Professional Documents
Culture Documents
1 www.ijeas.org
Missile Autopilot Design using Adaptive CMAC Supervisory Controller
x1 f1 x1 x2 g1 x3
u*
1
g x
f x d t yd K T E (8)
x 2 f 2 x1 g 2 x3 (3)
x3 a u a x3 where K k1 , k2 2 , in which ki i 1,2 are positive
T
where constants. Applying the control law (8) to system (6), the error
x1 , x2 q, x3 , u c dynamics is obtained.
f1 x1 C1z1 x1 z 2 x1 M m e k1e k2e 0 (9)
f 2 x1 C2 m1 x1 m2 x1 M m K is selected such that the real part of the solutions of
hs s 2 k1s k2 are strictly negative. This means that
QS Q Sd
g1 C1Bz , g 2 C2 Bm , C1 , C2 (4) tracking of the reference trajectory is asymptotically achieved
where lim et 0 when t for any starting initial
mV I yy
2 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-3, Issue-7, July 2016
u y + yd
Plant
w , m , v
Adaptive Laws
, C, H
+ uA + uCMAC E Tracking
CMAC
Error Vector
+ +
us uC
Estimation Law
Compensation
Controller
Adaptive CMAC
us E
Supervisory Controller
Fig. 1. Block diagram of the missile autopilot system based on the adaptive CMS controller.
where P nn is a positive-definite symmetric matrix that Substituting (8) and (15) into (14) and considering the
satisfies the Lyapunov equation case I 1 , yields
1
T P P Q (13) 1
VS E T QE E T PGm f x d t yd
and Q nn is also a positive-definite symmetric matrix.
2 g x
By using Eqs. (11), and (13) and taking the derivative of
K T E uA
1
g x
f U
x d U
yd K T E u A
VS with respect to time, we have L
1 1 1
VS E T PE E T PE VS E T QE 0 (16)
2 2 2
1 T T 1
E PE E T PE E T PGm u * u A u S Using the supervisory u S controller presented in (15),
2 2 when VS V , the inequality VS 0 can be obtained even for
1
E T QE E T PGm u * u A u S
2
non-zero value of the tracking error vector E . From (16), the
supervisory controller is capable of leading the tracking error
1
E T QE E T PGm u * u A E T PGmu S
2
(14)
to converge to zero.
However, due to the presence of sign function and the
In order to formulate the supervisory control law u S such selection of the bounds f U x , g L x , d U , an excessive and
that V 0 , it is necessary to know the bounds of the chattering control effort will be resulted. Moreover, the
S
transient tracking performance may be not satisfied.
functions f x and g x . Therefore, we make the following Therefore, to overcome these phenomena, the adaptive
assumption. CMAC will be formulated in the following subsection.
Assumption: The bound functions f U x , gU x and g L x are Association
Memory Space A
known such that f x f x and g L x g x g x for
U U Weight Memory
Space W
all x U C , where f x , g U x and g L x 0 .
U Input Space S
Output Space Y
Moreover, the aerodynamics uncertainties is bounded
by d t d U .
s1 y1
uS I sgn ET PGm
uA
1
g L x
f U x d U yd K T E (15)
where sgn is a sign function, and the operator index Receptive-Field
1 VS V Space T
I V is a positive constant. Fig. 2. Architecture of a CMAC
0 VS V
3 www.ijeas.org
Missile Autopilot Design using Adaptive CMAC Supervisory Controller
v v
T T nnR
discrete regions called elements. The number of elements, nE, and 1 ,, vTk ,, vTnR .
is designated as a resolution. In this design, the input state
variables are 4) Weight Memory Space W
S E e, e
T
(17) Each location of T is linked to a particular adjustable value
2) Association Memory Space A in the weight memory space, can be represented as
In this space, a block consists of several elements. The w11 w1o w1 p
number of blocks, nB , is generally larger than two. The
operating principle of two-dimension CMAC is depicted in
w w 1 , , w o , , w p wk1 wko wkp (21)
Fig. 3, with nE 9 and 4 ( is the number of elements
in a full block), blocks A, B, and C divide the input state s1 , wn 1 wn o wn p
R R R
and blocks a, b, and c divide the input state s2 . New blocks
where w o w1o ,, wko ,, wn o n
T
will be obtained by shifting each variable an element. For R
R
and wko represents
th
example, blocks D, E, and F for s1 , and blocks d, e, and f the connecting weight value of the o output associated with
the kth receptive-field.
for s2 are obtained by such shifts.
The weight wko is initialized to zero and is automatically
Each block defines a receptive-field basis function, which
can be represented as rectangular [16] or triangular or updated during online operation.
continuously bounded function (e.g., Gaussian [25], [26] or 5) Output Space Y
B-spline [27], [28]). Here, Gaussian function is formulated as The output of CMAC is the algebraic sum of the activated
the receptive-field basis function weights in the weight memory, and is represented as
s m 2
ik si exp i 2 ik
nR
for k 1, , nB (18) yo w To S , m, v wko bk S , m k , v k
vik k 1
Variable s2 for o 1,, p (22)
i
l 9 The outputs of the CMAC can be represented as
y y1 ,, yo ,, y p w T S , m, v
f
T
c
8
(23)
State (3,3)
7
Ee
k
6
Bb C. Compensation Controller
h
e
5
Jj Assume that there exists an optimal CMAC to approach to
b
4
the ideal control law such that
3
T
u* uCMAC S , w* , m* , v* w* *
j *
2
(24)
g
d
a 1
Gg
Variable s 1 where is the minimum approximation
error; w * , m* , v* and * are the optimal parameters
1 2 3 4 5 6 7 8 9
Layer 4
Layer 3
A B C Layer 1
Layer 2
Layer 1 of w , m, v and , respectively.
D E F Layer 2
Rewrite u A as follows
u A uCMAC S , w , v uC w
G H I
u
Layer 3
,m T C (25)
J K L Layer 4
Fig. 3. CMAC in two-dimension with = 4 and nE = 9. where w , v and are the estimates of the optimal
,m
parameters of w , m, v and . By subtracting (24) from (25),
where ik si represents the kth block of the ith input si with the
define an approximation error as
mean mik and variance vik .
T uC
u~ u* u A w* * w
T
3) Receptive-Field Space T ~ T * w ~
w T uC (26)
Several blocks forms areas called receptive-fields. Each ~ ~
where w w w
*
and .
*
location of A is compatible with a receptive-field. Define the
multidimensional receptive-field function as Based on Taylor theorem, and using the partial linear of the
multidimensional receptive-field basis functions [29], [30],
n n s m 2 ~
bk S , m k , v k ik si exp i 2 ik , the expansion of becomes
i1 vik
i 1
4 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-3, Issue-7, July 2016
b T
1
b T
1
uC p sgn E T PGm (36)
b~1 m v p E T PGm (37)
~ b T b T
~ b2 2
v
2
m m m
*
v* v Ot Then the stability of the proposed controller is guaranteed.
~ Proof: Define a Lyapunov function as
bn bn
R
T
R
b T
n
R
V E, w ~,~
~, m v, ~p , t
m m m v v v 1 T 1 ~T ~ 1 ~T ~ 1 ~T ~ 1 ~ 2
E PE w w m m v v p
C m~ HT~
T
v Ot (27) 2 2 w 2 m 2 v 2
~ (38)
~
where m m m
*
v v* v and bk bk* bk .
, ~
where ~p p p represents the estimation error of the
bk* is the optimal parameter of bk ; bk is the estimated uncertainty bound. Taking the derivative of V and using (13)
parameter of bk* ; and (32), it is concluded that
Ot nR is a vector of higher-order terms; ~,~
~, m
V E , w v, ~p , t
b b2 bnR nnR nR
1 T 1
E PE E T PE
1 ~T
w w
1 ~T
m m
C 1 2 2 w m
m m m
1 ~T 1 ~
b b2 bnR nnR nR v v p p
H 1 v
v v v
b b
where k and k are defined as
1 T T
2
E P P E E T Gm Pu~ u S
1 ~T
w
w w
m v
1 ~ T 1 ~T 1 ~
bk
T bk bk m m v v p p
m 0, ,0 , , , , 0 , , 0 (28) m v
m mnk
n k n
k 1n
1k
1 T ~ T w ~ HT~
E QE E T Gm P w T CT m v uC
R
bk
T bk bk 2
v 0 , , 0 , , , , 0 , , 0 (29)
D uS
1 ~T 1 ~ T 1 ~T 1 ~
v v w w m m v v p p
k 1n 1k nk n k n R
w m v
Rewrite (27) as
(39)
* CT m ~ HT~ v Ot (30) From (15) and (33)(37), (39) can be rewritten as
Substituting (30) into (26), yields ~,~
~, m
V E , w v, ~p , t
u~ w~ T C T m ~ HT~ v Ot
E T QE E T PGm D uC u S ~p p
1 1
w T T~
C m H v O u T~
t C 2
~ T w
w ~ HT
T CT m v w
~ ~ C T T ~ HT~
m v 1 T
E QE E T PGm D E T PGmuC E T PGmu S
w Ot uC*T 2
~ w
wT ~ HT~
T CT m
v D uC (31)
1
p p p
where D w ~ T CT m ~ HT~
v w Ot represents the *T
1
E T QE E T PGm D p E T PGm
uncertain term and it is assumed to be bounded with a small 2
positive constant p (i.e., D p ). From (26) and (31), the
p p E T PGm
error dynamics (11) can be rewritten in space state as
E E Gm u~ uS
1
E T QE E T PGm p D 0 (40)
2
E G w ~ T w
m
~ HT~
T CT m
v u D u C S Since V E , w
~, m~,~v, ~p , t 0 is a negative semi-definite
(32)
function, E, w ~, m v and ~p are all bounded. Consider the
~, ~
Theorem: Consider the nonlinear missile autopilot problem
presented in (6). The adaptive CMS control system is function t 1 2 ET QE V t and by
designed as (10) where the supervisory controller is described integrating t with respect to time
in (15) and the adaptive CMAC is formulated in (25). Here, in
d V E, w, m, v, ,0 V E, w, m, v, , t
t
the adaptive CMAC, the adaptive laws are chosen as ~ ~ ~ ~ ~ ~ ~ ~ (41)
p p
(33)(35) and the compensation controller as (36) with the o
estimation law given by (37), where w , m , v and are Since V E, w ~,~
~, m
v, ~p ,0 is
bounded,
strictly positive constants.
E T PG
~ ~
~ ~
and V E , w, m, v, p , t is non-increasing function and
w w (33) m bounded, it can concluded that:
m E T PG Cw (34) t
lim d
m m
(42)
v v E PGm Hw
T
(35) t
o
5 www.ijeas.org
Missile Autopilot Design using Adaptive CMAC Supervisory Controller
Then, t is bounded, so based on Barbalats Lemma [31], The adaptive CMS controller designed here needs to have
it can be noticed that lim t 0 . So, lim E t 0 . the bounds f U , g U and g L . In this system,
t t
Consequently, the stability of the proposed controller is f U 104.3 x3 and g U g L 1 are chosen.
ensured. Fig. 5 shows the simulation results under which the CMAC
controller is designed alone, and Fig. 5(a) illustrates the AOA
IV. NUMERICAL SIMULATIONS response and reference response for a step command. Also,
Fig. 5(b) - (d) depict shows the associated control input, pitch
To find out the efficiency of the proposed adaptive CMS
rate and control fin deflection angle respectively. Although a
controller for the missile pitch autopilot, simulations were
good response is obtained, the chattering phenomena of the
carried out. The missile model used in simulation is a generic
control efforts caused by the switching operation lead to the
short-range surface-to-air missile and are its aerodynamic
reduction of tracking accuracy.
coefficients given in Table I. First-order actuator model
a 150 rad/s is considered. Fig. 6 presents the simulation results of adaptive CMS
controller, and Fig. 6(a) illustrates the AOA response and
The control objectives are as follows: reference response for a step command. Also, Fig. 6(b)(e)
Maintain stability over the operating range specified by shows the associated control input, pitch rate, control fin
t , M m t such that deflection angle and the supervisory control, respectively.
10 t 10 and 1.6 M m t 2.6 Note that Fig. 6(e) shows one activation period [0, 0.0022]
Track step command in C , with time constant 0.2 sec, sec. After 0.0022 sec, the supervisory is deactivated.
The comparison between three controllers is summarized in
less than 10 % overshoot and steady-state error no greater Table II, which shows that the adaptive CMS controller
than 2 %. achieves the design requirement.
Table I: Details of pitch axis missile model
h4 13.53 rad -2 C2 50
V. CONCLUSION
In all simulations, a 20% uncertainties existing in all random In this paper, an Adaptive CMAC Supervisory controller,
constant perturbation pi i 1,,5 is taking into including an adaptive CMAC and a supervisory controller,
consideration. has been proposed to design a missile pitch autopilot for a
nonlinear model which is aerodynamically controlled and
A. Feedback Linearization Controller contains unknown parameters and aerodynamic uncertainties.
For the purpose of comparison, the feedback linearization From the simulation results, the proposed control system
control law presented in (8) was simulated. The controller achieves successfully the control objectives required. Future
gains are chosen as k1 50 and k2 225 . study will be applying this controller to other missile systems
Using the feedback linearization controller, the simulation in order to further check its performance.
results for a step command is depicted in Fig. 4. Fig. 4 (a)
illustrates the AOA response and reference response for a step REFERENCES
command. Also, Fig. 4 (b) - (d) depict the associated control
[1] M. Xin and S. N. Balakrishnan, Nonlinear H missile longitudinal
effort, pitch rate and control fin deflection angle, respectively.
autopilot design with -D method, IEEE Transactions on Aerospace
B. Adaptive CMS Controller and Electronic Systems , Vols. 44, no. 1, pp. 41-56, 2008.
[2] S. Pain and K. Tiwari, A graphical design methodology of missile
The adaptive CMS controller has been depicted in Fig. 1, autopilot, IEEE International Multi-Conference on Automation,
which uses Gaussian function as receptive field basis Computing, Communication, Control and Compressed Sensing,
functions. The input space was partitioned in a grid of 2013.
[3] C. Xianxiang; S. Jianmei and C. G. Key, Robust gain-scheduled
size 1 , and receptive fields are selected to cover the input autopilot design with LPV reference model for portable missile,
2
space 2,2, 2,2 along with each of the input dimension.
IEEE Aerospace Conference, pp. 1-10, 2012.
[4] F. K. Yeh; J. J. Huang and C. C. Wang, Observer-Based Adaptive Fuzzy
Therefore, the parameters are chosen as Sliding-Mode Controller Design for Missile Autopilot Systems,
IEEE SICE Annual Conference (SICE), pp. 546 551, 2012.
vik 2 2 and m 2.5,1.5,0.5,0.5,1.5,2.5 for all i and k. [5] C. Xun and L. Yongshan Design and Analysis of Autopilot Based on
Adaptive Control, IEEE Mechatronic Sciences, Electric
The design parameters are set as follows: Engineering and Computer (MEC), pp. 2918 2921, 2013.
400 15 [6] S.H. Kim, Y.S. Kim and C. Song, A robust adaptive nonlinear control
Q , w 1.0 , m v 0.75 , 0.01, V 1 approach to missile autopilot design, Control Engineering Practice,
15 1 Vol. 12, pp. 149-154, 2004.
6 www.ijeas.org
International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-3, Issue-7, July 2016
[7] F. Jun-fang, S. Zhong, and C. Zhen-xuan, Missile Autopilot Design and [19] F. J. Gonzalez-Serrano, A. R. Figueiras-Vidal, and A. Artes-Rodriguez,
Analysis Based on Backstepping, IEEE Systems and Control in Generalizing CMAC architecture and training, IEEE Trans. Neural
Aeronautics and Astronautics, pp. 1042 - 1046, 2010. Networks, vol. 9, pp. 15091514, 1998.
[8] A. A. Godbole, and S. E. Talole, Robust Feedback Linearization [20] J. C. Jan and S. L. Hung, High-order MS_CMAC neural network,
Approach to Pitch Autopilot Design, Advanced Materials Research, IEEE Trans. Neural Networks, vol. 12, pp. 598603, March 2001.
Vols. 403-408, pp. 4667-4673, 2012. [21] F. J. Lin, R. J. Wai and Ch. M. Hong, Hybrid supervisory control
[9] F. L. Lewis, A. Yesildirek, and K. Liu, Multilayer neural-net robust using recurrent fuzzy neural network for tracking periodic inputs,
controller with guaranteed tracking performance, IEEE Trans. IEEE Transactions on Neural Networks, vol. 12, no. 1, pp 68-90,
Neural Netw., vol. 7, no. 2, pp. 388399, Mar. 1996. 2001.
[10] X. X. Liu, Y. Wu, J. P. Shi, and W. G. Zhang, Adaptive fault-tolerant [22] A. Jayasiri, G. K. I. Mann, and R. G. Gosine, Modular Supervisory
flight control system design using neural networks, IEEE Control and Hierarchical Supervisory Control of Fuzzy
International Conference on Industrial Technology, 2008. Discrete-Event Systems, IEEE Transactions on Automation Science
[11] K. C. Qian, and Z. L. Chen, Dynamic inversion based on neural and Engineering, vol. 9, no. 2, pp 353-364, 2012.
network applied to nonlinear flight control system, IEEE [23] C. M. Lin and Y. F. Peng, Adaptive CMAC-Based Supervisory
International Conference on Future Computer and Communication, Control for Uncertain Nonlinear Systems, IEEE Transactions on
Vol. 1, pp 699-703, 2010. Systems, Man, and Cybernetics, vol. 34, no. 2, pp 1248-1260, 2004.
[12] Q. Lin, Z. H. Cai, Y. X. Wang, J. P. Yang, and L. F. Chen, Adaptive [24] L. X. Wang, Adaptive Fuzzy Systems and Control: Design and
Flight Control Design for Quadrotor UAV Based on Dynamic Stability Analysis. Englewood Cliffs, NJ: Prentice-Hall, 1994.
Inversion and Neural Networks, International Conference on [25] Y. H. Kim and F. L. Lewis, Optimal design of CMAC neural-network
Instrumentation, Measurement, Computer, Communication and controller for robot manipulators, IEEE Trans. Syst., Man, Cybern.
Control, 2013. C, vol. 30, pp. 2231, Feb. 2000.
[13] C. H. Wang and K. N. Hung, Intelligent Adaptive Law for Missile [26] C. T. Chiang and C. S. Lin, CMAC with general basis functions,
Guidance Using Fuzzy Neural Networks, International Journal of Neural Networks, vol. 9, pp. 11991211, 1996.
Fuzzy Systems, vol. 15, no. 2, pp. 182191, 2013. [27] S. H. Lane, D. A. Handelman, and J. J. Gelfand, Theory and
[14] C. K. Lin, Adaptive Critic Autopilot Design of Bank-To-Turn development of higher-order CMAC neural networks, IEEE Contr.
Missiles using Fuzzy Basis Function Networks, IEEE Transactions Syst. Mag., vol. 12, pp. 2330, 1992.
on Systems, Man, and Cybernetics, vol. 35, no. 2, pp 197-207, 2005. [28] S. Jagannathan, Discrete-time CMAC NN control of Feedback
[15] J. J Ma, P. Li, L. N. Geng and Z. Q. Zheng, Adaptive neural dynamic Linearizable nonlinear systems under a persistence of excitation,
surface control for a missile with input and output constraints, IEEE IEEE Trans. Neural Networks, vol. 10, pp. 128137, Oct. 1999.
Chinese Control Conference, 2014. [29] Y. G. Leu, T. T. Lee, and W. Y. Wang, Observer-based adaptive
[16] J. S. Albus, A new approach to manipulator control: The cerebellar fuzzy-neural control for unknown nonlinear dynamical systems,
model articulation controller (CMAC), J. Dyn. Syst., Measurement, IEEE Trans. Syst., Man, Cybern. B, vol. 29, pp. 583591, 1999.
Contr., vol. 97, pp. 220227, 1975. [30] Y. G. Leu, W. Y. Wang, and T. T. Lee, Robust adaptive fuzzy-neural
[17] S. H. Lane, D. A. Handelman, and J. J. Gelfand, Theory and controllers for uncertain nonlinear systems, IEEE Trans. Robot.
development of higher-order CMAC neural networks, IEEE Control Automat., vol. 15, pp. 805817, 1999.
Syst. Mag., vol. 12, no. 2, pp. 2330, Apr. 1992. [31] J. J. E. Slotine and W. Li, Applied Nonlinear Control. Englewood
[18] W. Yu, and X. Li, "Some New Results on System Identification with Cliffs, NJ: Prentice-Hall, 1991.
Dynamic Neural Networks," IEEE Trans. Neural Networks. vol. 12,
no. 2, pp. 412-417, 2001.
7 www.ijeas.org
Missile Autopilot Design using Adaptive CMAC Supervisory Controller
8 www.ijeas.org