Aerospace and Electronic Systems Magazine July 2018 - 8

EGT Prediction

Figure 3.

Experimental micro gas turbine engine and its test platform.

Olympus HP gas turbine engine and its test platform are shown
in Figure 3. Four tests were done on the engine and the average
values of the obtained data are shown in Figure 4.

RESULTS AND DISCUSSION
Figure 4 shows variations of EGT and engine rotational speed versus time, because the data gathering system measures all performance parameters of the engine according to the time.
Figure 4 displays the entire engine test process from start
to stop. The ECU will engage starter motor up to 3,000 RPM,
starts fuel pump and opens the starter solenoid valve to start the
engine. The main fuel valve opens as soon as EGT rises to 5°
C and the electric starter motor reaches maximum RPM. ECU
will automatically throttle up the turbine to its calibration point

at approximately 55,000 RPM and keeps the turbine at approximately 50,000 RPM for about two seconds, then it automatically
throttles back down to ideal RPM (i.e., 36,000 RPM) [23]. At
this point, the engine is fully calibrated and is ready for testing
and operator can change the RPM of the engine by throttle setting and perform desired experiments. The ECU regulates the
engine to 30% throttle (approximately 40,000 RPM) for about
four seconds and waits until exhaust temperature stabilizes for
engine shut down process. The engine still has about 2 kg of
thrust in 40,000 RPM when ECU stops the engine. Then, the
ECU automatically switches electric-starter on and off several
times to reduce EGT below 88° C [23]. The ECU is now ready
for a new start up.
Based on Figure 4, it can be concluded that by increasing engine rotational speed, EGT increases and vice versa. In fact, increasing engine rotational speed sucks more air into the engine
combustion chamber, and thereupon more fuel enters the combustion chamber, making the combustion process longer and more
complete, thus EGT increases.
In this study, the relation between EGT, as the output quantity
of engine and engine RPM, and as the input quantity, was investigated using two different data mining techniques. For this purpose, ANN (i.e., Multilayer Perceptron and Radial Basis Function
neural networks) and MPR approaches were employed to estimate
the nonlinear relation between input and output of the engine and
to predict the outputs corresponding to randomly selected sets of
input data. In other words, almost 20% of experimental data was
selected randomly from the whole experimental data set and used
to evaluate capabilities of these two approaches in predicting EGT.
Correspondence between predicted and measured values of EGT
was quantified by calculating the root mean squares of errors between predicted and measured values [24]. Based on the calculated root mean squares of errors, predicting capability of these
two data mining approaches was compared [24]. ANN and MPR
implementing procedure is discussed in more details in following
sections.

Figure 4.

Variation of EGT and RPM versus time.

IEEE A&E SYSTEMS MAGAZINE

JULY 2018

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine July 2018