1
Copyright© 2003 by Roth et al. Published by the American
Institute of Aeronautics and Astronautics Inc., with permission.
Estimation of Turbofan Engine Performance Model Accuracy and Confidence Bounds
Bryce A. Roth, Dimitri N. Mavris
School of Aerospace Engineering
Georgia Institute of Technology
Atlanta, GA 30332-0150
and
David L. Doel
GE Aircraft Engines
Cincinnati, OH 45215
Abstract
This paper explores the application of Inference
and Bayesian Updating principles as a means to
efficiently incorporate probabilistic data into the turbine
engine status model matching process. This approach
allows efficient estimation of nominal model match
parameters from test data and also enables
quantification of model accuracy and confidence
bounds. The basic concepts are developed in detail and
formulated into a status matching approach. This
method is then applied to a simple surrogate matching
problem using a cantilever beam matching exercise to
illustrate the methods in a clear and easy-to-understand
way. Typical results are presented and are directly
analogous to status matching of a gas turbine engine
cycle model.
Introduction
This paper focuses on developing improved
methods for matching high-accuracy engine
performance models to test data. The chief motivation
for this work is the need to make model predictions
match test data as closely as possible and to do so in the
minimum possible time and cost. The basic engine
performance matching problem is complicated by the
fact that the test data has inherent uncertainty which
must be accounted for in the matching process.
Furthermore, the data typically comes from many
sources, each with a varying degree of relevance
(legacy test data versus new test data, for instance).
Appropriate weight must be apportioned to each piece
of data according to its relevance to the current engine.
Finally, some parameters must be matched with greater
precision than others. Challenges are also presented by
continued increases in engine complexity (particularly
with regard to control laws), and the need to model an
ever-increasing number of phenomena that impact
engine performance. The confluence of these factors is
a clear and present need to develop rigorous, efficient,
and methodical approaches to matching model
predictions to test data.
While it is important that the data match be as
accurate as possible, it is equally important to be able to
quantify the accuracy of the match. Engine
performance models are used throughout the design
process and impact significant decisions regarding the
design and specific commitments to customers. A
failure to appreciate the accuracy of the performance
model can cause customer commitments to be
unattainable or, conversely, to be overly conservative.
Similarly, the same issues can lead to overly
conservative designs or to field problems resulting from
optimistic assumptions about the conditions an engine
component will experience in service. Thus, a second
ingredient of a useful engine performance model is the
ability to specify its uncertainty.
A Typical Status Matching Approach
A general feature of status matching is that the
problem becomes more complex and the flexibility
greater as more data is available to be considered. The
easiest case is a production status match where there are
many engines, but few parameters to be matched.
Another thing that makes the production status match
simpler is that all the data is at sea level static
conditions.
The goal of a production status match is to
simultaneously match thrust at fan speed, core speed at
fan speed, specific fuel consumption (SFC) at thrust
and exhaust gas temperature (EGT) at fan speed.
Although internal pressures and temperatures are
available, they are normally single element probes
(control sensors) and are not considered in the matching
process.
Typically, the first step is to match thrust at fan
speed. This is done by varying the fan pumping (air
flow at fan speed) until a thrust match is achieved. In
this, or any subsequent step, a reasonableness check is
performed to make certain that the resulting map makes
sense. This might involve an observation that very little
change from the previous representation was required
to achieve the match or it could involve a review with
the appropriate aero designer or other expert(s).
Step two is to match core speed at fan speed. This
will usually be done by adjusting the high pressure
compressor flow at speed characteristic. If either high
pressure or low pressure turbine flow functions have
changed (because of a deliberate design decision), they
ISABE 2003-1208