Turbofan Engine Modelling and Control Design using Linear Quadratic Regulator (LQR)

The International Journal Of Engineering And Science (IJES)
|| Volume || 6 || Issue || 2 || Pages || PP 49-58 || 2017 ||
ISSN (e): 2319 – 1813 ISSN (p): 2319 – 1805
DOI : 10.9790/1813-0602024958 www.theijes.com Page 49
Turbofan Engine Modelling and Control Design using Linear
Quadratic Regulator (LQR)
Jackson Lutambo & Jiqiang Wang
College of Energy and Power Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing,
210016 China
--------------------------------------------------------ABSTRACT-----------------------------------------------------------
There are many applications in which gas turbine engine is used today, including aircraft propulsion for both
commercial and military purposes, and power generation and in all these Control systems technology has
played a fundamental role in enhancing performance. Modelling plays a significant role in the development of
the entire engine system performance. This paper investigated Linear Quadratic Regulator (LQR) model-based
control method to obtain estimates of performance parameters. The main control variable selected is the fuel
flow to control the rotational speed of high-pressure spool speed of the turbofan engine. Firstly a suitable
mathematical model of the engine is developed in MATLAB Simulink environment with both the inter-
component volume and the constant mass flow methods used. Equations of the mass flow rate and the torque
balance are incorporated in the steady state and dynamic state of the thermodynamic engine model. This
represents the engine model by a set of first-order differential and algebraic equations and a linearized model is
extracted for the analysis and design of a controller by LQR. It is demonstrated that LQR based controllers can
perform better than conventional PID controllers. The settling time, rise time and maximum overshoot for LQR
based controller are all less than those for PID based controller. The input also changes more accurately for
LQR than the PID controller compared.
Keywords: Control System Design, Linear Quadratic Regulator (LQR), Modelling and Simulation, Transient
Performance, Turbofan Engine.
-------------------------------------------------------------------------------------------------------------------------------------
Date of Submission: 26 January 2017 Date of Accepted: 28 February 2017
--------------------------------------------------------------------------------------------------------------------------------------
NOMENCLATURE
A = state coefficient matrix
B = input coefficient matrix
Cp = specific heat capacity
HV = Heating value of fuel
J = LQR performance measure
J = moment of inertia
K = LQR gain matrix
LQR = linear quadratic regulator
m = mass flow rate
N = rotational speed
P = pressure
Q = matrix of error weights
R = control weights
T = temperature
U = input matrix
X = system state matrix
 = temperature ratio
 = pressure ratio
W= work
 = ratio of specific heats
 = efficiency

Turbofan Engine Modelling and Control Design using Linear…
I. INTRODUCTION
The turbofan engine has found its use mostly in aircraft propulsion. It operates by absorbing the excess
energy from the high-pressure turbine to drive a fan and low-pressure compressor located upstream of the main
compressor section [1]. In aircraft propulsion, it performs a number of key functions, firstly it provides the
propulsive energy to propel the aircraft throughout its route while achieving appropriate performance and
efficiency. Secondly, it provides the prime source of energy for the on-board systems by generating electrical
power and driving other components responsible for producing hydraulic and pneumatic power. Finally, it
provides the air to create a habitable environment for the crew, passengers, and avionics equipment. It also costs
a substantial amount of investment and its incorporation into the aircraft affects both airframe and systems in
terms of safety, reliability, and cost of ownership [2].
The goal of any engine control system is to enable the engine to realize high accuracy, reliability,
performance and efficiency in all exploitation conditions [3]. The modern turbofan is highly fuel efficient and
control technology has contributed significantly to the attainment of this effect. In recent years great research
work has been done in the field of modelling and simulating of gas turbines. Making models of gas turbines and
their related control systems have been a useful technical and cost-saving strategy for performance enhancement
and testing before manufacturing [4].
At first, a suitable mathematical model for the turbofan engine is built in Matlab/Simulink software [5].
After building the model it is desirable to extract a linearized model to be used for analysis and design. Most of
the analytical techniques that are commonly used for analysis and design of control systems can only be applied
to linear models and this is accomplished within Matlab/Simulink. The idea of using fuel flow rate to control
thrust as in the earlier hydromechanical systems is still the strategy in many gas turbine engines today. Other
control signals are often used for limiting the engine from exceeding its operating range [6]. In this model, the
input signal is the fuel flow and the output signal is the high-pressure compressor rotational speed. The more
accurate control of engine thrust which is the primary control objective of the gas turbine engine can be obtained
through the control of compressor speed [7]. The desired performance requirements are to obtain faster response
without any overshoot. In this paper, controller design has been carried out for speed control of twin spool
turbofan engine by LQR control method and fuel flow rate was used as the input. The objective of the LQR
control design is to calculate the optimal feedback gain matrix K such that the feedback law
u K x  : (1)
Minimizes the cost function
0
( )
t
T T
J x Q x u R u  : (2)
With the constraint equation being
x A x B u  : (3)
After obtaining the A and B matrices in Simulink the main task is selecting Q and R matrices subject to the
further constraints that Q be symmetric positive semi-definite and R be symmetric and positive definite.
II. MODELLING OF TURBOFAN ENGINE.
Many software packages are now available for modelling systems and a considerable research work has
been carried out in the field of modelling and simulation of turbine engines with the view of understanding the
subject well before building physical models. This great work has been carried out because the need for and use
of gas turbines have become more apparent in the modern industry. Modelling of turbine engines and control
systems has been a useful technical and cost-saving strategy for performance optimization of the equipment
before making and testing the final product [7]. Gas turbine models and simulators can be used for off-design
performance prediction, performance deterioration and evaluation of emissions, turbine creep life usage and the
engine control systems [8]. Mathematical modelling is considered as a general approach to system modelling and
uses mathematical language to describe and predict the behaviour of a system.
A mathematical model of a gas turbine engine consists of mathematical equations, maps, and tables
which describe the thermodynamic relationships of various variables in the engine. The model developed here
consists of individual components such as inlet, fan, compressors, combustor, turbines and nozzle [9]. The
components require a number of input variables to generate the output variables. The schematic representation of
an axial flow, unmixed-flow, twin-spool and high-by-pass ratio turbofan engine typical of the one developed in
this paper is shown in fig 1. The model is built in the Matlab simulation environment and its Simulink toolbox
with inter-component volume and constant mass flow assumed. The complete model of turbofan engine
components with the control volumes and torque balance on the rotating group developed in Simulink is shown

in fig 2. The thermodynamic equations of the components and the set of static and dynamic equations were used
in this model.
Figure 1 Turbofan Unmixed Flow Engine
Figure 2 Simulink Model of a Turbofan Engine
2.1 Compressor
The task of the compressor is to increase the pressure of air required by the engine. The axial flow
compressor is considered here i.e. one in which the flow enters in an axial direction parallel to the axis of
rotation. The overall performance maps are used to represent the steady performance of the compressor which is
good enough for real-time simulation. The typical performance maps and the governing equations are similar for
the fan, low-pressure compressor, and high-pressure compressor. The map values of airflow and efficiency are
computed from functions of pressure ratio and corrected speed using 2-D lookup tables were through
interpolation the mass flow and efficiency are determined [10]. The following equations are used to calculate
various parameters together with the map.
Temperature ratio,
/t std
T T  : (4)
Pressure ratio,
/t std
P P  : (5)
Normalized speed
 / *co rr d es
N N N  : (6)
Mass flow rate,
3 3
* /corr
m m    : (7)
Temperature,

 
 
 
( 1 ) /
3 2
* 1 1 / * 1fa n fa n
T T
 
 

   
  
: (8)
Work,
 3 3 2
* *fan p
W m C T T  : (9)
2.2 Combustion Chamber
The combustion chamber provides an environment where fuel/air mixture burns and release the hot
gases with its energy to the turbine and then the nozzle sections of the engine. The combustion process in the
model is simplified and assumed to be instantaneous after the fuel injection but the actual process is a complex
one.
Theoretically, in the combustion process the inlet and outlet pressures of the combustion chamber are
assumed to be the same but in practice, the pressure drops in the combustion chamber. Because of this pressure
drop, the gases in the combustion chamber can flow rearwards. Temperature is given by the Energy balance
equation
 
5
6
* * * *
*
co m b eff f a ir p
a ir f p g a s
H V m m C T
T
m m C
 




 
 
: (10)
2.3 Turbine
The Turbine is used to extract sufficient energy from the hot gases of the combustor to drive the
compressor and other components attached, like pumps and generators. In this model the turbine is assumed to
be an axial flow turbine and both low and high-pressure turbines are modelled with same equations [11]. But
while modelling the engine in Simulink, the turbine is modelled as a single block by stacking the stages of the
turbine into a single block. Dynamic behavior of all the individual stages is also stacked into a single block with
only the inlet and final exit conditions of the turbine. From the values of the corrected mass flow rate and the
turbine efficiency, the actual mass flow rate and the temperature can be calculated from the expressions,
Actual mass flow rate,
6
7
*c o rr
m
m





 : (11)
Temperature,
 
1
7 6
* 1 * 1
g a s
g a s
h p t h p t
T T


 

   
    
   
: (12)
Work of the turbine,
7 6 7
* * ( )h p t p g a s
W m C T T
  : (13)
2.4 Nozzles
A convergent nozzle is considered in modelling the engine. The partially expanded gas coming from the
turbine is accelerated to a high velocity in the nozzle and expand to the ambient pressure.
The nozzle is to be checked first to define whether it is choked or not. Thus the critical pressure is
calculated from the relationship:
1
8
2
*
1
c
P P




 
  
 
: (14)
Thrust can be calculated from
9 0 9
* ( ) * ( )n o zzle a m b
F m V V A P P    : (15)
9 8 9
2 * * ( )p
V C T T  : (16)
9 8
* ( 2 / ( 1))T T   : (17)
9 a m b
P P : (18)
For unchoked nozzle and
9
* *V R T : (19)

1
9 8 9 8
* ( / )T T P P

 
 : (20)
9 c ritic a l
P P : (21)
For choked nozzle
2.5 Volume Dynamics
Inter-component volume dynamics between various components to account for incoming and outgoing
mass flows is considered at engine locations where the dynamics are important or required to avoid the need for
an iterative solution. The storage of energy and mass take place in these volumes and the dynamic pressure can
be calculated by the ideal gas law integration
 in o u t
m m
P d t
V

 
 
: (22)
2.6 Shaft Dynamics
At steady state, the operating points of the compressor must match with that of the turbine. Any mismatch
between the compressor and turbine will produce unbalanced torque. Therefore the change of torque is a
function of the energy difference between the turbine and the compressor and is given by
 
2
30 1
* t c
N P W P W
JN
 
 
 
 
 : (23)
III. CONTROLLER DESIGN BY LINEAR QUADRATIC REGULATOR
The objective of the LQR controller is to select the gain K equation (1) that minimizes the performance
index (J) equation (2) [13]. A controller is designed to keep a stationary system within an acceptable deviation
from a reference condition using an acceptable level of control. The controller moves the system from an initial
state x(t0) to a terminal state using input control u(t) without exceeding acceptable levels of the state. The linear
quadratic control refers to a body of techniques using optimal control developed for control system design. The
plant is assumed to be a linear system in state space form and the cost function is a quadratic function of the
plant states and control input.
The cost function J represents the weighted sum of energy of the state and control. The weighting
matrices Q and R control the characteristics of the system. By weighting R more than Q the control energy is
penalized heavily translating into smaller actuators, motors, and gains required for implementing the control law.
By weighting Q much larger than R the state is similarly affected resulting in a damped system, which avoids
larger fluctuations or overshoots in system states. An appropriate choice of Q and R must be made to obtain
acceptable levels of x(tr), x(t) and u(t) [14]. The system can be represented in state space as shown in fig 3
Figure 3 State-Space Representation of the System
To solve using LQR controller design the following assumptions are made:
1 the entire state vector x(t)is available for feedback.
2 the system is both controllable and observable.
The design steps are as follows:
Select design parameter matrices Q and R
Solve the algebraic Riccati equation for P

1
( ) ( ) ( ) 0
T T
A P t P t A Q P t B R B P

    : (17)
Determine the optimal state feedback gain K
1
* *
T
K R B P

 : (18)
*u K x  : (19)
The system is described by
x A x B u
y C x D u
  
 
  

: (20)
The performance index is defined by
 
0
T T
J x Q x u R u d t

  : (21)
Where Q: nxn semi-positive definite matrix
R: mxm positive definite matrix
IV. SIMULATIONS AND RESULTS
After the model is built in Simulink as described in section 2 above, first there is a need to extract the
linearized state space model from the entire system. In order to extract we need to define the input which is fuel
flow in this case and the output signal which is high-speed spool speed of the model. At the Matlab prompt using
the command linmod.m the linearized model is obtained from the systems of ordinary differential
equations(ODEs) containing the matrices A,B,C and D. The equivalent transfer function obtained is
X(s)=
1.064e04 s^5 + 2.009e07 s^4 + 8.465e09 s^3 + 1.122e12 s^2 + 3.71e13 s + 7.279e13
s^6 + 1807 s^5 + 7.462e05 s^4 + 9.662e07 s^3 + 3.049e09 s^2 + 1.19e10 s + 1.29e10
Initially, the system is simulated to obtain the open loop response and fig 7 shows the simulation of
high-speed spool speed in open loop and fig 8 shows the associated bode plot. As can be seen from the figure the
system has slow rise time (0.894 seconds), settling time (1.53 seconds) and has no overshoot so it needs to be
improved by adding a controller.
0 0.5 1 1.5 2 2.5 3
0
1000
2000
3000
4000
5000
6000
From: Fuel flow(Wf) To: high speed(Nh)
Open Loop Step Response
Time (seconds)
HighPressureRotationalSpeed(Nh)
Figure7 Open Loop Response

-50
0
50
100
From: Fuel flow(Wf) To: high speed(Nh)
Magnitude(dB)
10
-1
10
0
10
1
10
2
10
3
10
4
10
5
-135
-90
-45
0
Phase(deg)
Bode Diagram for Open Loop Response
Frequency (rad/s)Figure 8 Bode Diagram for Open Loop Response without the Controller.
4.1 Proportional Integral Controller
The PI controller was implemented and was found to give good enough response. After several trials the gains
Kp=0.3 and Ki=6.3101 gave the desired response. The settling time was found to be 0.000103 seconds, rise time
was 0.000643 seconds and 1 % overshoot and fig 9 and fig 10 show the step response and bode plot respectively
after simulation.
Step Response with PI Contoller
Time (seconds)
Amplitude
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
x 10
-3
0
0.2
0.4
0.6
0.8
1
1.2
1.4
From: Fuel flow(Wf) To: Out(1)
Figure 9 Step Response with PI Controller
-30
-20
-10
0
10
From: Fuel flow(Wf) To: Out(1)
Magnitude(dB)
10
0
10
1
10
2
10
3
10
4
10
5
-90
-45
0
Phase(deg)
Bode Diagram with PI controller
Frequency (rad/s)
Figure 10 Bode Diagram with PI Controller
4.2 LQR Controller

When designing LQR controller, the function in Matlab can be used to determine the value of the gain
vector K to determine the feedback control law. This is done by initially choosing parameter values using
Anderson approximation which is given by
*
T
Q C C : (22)
and R=1.
1 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 0 0 0 0
0 0 0 0 0 0
Q
 
 
 
 
  
 
 
 
 
Using the Q obtained and R=1 the feedback gain is found to be
K= [0.9997 0.0000 0.0000 0.0000 0.0000 0.0000]
The symmetric matrix P is found to be
0 .0 0 0 0 9 3 3 2 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 0 0 0 0
0 0 0 0 0 0
P
 
 
 
 
  
 
 
 
 
And the system poles were
-10630, -1359, -316, -167, -46, -2
The step response of the closed loop system with LQR controller is as shown on fig 11 and the
associated bode plot in fig 12. From the graph and the poles of the system, the system behaves as if it were first
order system because the poles are far apart. The pole at -2 dominates the transient response behaviour as it is
the term that decays slowly. The poles of large negative values can be ignored as their transient response
behavior [15] is assumed to be too fast to contribute to the system dynamics as this can be seen from the figure.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
x 10
-4
0
0.2
0.4
0.6
0.8
1
From: Wf To: High Speed(Nh)
Closed Loop Step Response
Time (seconds)
Amplitude
Figure 11 Closed Loop Response with LQR Controller

-40
-30
-20
-10
0
From: Wf To: High Speed(Nh)
Magnitude(dB)
10
0
10
1
10
2
10
3
10
4
10
5
10
6
-90
-45
0
Phase(deg)
Bode Diagram For Closed Loop with LQR Controller
Frequency (rad/s)
Figure 12 Bode Diagram for Closed Loop with LQR Controller
As shown from fig 11 the response is much faster and it meets the specifications. The settling time is
now 0.00036 seconds, rise time is 0.000206 seconds and there is no overshoot. Table 1 below shows the
different responses of the system in open loop, with PI controller and LQR controller. The LQR controller
performance was better than the PI controller as it shows faster responses, in terms of settling time, rise time and
overshoot. Therefore LQR controller better meets the requirements of design specifications than then the PI
controller in this case.
Table 1 Comparison between PI and LQR
Characteristic Open Loop PI LQR
Settling Time 1.53 sec 0.00103 sec 0.000365 sec
Rise Time 0.894 sec 0.000643 sec 0.000206 sec
Steady State 1 1 1
Overshoot 0.0198% 1% 0%
V. CONCLUSION
In this paper, the LQR control technique was used to design a controller for the turbofan engine. The
model was first built in Matlab Simulink and a linearized state space representation was extracted to be used for
analysis and design of the controller. Results show that the Linear Quadratic Regulator is capable of stabilizing
the system and produce system responses with smaller rise time, settling time, and no overshoot as compared to
the PI controller. The LQR controller is observed to have performed satisfactorily for the SISO case considered
in this paper. In case where there are too many variables and inputs LQR method may be a viable approach were
classical design methods are limited.
ACKNOWLEDGEMENT
I would like to thank the university and staff of Nanjing University of Aeronautics and Astronautics for funding
to make this paper published.
REFERENCE
[1] H. Klause, Jet Engines, Fundamentals of theory, Design and operation, Motorbooks International Publishers and wholesalers,
Osceola, USA 2003.
[2] Bernie M, Roy L. Gas Turbine Propulsion Systems. A. John Wiley & Sons, Ltd. West Sussex, UK, 2011.
[3] J. Sun, V. Vasilyev, Ilyasov, Advanced Multivariable Control Systems of Aeroengines, buaa press. Beijing, China, 2005.
[4] E. Tsoutsanis et al. Dynamic Performance Simulation of an Aeroderivative Gas Turbine Using The Matlab Simulink
Environment, Proceedings of the ASME International Mechanical Engineering Congress & Exposition IMECE November,
2013 San Diego, California, USA.
[5] A. M. Y. Razak, Industrial Gas Turbines: Performance and Operability, Woodhead Publishing Limited, Cambridge, England,
2007, First Ed.
[6] H. Melker, Multivariable Control Design for a Jet Engine, Proceedings of 13th
Triennial World Congress, San Francisco, USA
1996. Volvo Aero Corporation, Sweden.
[7] A. J. Sobey and A. M. Suggs, Control of aircraft and Missile Power Plants, John Willey and Sons inc. New York, 1983.
[8] M. Bazazzadel, H. Badihi and A. Shahriari, Gas Turbine Engine Control Design Using Fuzzy Logic and Neural Networks,
International Journal of Aerospace Engineering, 2011, Volume 2011.
[9] Q. R. Ali and S. Mahmoud Application of Optimal control and eigenvalue/eigenvector assignment for Jet engine control. I mech
E proceedings C 16/87, 2011, 103-109.

[10] A. Hamid, Modelling, Simulation and Control of Gas Turbines Using Artificial Neural Networks. University of Canterbury
Christchurch, New Zealand,2014.
[11] M. H Ahmad, S.S Mohamed et al. Gas Turbine LQR, Integral Controllers and Optimal PID Tuning by Ant Colony Optimization
Comparative Study. International Journal of Computer Science and Telecommunications 2013,Volume 4, Issue 1.
[12] A. T. Alouani Application Of Linear Multivariable Control Design For The Ge-21 Turbofan Jet Engine, Tennessee
Technological University Cookeville, Tennessee USA, 1989.
[13] A. J. Ujam, Modeling Performance Characteristics of a Turbojet Engine, International Journal of Manufacturing, Material and
Mechanical Engineering Research 2013,Vol.1, No. 1, pp.1-16.
[14] D.S. Naidu, Optimal Control Systems, CRC Press, Idaho State University,Pocatello, Idaho, USA, 2003.
[15] K. Ogata Modern, Control Engineering, Tom Robbins, New Jersey, USA, 1997, 3 Ed.

Turbofan Engine Modelling and Control Design using Linear Quadratic Regulator (LQR)

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Turbofan Engine Modelling and Control Design using Linear Quadratic Regulator (LQR) (20)

Recently uploaded (20)

Turbofan Engine Modelling and Control Design using Linear Quadratic Regulator (LQR)