service provider is developed to generate the
2D/3D blade-to-blade grid. GE in-house application
is the service provider for the 3D gridding. Another
GE proprietary code is the service provider for the
2D gridding. A blade-to-blade CFD analysis service
provider is developed to execute the 2D/3D blade-to-
blade CFD analysis. Here, an internally developed
GE code is used for the CFD analysis. The code used
Figure 6: Turbine blade geometry parameters is a 3D structured grid Euler/Navier-Stokes solver for
turbomachinery blade rows. Finally, a post-
In this paper, the results of this optimization study for processing service is developed to post-process the
both 2D as well as 3D cross-section designs are CFD results to calculate the design objectives and
presented. In the 2D mode, only the cross-section at constraints.
one streamline section is cut and designed for optimal
performance, while in the 3D mode the entire 3D

5
American Institute of Aeronautics and Astronautics
AIAA-2002-5479

Copy baseline design to


optimization workspace

Update airfoil shape


parameters

Generate
DOE Table Morph the airfoil
sections

Stack the airfoil sections Figure 8: Integration of 2D/3D CFD Design


to generate 3D airfoil Process Using FIPER Architecture
All services are published at a remote HP
workstation. Emerging JavaSpacesTM technology is
Generate 2D/3D grid used to develop a distributed DOE execution
environment. In particular, a FIPER space is created
in which the various DOE runs are dropped as tasks.
Depending on the availability of the various service
Execute 2D/3D CFD providers (multiple instances of these service
Analysis providers were published across the network), the
tasks are picked up by the various providers and are
executed and the results attached to the context model
Calculate optimization and sent back to the user for review. After execution,
objective function each service sends back a detailed report which can
be perused by the experienced engineer to make
appropriate design decisions. In order to speed-up the
convergence of the CFD analysis the flow results
Evaluate design from the baseline case were used as the initial starting
constraints flow solution for the various DOE runs. Fig. 9 shows
the initial and optimal blade profiles calculated using
this system. This case corresponds to the 1st rotor in a
2-stage high-pressure turbine. (Note: The turbine
blade example considered herein is strictly a
Generate Response caricature of a real turbine blade problem; no
Optimizer Surface information contained in this paper is representative
of turbine blades produced by GE. Also, due to
restrictions dealing with sharing proprietary
information, only normalized quantities will be
presented in this paper). All the angle blade shape
Figure 7: 2D/3D CFD Optimization Process Map
parameters (stagger, wedge, overturning) were varied
from the baseline values by +/- 5 degrees while the
trailing edge thickness and leading edge radius were
varied up to a maximum of 10% from their nominal
values. A total of 59 DOE runs followed by an
additional 18 validation runs were required to
generate an appropriate response surface for this
design problem. Fig 10. compares the results for a
similar optimization study on the 2D pitchline section
of the same rotor blade. Using the CFD solution of

6
American Institute of Aeronautics and Astronautics
AIAA-2002-5479

the nominal design as the starting flow solution (this network. As an extension of this work, currently CFD
technique was employed to accelerate the solution of multiple blade rows is being considered.
convergence of the CFD solution), the 3D-CFD This will allow the designers to consider the entire
design took close to 45 minutes per DOE iteration stage performance as the objective function rather
while the 2D-CFD case took about 10-15 minutes per than limit themselves to blade-row efficiency
DOE iteration. All the DOE runs were executed in estimates.
parallel on 5 HP servers distributed across the

Normalized Stagger = 1.0


Normalized leading edge radius = 1.0
Normalized trailing edge radius = 1.0
Normalized over turning = 1.0
Normalized wedge angle = 1.0
Normalized blade row efficiency = 1.0

Normalized Stagger = 1.1208


Normalized leading edge radius = 1.009
Normalized trailing edge radius = 0.971
Normalized over turning = -0.4548
Normalized wedge angle = 0.8115
Normalized blade row efficiency = 1.0172

Figure 9: Initial and Optimal 3D Blade Designs

7
American Institute of Aeronautics and Astronautics
AIAA-2002-5479

JavaSpacesTM technology. Email notification after the


execution of jobs provides URL links to the
significant output which the user can study through a
Normalized stagger = 1.0
web browser, from anywhere in the world.
Normalized leading edge radius = 1.0
Normalized trailing edge radius = 1.0 Currently, the FIPER team is expanding the capability
Normalized over turning = 1.0 to do system-level optimization studies wherein
Normalized wedge angle = 1.0
Normalized blade row efficiency = 1.0 disciplines such as aerodynamics, heat transfer and
mechanical analysis are all coupled. The results from
these efforts will be presented in subsequent
presentations.
Acknowledgments
This research is jointly funded through the National
Institute for Standards and Technology-Advanced
Normalized stagger = 1.140 Technology Program (NIST-ATPTM) and the General
Normalized leading edge radius = 1.049 Electric Company. The authors would like to
Normalized trailing edge radius = 1.1003
Normalized over turning = -0.0369 acknowledge this support, as well as the valuable
Normalized wedge angle = 1.0665 input from the whole FIPER team. In particular, the
Normalized blade row efficiency = 1.021
authors would like to acknowledge the help and
valuable input from Shashi Talya, Anurag Gupta,
Sanjay Goel, Rohinton Irani and Michael Sobolewski
in developing the CFD-based optimization tool within
the FIPER system.
Figure 10: Initial and Optimal 2D Blade Designs References
CLOSING REMARKS [1] Federated Intelligent Product Environment,
Technical Proposal, OAI, General Electric,
A robust network-centric MDA/MDO environment Goodrich, Parker Hannifin, Engineous Software,
has been developed and applied to an engineering Ohio University, Stanford University, April,
design problem. The concepts of an engineering 1999.
analysis code as a service provider and of a “FIPER [2] Röhl P.J.; Kolonay R. M. et al.: A Federated
context” as a generic means of supplying problem- Intelligent Product Environment. Proceedings,
specific information to the generic service provider 8th AIAA/USAF/NASA/ISSMO Symposium on
have been implemented. The FIPER framework has Multidisciplinary Analysis and Optimization,
been applied to a number of sophisticated engineering Long Beach, CA, September 2000.
and design problems, one of which, the aerodynamic [3] Rohl, P. J.; Kolonay R. M. et al.: Intelligent
analysis and CFD based design of a turbine blade, Compressor Design in a Network-Centric
forms the basis of this work. The framework takes Environment. Proceedings, ASME Computers in
advantage of a number of key technologies for Engineering Conference, Pittsburgh, PA,
distributed network-centric computing, primarily September 2001.
JINITM, RMITM, and JavaSpacesTM. [4] Intent User Manual, Heide Corporation,
The user has the ability to access the system via a Medfield, MA, 2000.
web browser from anywhere in the world to submit [5] Rohl, P. J., A Multilevel Decomposition
jobs, check the status of jobs, or to review results. Procedure for the Preliminary Wing Design of a
The system also provides the user the capability to High-Speed Civil Transport Aircraft, Doctoral
monitor jobs in real-time – giving essentially Thesis, Georgia Institute of Technology, Atlanta,
unlimited flexibility in editing and reviewing GA, 1995.
submitted jobs. This is particularly attractive in the [6] Raymond H. Myers and Douglas C. Montgomery,
context of solving large-scale multidisciplinary Response Surface Methodology: Process and
optimization problems, wherein, the probability of Product Optimization Using Designed
tasks failing is significant. The system also provides Experiments, John Wiley & Sons, 1995.
for distributed computing using the emerging

8
American Institute of Aeronautics and Astronautics
R. Sampath 2D/3D CFD Design Optimization Using the Federated Intelligent 2002GRC202
R.M. Kolonay Product Environment (FIPER) Technology August 2002
C. Kuhne

You might also like