Professional Documents
Culture Documents
Depth Control of A High Speed Underwater Vehicle Using Model Predictive Control
Depth Control of A High Speed Underwater Vehicle Using Model Predictive Control
Indian Institute of Technology (Banaras Hindu University) Varanasi, India, Dec 9-11, 2016
Abstract: Super cavitation is an advanced technology that friction reduces and almost becomes zero. The main
makes underwater vehicles to reach high speed. These vehicles problems associated with supercavitating vehicles are
are in general highly complex, nonlinear, multi input multi complex dynamics, instability, nonlinearity and time delay
output, and coupled systems. Modeling and control plays a between vehicle and cavity interaction [20-21].
crucial role in designing the supercavity vehicles. Model
Predictive Control technique has been applied to diving plane
of the vehicle to analyse pitch and depth control of an
underwater vehicle. The supercavitating vehicle has cavitator in the front portion
and four fins at the aft portion. These two places play a
Key Words – Supercavity vehicle, MPC, stability and LQR crucial role to control and stabilize the vehicle. The main
I. INTRODUCTION targets in supercavitating vehicles are a) balancing the
weight of the vehicle b) generation and maintenance of
One of the limitations of the underwater vehicle is limited cavity c) guidance and control d) vehicle stability [8]
speed. Maximum speed limit is around 40 m/s for common
underwater vehicles. The reason behind this is due to the
drag induced by the skin friction. To overcome this
difficulty a concept of super cavitation has been introduced
to improve the speed of the vehicle up to nearly 100 m/s.
Cavitation is defined as the process of formation of the
vapour phase of a liquid when it is subjected to reduced
pressures at constant ambient temperature [3]. The
relationship between velocity and pressure is inversely
proportional as per the Bernoulli’s law. The pressure
decreases when the speed of the vehicle increases.
Cavitation is mainly found at sharp corners of a body where
the velocity is also very high [2].
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:24:09 UTC from IEEE Xplore. Restrictions apply.
െ݊ െ݊ ͳ ͳ ʹ
ܽʹͶ = VTS ቀͳܥ ቁ-VTቀͳܥ ቁ ;ܮ
݉ ͻ ݉ ͵ ͵
Few control techniques have been applied on
supercavitating underwater vehicles. Sliding mode control
has been applied to analyse the dynamics and control of ܽͶʹ =
ܸܶ ͳܥ
ቀ
ͳ
െ
ͳͳ
݊ቁ; ܽͶͶ =
െͳͳ ܣܮܸ݊ܶ ͳܥ
* ;
underwater vehicle [6]. Robust control technique is also ݉ ͵ ͵ ͵ ݉
The structure of this paper is as follows: section-I describes ͳͳ ݊ܶ ʹ ܸ ͳܥ ͳ ܶ ʹ ܸ ͳܥ
about introduction. Section II is about the modeling of a ܾͶͳ =- * ; ܾͶʹ = * ; ܿʹ =g ;
͵ ݉ ͵ ݉
vehicle. Control techniques have been developed in section-
III-a & b. Section –IV and Section V highlights the results
ܶ െͳ ܵ ͳͳ ܶ
and conclusion of the paper. ݀ʹ = ቀ ቁ*ቀ ܮ ቁ ; ݀Ͷ = *
݉ ͵ ܮ ͵ ݉
y= [1 0 0 0;0 0 1 0]x;
219
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:24:09 UTC from IEEE Xplore. Restrictions apply.
State vectors are x = [z w θ q]T The main advantage of quadratic optimal control scheme is
to stabilize the system and its responses are plotted in Fig.2
ߜ
and the input vector u= ݁ ൨
ߜܿ
and its transfer functions for depth and pitch are as follows
ܼሺݏሻ ʹͲͷ Ǥͻ ʹ ݏͺǤͲͻʹ݁ͲͶݏെ͵ǤͲͷ݁ͲͶ
=
ߜ ݁ ሺݏሻ ݏͶ ͳͷǤͷ ͵ ݏ͵Ͷͺͷ ʹ ݏ
Equation (4) can be written as Fig.2 Stabilization of states using LQR technique
III-a Control technique Fig.3 shows a block diagram of model predictive control.
LQR is a linear optimal control law for a linear time MPC technique has been attempted to control the pitch and
invariant system with quadratic performance index. It has depth of supercavitating vehicle. The controlled variables of
been developed and widely used in various applications. vehicle are pitch and depth and the disturbance variables are
Selection of Q and R plays an important role in LQR ocean current, ocean wave and winds. Controller output is
method [8]. called as manipulated variables. The Controlled Variable-
CV, Disturbance Variable-DV and Manipulated Variable-
MV are very important while designing a MPC. The three
The standard dynamic model of the system is described by: important components of MPC are 1) A supercavitating
vehicle model which predicts the output in the future up to
ݔሶ =Ax+Bu
the prediction horizon (around 120 scans) 2) A set point
y =Cx+Du
future trajectory for the same number of scans as the
predicted output trajectory. 3) The difference between
The performance index is defined [18] as
predicted output and set point future trajectory generates an
λ error vector and it is used to compute the control algorithm
J= ݔܳ ܶ ݔ Ͳ ݑܴ ܶݑሻ݀ݐ (5) [17, 22].
Q=C’*C;R=I
The vehicle model plays an important role in model
Where Q is a positive definite/semi definite or real predictive control. The manipulated variable is applied on
symmetric matrix, and R is the weight matrix. In equation vehicle to minimize the disturbances variables (ocean winds
(5) I denotes identity matrix. The control law is and waves) acting on vehicle. The main aim of MPC is to
maintain pitch and depth control of vehicle in desired
u=-Kx (6) trajectory. If CV’s (pitch and depth) does not follow desired
trajectory then CV is taken as feedback and applied to error
K can be found as detector to compare with set point reference trajectory. The
K=ܴ െͳ ܶܤP error detector output is actuating signal which is applied on
MPC input. The cost function of MPC depends on the
K=[0.9867 0.0430 -20.9121 -0.3752; quadratic error between the future control variable and the
-0.1627 -0.9919 -0.2221 0.4866] future set point reference variable within limited time
horizon. The predicted control algorithm develop outputs to
and P can be determined by solving the continuous time minimize the sum of the squared errors over the prediction
algebraic Riccati equation [18]: horizon, taking into account several future manipulated
variables moves
ܲ ܶܣ ܲ ܣെ ܲ ܴܤെͳ ܲ ܶܤ ܳ ൌ Ͳ (7)
220
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:24:09 UTC from IEEE Xplore. Restrictions apply.
Set Point
Control Supercavitating
+/- Algorithm Vehicle
Controlled Variable
Prediction
Model output
Distrubance
+/-
Model Output
Correction
Fig 3 Model Predictive Controller of a supercavitating vehicle
in its calculations. MPC can easily handle all constraints on Prediction of x at sample (k+4) where prediction was made
inputs and states and it is also useful for control and at sample k and prediction of y at sample (k+6) where
optimization problems. prediction was made at sample (k+2)
A general expression of the n step ahead prediction is given
Consider the generalised discrete time state space model
as
[22]
x(k+1)=Ax(k)+Bu(k) (8)
x(݇ ݊ᑻ݊ሻ= ݊ܣx(k) + ݊ܣെͳ Bu(kǀk) +
ݕሺ݇ሻ ൌ ݔܥሺ݇ሻ ݀ݑሺ݇ሻ(9) ݊െʹ
ܣBu(k+1ǀk)+………..+ABu(k+n-2ǀk)+Bu(k+n-1ǀk)
A general n-th order expression of the n step ahead
(13)
prediction is given as follows
x(k+n)= ݊ܣx(k)+݊ܣെͳ Bu(k)+݊ܣെʹ Bu(k+1)+….+ABu(k+n-
y(k+nǀn)=C ݊ܣx(k) +C(݊ܣെͳ Bu(kǀk) +
2)+Bu(k+n-1) (10) ݊െʹ
ܣBu(k+1ǀk)+………..+Bu(k+n-1ǀk))+d(k) (14)
The system output can be determined simply using
The double subscript makes it clear where a value is
y(k+n)=Cx(k+n)+d(k+n); d(k+n)=d(k); (11)
notionally ‘in the future or a predicted value’ as opposed to
known.
substituting eq.no (10) in eq no (11), we get
U= [Δ( ܶݑk) ο( ܶݑk+1) … ο( ܶݑk+Nc-1)]T (15)
y(k+n)= C ݊ܣx(k) x(k)= [( ܶ ݔk+1ǀk) ( ܶ ݔk+2ǀk) …. ( ܶ ݔk+Npǀk)]T (16)
+C(݊ܣെͳ Bu(k)+݊ܣെʹ Bu(k+1)+…+ABu(k+n-2)+Bu(k+n- The state space model is used to compute the future state
1))+d(k) (12) vectors and output vectors
Y=Fx+u (17)
This prediction mixes up past and future data, so it is ܣܥ
ې ʹܣܥ ۍ
advisable to be more careful with notation and construction ۑ͵ ێ
of the predictions [16,17] Where F= ۑ ܣܥ ێand
ێǤ ۑ
A common notation used in the literature is double subscript ێǤ ۑ
[18]. The first term denotes the sample of the prediction ے ݊ܣܥۏ
ܤܥ Ͳ Ͳ Ͳ ǥǤ Ͳ
(how many steps ahead) and the second term denotes the ܤܣܥ ۍ ܤܥ Ͳ Ͳ ǥǤ Ͳ ې
sample at which prediction was made (only used for ێ ʹ ۑ
prediction and not past) ܤ ܣܥ ێ = ܤܣܥ ܤܥ Ͳ ǥǤ Ͳ ۑ
ێ Ǥ Ǥ Ǥ Ǥ Ǥ Ǥ ۑ
ێ Ǥ Ǥ Ǥ Ǥ Ǥ Ǥ ۑ
݊ܣܥۏെͳ ݊ܣܥ ܤെʹ ܤ ݊ܣܥെ͵ ܤǤ Ǥ ݊ܣܥെܰܿ ے ܤ
For example x((݇ Ͷሻᑻ݇ሻ & y(݇ ᑻ݇ ʹሻ
Where n=Np
Constraints to be defined as
221
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:24:09 UTC from IEEE Xplore. Restrictions apply.
yminyሺ݇ሻ ymax Table-II
Δumin ߂uሺ݇ሻ ߂umax Parameters used in MPC Controller
uminuሺ݇ሻ umax Parameter Value
Control interval 0.2
Prediction horizon 10
Optimization criteria used in MPC is Control horizon 2
Estimator gain 0.35
ܰ
σ݇ൌͲ ሼܳͳ ሾ݁ͳ ሺ ݇ݐሻሿʹ ܳʹ ሾ݁ʹ ሺ ݇ݐሻሿʹ ܳ͵ ሾ݁͵ ሺ ݇ݐሻሿʹ Input Weight 0.2
Output Weight 1
݊ܳ ڮሾ݁݊ ሺ ݇ݐሻሿʹ ሽ+
ܴͳ ሾ߂ ͳݑሺ ݇ݐሻሿʹ
σܰܿ
݇ൌͲ ൜ ൠ
ܴʹ ሾ߂ ʹݑሺ ݇ݐሻሿ ܴ͵ ሾ߂ ͵ݑሺ ݇ݐሻሿʹ ܴ݊ ڮሾ߂ ݊ݑሺ ݇ݐሻሿʹ
ʹ
IV Results
(18) All the simulations are carried out in MATLAB
=sum of the future weighted squared control errors + sum of environment.The MPC has been applied to the underwater
the future weighted increments of control variables vehicle taking sinusoidal input in Fig.4. The trajectories of
where Np and Nc are prediction and control horizons. Q and vehicle states are following desired trajectories. Further step
R is the symmetric positive definite matrix. response of vehicle with respect to planning force, depth
and pitch are also analyzed and plotted in Fig.5 and Fig.6.
A convenient orthogonal basis for discrete time system
is
Tuning of MPC
Model Horizon H: ‘H’ should be more than the open-loop
settling time (defined as 99 or 95% of response). Commonly
25 ≤ H ≤ 75.
Sampling Period
It should be selected to capture important dynamic Fig.4 Sinusoidal trajectory of vehicle states
information.
Prediction and control horizon P & M
Generally, PM. Increasing M increases controller
aggressiveness and decreases robustness. A suitable first
guess P Үt60/T (number of samples to reach 60% of open-
loop response).
Weighting Matrices Q and R
Q is the output weighting matrix and R is the input weight
matrix.
222
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:24:09 UTC from IEEE Xplore. Restrictions apply.
[13] J Dzielski and A Kurdila,“A benchmark control problem for
supercavitating vehicles and an initial investigation of solutions,”
Journal of Vibration and Control, vol. 9, no. 7, pp. 791–804, 2003
[14] Ankul Goel. “Robust control of supercavitating vehicles in the
presence of dynamic and uncertain cavity”. Florida:University of
Florida, 2005
[15] Fossen TI. Guidance and Control of Ocean Vehicles, J.Wiley&
Sons, New York, USA, 1994
[16] Liuping Wang, “Model Predictive Control System Design and
Implementation Using MATLAB”, Springer, Advances in Industrial
control, 2009
[17] Kirschner IN, Kring DC, Stokes AW, Uhlman JS. “Control strategies
for Supercavitating vehicles”. J Vibration and Control 8:219-242,
2002
[18] D Mayne, M Seron, and S Rakovic,“Robust model predictive control
of constrained linear systems with bounded
disturbances,”Automatica, vol. 41, pp. 1136–1142, 2006
Fig.6 Step response of vehicle states depth and pitch. [19] Ivan N Kirschner, James S.Uhlman “Overview of High-speed
V Conclusion supercavitating vehicle control”. AIAA Guidance, Navigation and
Control Conference and Exhibit.21-24 August, Keystone, Colorado,
The main contributions of this paper are 1) vehicle model is 2006
stabilized with the help of LQR. 2) Model Predictive [20] E Paryshev, “Approximate Mathematical Models in High-Speed
Control is then applied on longitudinal plane to control the Hydrodynamics,” J. Eng. Math., 55, pp. 41–64, 2006
[21] G Lin, Balachandran, and E Abed, “Nonlinear Dynamics and
speed, depth, pitch angle and pitch rate. Considering the
Bifurcations of a Supercavitating Vehicle,” IEEE J. Ocean. Eng.,
model from [1], all the simulations were carried out in Vol.32, pp. 753–761, 2006
MATLAB. Set point tracking of vehicle was analysed. It has [22] Guojian Lin, Balachandran and Eyad H Abed “Nonlinear Dynamics
observed that Vehicle is both controllable and stable. The and Bifurcations of a Supercavitating Vehicle.”[J] IEEE Journal of
Oceanic Engineering, 2007:753-761, 2007
future scope of the paper is non minimum portion of
response can be eliminated using smith predictor control
algorithm.
References
223
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:24:09 UTC from IEEE Xplore. Restrictions apply.