Journal of Physics: Conference Series

PAPER • OPEN ACCESS You may also like

- Research on Flight Operational Quality
Fitting Operation Curve of Civil Aviation Turbo-fan Assurance Based on ROM Concept
Haofeng Wang, Kai Qi and Yin Tang
Engine’s Variable Bleed Valve based on MATLAB - Carbon dioxide emissions from
international air transport of people and
freight: New Zealand as a case study
To cite this article: Yan Tan 2019 J. Phys.: Conf. Ser. 1176 052040 Anna P Tarr, Inga J Smith and Craig J
Rodger

- Design and Implementation of

International Civil Aviation Security
Information Database Management
View the article online for updates and enhancements. System
Hai Li, Xiaoyong Yang and Sihai Feng

This content was downloaded from IP address 197.244.160.246 on 27/01/2023 at 13:23

SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

Fitting Operation Curve of Civil Aviation Turbo-fan Engine’s

Variable Bleed Valve based on MATLAB

Yan Tan*
Aero Engine Maintenance Training Center, Civil Aviation Flight University of China,
Sichuan, China
*Corresponding author e-mail: [email protected]
Abstract. VBV (variable bleed valve) is an important mechanism used by civil
aviation turbo-fan engine to control compressor airflow, thus to prevent engine surge.
However, since most of civil aviation turbo-fan engine are produced by a handful of
manufacturers, the component characteristics and control principles of engine valves
are not known. As a result, condition monitoring and fault diagnosis of VBV cannot
be done by airlines themselves without the help of manufactures. Adopting MATLAB
simulation software, using the quick access recorder (QAR) data, by function mining,
this paper found a fitting function expression of the operation curve of VBV. The
results of validation shows that, this function expression has high relevance and small
error which can meet the needs of airlines for VBV condition monitoring and fault
diagnosis; using MATLAB software for operation curve fitting of valves is simple and
efficient, and it can be further applied in aero engine modeling.

1. Introduction
For an aero engine, the surge of the compressor will lead to strong mechanical vibration and over
temperature of hot section, which may cause serious damage to the engine in a very short time. Now,
most of civil aviation turbo-fan engines use variable bleed valves (VBV) to discharge some
compressor air to prevent surge. Therefore, VBV’s researches mean a lot for engine condition
monitoring, and then discovering adverse change in time to prevent flight accidents.
At present, most of civil aviation turbo-fan engine in China are produced by a handful of foreign
manufacturers, the component characteristics, structural parameters and control principles of engine
valves are not known, so it is impossible to establish the mathematical model for engine condition
monitoring and fault diagnosis. However, during engine operation, airlines have accumulated huge
QAR (Quick access recorder) data. These QAR data basically cover all performance parameters of
engine gas path, control, vibration and so on. So by mining these QAR data, that is, using the input
parameters and output parameters of system to recognize engine system, we can find a corresponding
functional relationship, establish a relatively real mathematical model to reflect system.
The tool used in this paper for data processing and function mining is MATLAB, which is a
commercial mathematics software developed by an American company Mathworks. MATLAB is an
advanced computing language and interactive environment for algorithm development, data
visualization, data analysis and numerical calculation. MATLAB mainly includes MATLAB and
Simulink. MATLAB has developed powerful module sets and tools for many specific fields. Usually,
they are developed by experts in specific fields. Users can use toolbox directly to learn, apply and
evaluate different methods without writing their own code. Data acquisition, database interface,
probability statistics, spline fitting, optimization algorithm, partial differential equation solution,
neural network, control system design, linear matrix inequality control, robust control, model

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

prediction, fuzzy logic and so on, they are all have a place in the toolbox.
2. Data preprocess
2.1 Data selection
For getting better operation curve of VBV, this paper selects a Boeing 737-800 aircraft with two
CFM56-7B26E engines as research objects. Both engines are installed on wing in May 2015, CSN
(cycles since new) is about 4200.
VBV operation curve refers to the functional relationship between VBV in standard state and its
control parameters when engine is in good operation state. According to the manual of CFM Company
which is the manufacturer of CFM56-7B engine, the qualitative operation curve of VBV is:
VBV=F (N1K12, N2K25, VSV, TAT, T2.5, TLA, N1ACT, P0)
Where: VBV, opening of VBV (degree); N1K12, corrected low pressure rotor speed (％); N2K25,
corrected high pressure rotor speed (％); VSV, opening of variable stator vane (degree); TAT, total air
temperature (Celsius); T2.5, high pressure compressor inlet temperature (Celsius); TLA, thrust lever
angle (degree); N1, low pressure rotor speed (％); P0, ambient static pressure (psia).
According to the simplified curve, this paper downloads and decodes this Boeing 737-800 aircraft’s
38 flights, total 76 groups QAR data. Depending on different flight time, each flight data varies from
7500 to 12000. Part of selected QAR data is shown in table 1.
Table 1 Part of selected QAR data
N11C N12C N21C N22C TATS TATS VBV VBV VSV VSV T251 T252
EL1 EL2 SEL1 SEL2 P1 P2 1 1
/% /% /% /% /℃ /℃ /degr /degr /degr /degr /℃ /℃
ee ee ee ee
0 0 0 0 -5.80 11
08
0 0 0 1 34.45
31
0 0 0 3.1 0 0 37.79 19.5
3
0 0 0 5.6 31.81
64
0 0 0 8.3 -90 -128
0 0 0 10.8 -90
0 0 0 13.1 0 0 -5.80 18.75
08
0 0 0 15.3 10.89
84
0 0 0 17.1 -5.80 11
08
0 0 0 19 34.45
31
0 0 0 20.5 0 0 38.14 17.75
45
0 0 0 21.9 34.27
74
0 1.4 0 23.1 0 0
0 1.8 0 24.3 0
0 2.1 0 25.3 0 0 38.32 17.25
03
0 2.5 0 26.1 33.75
0 2.9 0 26.8 -5.80 11
08

2
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

Where: N11C, corrected low pressure rotor speed of engine one; N12C, corrected low pressure
rotor speed of engine two; N21C, corrected high pressure rotor speed of engine one; N22C, corrected
high pressure rotor speed of engine two; TATSEL1, total air temperature of engine one; TATSEL2,
total air temperature of engine two; VBVSEL1, VBV opening of engine one; VBVSEL2, VBV
opening of engine two; VSVP1, VSV opening of engine one; VSVP2, VSV opening of engine two;
T2511, high pressure compressor inlet temperature of engine one; T2521, high pressure compressor
inlet temperature of engine two.

2.2 Data processing

For better mining of VBV operation curve, the following principles are carried out to filter data:
select data with good engine operating performance; eliminate data with significant bias; process
missing data by interpolation. After filter, data are processed as: separate each flight data to engine one
and engine two; transfer valves opening from degree to radians, 1 degree = 0.01745 radians. Table 2
and table 3 are the processed results of table 1.

Table 2 Processed data of engine one

N11C N21C TATSEL1 VBVSEL1 VSVP1 T2511

/% /% /℃ /radians /radians /℃
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11
0 0 0 0.60121 -0.10122 11

3
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

Table 3 Processed data of engine two

N11C N21C TATSEL1 VBVSEL1 VSVP1 T2511
/% /% /℃ / radians / radians /℃
0 0 0 0.5552 0.65949 19.5
0 1 0 0.5552 0.65949 19.5
0 3.1 0 0.5552 0.65949 19.5
0 5.6 0 0.5552 0.65949 19.5
0 8.3 0 0.5552 0.65949 19.5
0 10.8 0 0.5552 0.65949 19.5
0 13.1 0 0.5552 -0.10122 18.75
0 15.3 0 0.19018 -0.10122 18.75
0 17.1 0 0.19018 -0.10122 18.75
0 19 0 0.19018 -0.10122 18.75
0 20.5 0 0.19018 0.66562 17.75
0 21.9 0 0.59814 0.66562 17.75
1.4 23.1 0 0.59814 0.66562 17.75
1.8 24.3 0 0.59814 0.66562 17.75
2.1 25.3 0 0.59814 0.66869 17.75
2.5 26.1 0 0.58894 0.66869 17.75
2.9 26.8 0 0.58894 0.66869 17.75
3.1 27.4 0 0.58894 0.66869 17.75
3.5 28.4 0 0.58894 0.67176 15.5
4.3 30 0 0.5828 0.67176 15.5
2.3 Correlation analysis
Since each independent variable has different effects on variables, we need to identify the most
influential parameters for VBV opening as the basis of fitting. So, first of all, we need to analyze the
correlation coefficient and determination coefficient of independent variables to variable. After
analyzing the data of one engine during one flight which is randomly selected, the results are
following:
Table 4 Correlation coefficient and determination coefficient
N1 N2 TAT VSV T25
Correlation -0.8440 -0.2073 0.5836 0.9558 -0.0540
coefficient
Determination 0.7124 0.0430 0.3406 0.9136 0.0029
coefficient
The results show that, VBV opening is closely correlated to VSV opening, then the corrected N1
speed. Therefore, this paper will find the functional relationships between VSV opening, corrected N1
speed and VBV opening first.
3. Functional curve fitting based on MATLAB
3.1 Single parameter curve fitting
Take VSV opening as independent variable x, VBV opening as variable y, single parameter
functional curve is fitted by fit function of MATLAB. The following shows the fitting results of a
random selected flight and after 4:1 data compression.
Table 5 Results of single parameter linear function fitting
Fitting function y=-0.06489+1.214x
Residual sum of squares 10.1454
Degree of freedom 2539
Determination coefficient 0.9136
Mean square error 0.0632

4
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

This fitting uses linear function, but obviously VBV control is nonlinear. And the same time, from
the results of table 5, the determination coefficient is 0.9136, which has lower accuracy, so sine-cosine,
exponential function and so on are introduced to single parameter fitting, then the results are shown in
table 6.
Table 6 Results of single parameter various function fitting
Fitting function y=-283.5-504.4*x+123.7*cos(x)+342.4*sin(x)+159.9*exp(x)
Residual sum of squares 3.8416
Degree of freedom 2539
Determination coefficient 0.9672
Mean square error 0.0389
By the results of table 6, residual sum of squares decrease obviously, determination coefficient
increases to 0.9672, which means the accuracy increases a lot. By now, the deviation between fitting
functional curve and actual curve is shown as Fig. 1. In this Fig., the solid line is actual curve; the
point line is the fitting curve.

Fig.1 VBV single parameter fitting curve and actual curve

Fig.2 shows the difference between VBV single parameter fitting curve and actual curve

Fig.2 Difference between VBV single parameter fitting curve and actual curve
By Fig.1 and Fig.2, we can say, besides where the actual curve varies dramatically, the
approximation between fitting curve and actual curve is high, and the error of fitting curve is small.
3.2 Various parameters curve fitting
In order to further improve accuracy, after single parameter curve fitting, other parameters are

5
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

joined to various parameters curve fitting. This various parameters curve fitting is based on a
supposition which the influences of other independent variables to variable are linear combination. By
the various parameters fitting function regree () of MATLAB, the results of fitting are shown in table
7.
Table 7 Results of various parameters various function fitting
Fitting function y=-283.5-504.4*x+123.7*cos(x1)+342.4*sin(x1)+159.9*exp(x1)+0.0028
x2-0.0021 x3+0.0013 x4-0.0015 x5
Residual sum of 2.1389
squares
Degree of freedom 2539
Determination 0.9819
coefficient
Mean square error 0.029
In this fitting function, independent variables x1, x2, x3, x4, and x5 are VSV, N1, N2, TAT and T25.
In the results, the determination coefficient is 0.9819, mean square error is 0.029, which fully prove
the perfect fitting effect. Fig.3 shows the deviation between fitting curve and actual curve, the solid
line is actual curve; the point line is the fitting curve.

Fig.3 VBV various parameters fitting curve and actual curve

Fig.4 shows the difference between VBV various parameters fitting curve and actual curve

Fig.4 Difference between VBV various parameters fitting curve and actual curve

By fig.3 and fig.4, we can say, the error between the fitting curve fitted by various parameters VSV,

6
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

N1, N2, TAT, T25 and actual curve is very small, the fitting effect is good.
4. Verification of fitting curve
The previous fitting is based on a random fight, and after data compressed. So in order to verify
effect of fitting curve, this paper verifies with other 37 flights, and gets 74 results as table.8 shows.
Table 8 Verification results of 38 flights
flight/engi Correlat Determina Residual Mean flight/engi Correlat Determina Residual Mean
ne ion tion sum of squar ne ion tion sum of square
coeffici coefficient squares e coeffici coefficient squares error
ent error ent
20180107- 0.98786 0.97586 21.0696 0.045 20180111- 0.97216 0.94510 37.1265 0.07329
1-1 0 53 2-2 8
20180107- 0.99001 0.98011 19.1534 0.043 20180112- 0.98228 0.96487 25.6128 0.04657
1-2 4 41 1-1 6
20180107- 0.98847 0.97708 24.7984 0.054 20180112- 0.98560 0.97140 23.6340 0.04474
2-1 4 35 1-2 2
20180107- 0.99035 0.98079 22.5249 0.051 20180112- -0.0900 0.00811 162128. 3.67263
2-2 0 80 2-1 7 363
20180107- -0.1231 0.01517 162125. 4.019 20180112- 0.98984 0.97977 20.2052 0.04100
3-1 6 395 25 2-2 4
20180107- 0.98171 0.96376 16.6165 0.040 20180112- 0.99006 0.98023 22.3506 0.04495
3-2 4 69 3-1 5
20180107- 0.98557 0.97135 21.0099 0.048 20180112- 0.99110 0.98229 21.7607 0.04436
4-1 4 58 3-2 4
20180107- 0.98672 0.97361 21.2444 0.048 20180112- 0.98925 0.97861 19.5238 0.04829
4-2 3 85 4-1 1
20180108- 0.98679 0.97376 25.8756 0.042 20180112- 0.98919 0.97850 21.3051 0.05045
1-1 3 89 4-2 4
20180108- 0.98718 0.97452 26.3964 0.043 20180113- 0.98702 0.97421 23.0779 0.04602
1-2 1 32 1-1 7
20180108- 0.98614 0.97248 25.0010 0.046 20180113- 0.98929 0.97870 20.8766 0.04377
2-1 3 38 1-2 8
20180108- 0.98671 0.97359 25.0605 0.046 20180113- 0.99079 0.98167 21.3086 0.04421
2-2 9 43 2-1 7
20180108- 0.98173 0.97379 27.5915 0.056 20180113- 0.99267 0.98540 19.7830 0.04260
3-1 4 42 2-2 9
20180108- 0.98480 0.96983 25.8534 0.054 20180113- 0.98060 0.96158 16.7805 0.05492
3-2 2 61 3-1 3
20180109- 0.98114 0.96264 24.7646 0.059 20180113- 0.98244 0.96519 16.1183 0.05538
1-1 4 84 3-2 0 2
20180109- 0.98422 0.96869 22.9765 0.057 20180113- 0.95360 0.90936 53.9315 0.07353
1-2 0 64 4-1 9
20180109- 0.99004 0.98018 22.8375 0.045 20180113- 0.98355 0.96738 25.7718 0.05083
2-1 4 46 4-2 1
20180109- 0.99107 0.98222 21.7517 0.044 20180113- 0.09665 0.00934 158650. 4.04172
2-2 3 36 5-1 341
20180109- 0.98860 0.97734 21.8111 0.051 20180113- 0.98985 0.97980 19.2233 0.04449
3-1 5 60 5-2 9
20180109- 0.99076 0.98160 19.1133 0.048 20180113- 0.97717 0.95486 23.3939 0.06762
3-2 4 30 6-1 7
20180109- 0.98571 0.97163 25.1478 0.047 20180113- 0.98015 0.96069 21.6189 0.06501
4-1 1 42 6-2 1
20180109- 0.98711 0.97439 24.5046 0.046 20180114- 0.98304 0.96636 21.9265 0.04348
4-2 7 81 1-1 7
20180109- 0.98677 0.97371 22.4526 0.052 20180114- 0.98617 0.97253 20.7667 0.04232
5-1 4 07 1-2 9
20180109-5 0.98946 0.97902 19.86960 0.048 20180114-2 0.98740 0.97495 23.04935 0.04461
-2 99 -1
20180110-1 0.98122 0.96279 22.06120 0.051 20180114-2 0.98865 0.97744 21.95433 0.04354
-1 62 -2
20180110-1 0.98321 0.96670 20.64852 0.049 20180114-3 0.98175 0.96384 24.52669 0.04620
-2 94 -1
20180110- 0.97257 0.94589 27.5331 0.061 20180114- 0.98384 0.96794 23.1292 0.04486
2-1 1 40 3-2 9

7
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

Table 8, cont.
20180110- 0.97397 0.94863 26.2970 0.060 20180114- 0.98702 0.97421 23.0779 0.04602
2-2 0 00 4-1 7
20180110- -0.1320 0.01743 162128. 5.340 20180114- 0.98929 0.97870 20.8766 0.04377
3-1 4 037 75 4-2 8
20180110- 0.98235 0.96502 20.5213 0.060 20180115- -0.1080 0.01167 162123. 3.56226
3-2 8 09 1-1 4 594
20180110- 0.97263 0.94600 22.7488 0.070 20180115- 0.98692 0.97400 19.1958 0.03876
4-1 2 11 1-2 4
20180110- 0.97486 0.95035 21.3668 0.067 20180115- 0.98747 0.97510 22.3390 0.04532
4-2 9 95 2-1 5
20180110- 0.97701 0.95455 18.2964 0.060 20180115- 0.98845 0.97704 22.0284 0.04500
5-1 9 44 2-2 9
20180110- 0.98151 0.96337 15.1429 0.054 20180115- 0.98461 0.96945 27.9076 0.04697
5-2 7 99 3-1 3
20180111- 0.98581 0.97182 19.7683 0.046 20180115- 0.98940 0.97892 23.5977 0.04319
1-1 8 69 3-2 7
20180111- 0.98725 0.99467 19.1432 0.045 20180115- 0.97630 0.95317 40.6734 0.05932
1-2 4 95 4-1 4
20180111- 0.96975 0.94041 38.9376 0.075 20180115- 0.97845 0.95736 37.9350 0.05729
2-1 0 06 4-2 4

By table 8, we know, after verification with all flights, there are 4 poor results in all 75 results. The
4 poor results are 20180107-3-1, 20180110-3-1, 20180112-2-1 and 20180115-1-1. Besides these 4
results, the other results all have better accuracy. Their correlation coefficient is between 0.95 and 0.99,
the average of correlation coefficient is 0.98. Their determination coefficient is between 0.90 and 0.98,
the average of determination coefficient is 0.97; Residual sum of squares and mean square error are all
small. Therefore, we can prove this result of fitting curve is good and effective. About these four poor
results, they are all happened in engine one of these flights and they are confirmed later that this
engine had some unexplained data recording interference during these flights.

5. Conclusions
In this paper, a VBV operation curve is fitted using five parameters of VSV, N1, N2, TAT and T25
of civil engines during multiple flights. Using statistics and curve fitting toolboxes of MATLAB, after
data processing, correlation analysis, single parameter curve fitting and various parameters curve
fitting, there is the final fitting function expression of the VBV operation curve.
y=-283.5-504.4*x1+123.7*cos(x1)+342.4*sin(x1)+159.9*exp(x1)+ 0.0028x2 -0.0021x3 +0.0013x4
-0.0015x5
Where: independent variables x1, x2, x3, x4 and x5 are VSV, N1, N2, TAT and T25.
At last, verification with all flights show that, besides engine one during 4 flights, the results are
good with high relevance, small error which proves this fitting function expression is valid. And the
same time, since using MATLAB software for operation curve fitting of valves is simple and efficient,
and it can be further applied in aero engine modeling.

References
[1] Liu hao, Han jing, MATLAB R2016A self-study all in one, Electronics industry press, Beijing,
2016, pp.24-27.
[2] Zhuo jin wu, MATLAB application in mathematical modelling, Beijing university of aeronautics
and astronautics press, Beijing, 2014,pp. 123-126
[3] CFMI, CFM56-7B line maintenance course, Customer technical education centre, Cincinnati,
2017, pp. 300-302
[4] Wang Huan, Research on modelling and control of turbofan engine’s actuator (master degree
thesis), Nanjing university of aeronautics and astronautics, 2012, pp.68-71

8
SCSET 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1176 (2019) 052040 doi:10.1088/1742-6596/1176/5/052040

[5] Zhong shi sheng, Cui zhi quan, Fu xu yun, Baseline mining method of RR’s engine, Computer
integrated manufacturing systems, 2010, pp. 2265-2270
[6] Kong cheng an, Li wen hua, yin zhan, Monitoring aircraft performance with QAR data, China
academic journal, 2008, pp. 54-56
[7] Ding Shui Ting, FHA method for VBV position control function of FADEC system based on
aero-engine dynamic model, Procedia engineering, 2011, pp.567-579
[8] Yan feng, Fu wei gang, A modelling method for high compressor rotor speed (N2) baseline of the
aero-engine, Machinery design & Manufacture, 2015, pp. 148-151