2008 5th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE 2008)

Output Linear Feedback for a Class of Nonlinear Systems based on the Invariant Ellipsoid Method
S. Gonzalez-Garcia1 , A. Polyakov1,2 , A. Poznyak1 Department of Automatic Control. CINVESTAV Av. Instituto Polit ecnico Nacional 2508. 07300 M exico D.F., M exico 2 Post-doc position from Department of Applied Mathematics, Voronezh State University, Universitetskaja pl.1, Voronezh 394693, Russia. E-mail: sgonzalez{apolyakov,apoznyak}@ctrl.cinvestav.mx
Abstract The robust output stabilization problem for a class of uncertain nonlinear systems with disturbances is considered. The suggested control scheme as well as the observation algorithm are based on the invariant ellipsoid method, which allows to obtain the robust linear feedback as a solution of the special linear optimization problem with bilinear constraints. The methods for solving this optimization problem involves the LMI technique. The stabilization of a single link exible joint manipulator is considered as an illustrative example.

Keywords: state estimation, uncertain systems, linear matrix inequalities. I. INTRODUCTION The problem of persistent disturbance attenuation consists in the minimization of the perturbation eect in system through an appropriate feedback control. The synthesis problem for a linear system with unknown but bounded perturbations traces back to the work of Bertsekas and Schweppe [1] and [2] where an admissible control tube was found such that it maintained the state in a specied region and reached a given target set. This was done using a dynamic programming scheme and convex set operations. A class of state feedback controls in order to guarantee uniform ultimate boundedness (UUB) for uncertain dynamic systems were presented in [3] and summarized in [8] where additional results were presented for the analysis of perturbed systems subjected to nonvanishing perturbations. For the case of L2 perturbations, the explicit solutions for the optimal controller with the H2 theory were obtained as it was shown in [4] and [5]. The restriction of bounded perturbations was changed in [6] where only the maximum amplitude of the perturbations (l1 norm) was considered, although the optimal controller designed with the l1 scheme can be of high order. An interesting methodology for the control of nonlinear systems was presented in [7] that combines feedback linearization and linear compensation based on the stable factorization approach to guarantee robustness against parameter uncertainty and external disturbances.

In [9] the problem of peak to peak gain minimization with Linear Matrix Inequalities (LMI) was considered. The notion of invariant sets and its close connection with the Lyapunov theory was also exploited for the analysis and control of dynamical systems such as constrained control, robustness analysis, and disturbance rejection as was stated in [11]. The concept of invariance set states that any element of this set for t = 0, will remain in the invariance set for all t R. The most commonly used forms of invariant sets are ellipsoids and polytopes. The former are simple to characterize the invariant set (a center and the matrix are only required) but it is rather conservative. The latter are the most natural and exact form to dene constraints although with a more complex representation. In [12] the problem of synthesis of a static state-feedback controller for an LTI system that minimizes the size of the corresponding invariant ellipsoid was reduced to the optimization of a linear function under some LMI constraints. Although many robust control problems can be formulated in terms of linear matrix inequalities (LMI) and be solved with semidenite programming [10], a signicantly wider class of problems can be formulated in terms of Bilinear Matrix Inequalities (BMI) as in [13]. As for their numerical solution, it can be mentioned that only in very few cases (such as static state feedback and dynamic output feedback) it is possible to convexify the problem with the appropriate change of variables and derive equivalent LMIs. This paper is concerned with the problem of stabilization of a nonlinear system in the presence of bounded perturbations and output measurements. The stabilization scheme is based on the Lyapunov function approach and the invariant ellipsoid method to construct a Luenberger observer and the observed state feedback controller. Due to the nonlinearities of the system, the constraints of the optimization problem are bilinear. However, with the use of some appropriate transformations, the obtained BMI constraints are shown to be dependent only on two scalar parameters, i.e., for xed scalar parameters, the BMI constraints becomes LMI and the optimization problem can be solved through standard semidenite programming

IEEE Catalog Number: CFP08827-CDR ISBN: 978-1-4244-2499-3 Library of Congress: 2008903800 978-1-4244-2499-3/08/$25.00 2008 IEEE

2008 5th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE 2008)

technique. The outline of this paper is as follows. In section 2, the problem statement and basic assumptions are introduced. Then, in section 3 the main results and its proofs are given; some numerical aspects of the robust feedback synthesis are given in section 4. Next, section 5 presents the application of the obtained result to the stabilization of a single link exible joint robotic manipulator. Finally, some remarks conclude this paper. II. PROBLEM STATEMENT Consider the nonlinear dynamic system with linear state-output mapping given by x = f (x, t) + Bu y = Cx + wy (1)

example, [12]). Here we just present some basic ideas of this method. The ellipsoid (0, P ) = x Rn : xT P 1 x 1 , P > 0 with center in the origin and shape matrix P is said to be state-invariant for the system (3),(4) under the disturbances (2) and nonlinearities (5) if 1) the condition x(0) (0, P ) implies x(t) (0, P ) for all t 0; 2) if x(0) / (0, P ) then x(t) (0, P ) for t . In other words, any trajectory of the system starting in the invariant ellipsoid stays in it for all t 0, but the trajectory starting outside of the ellipsoid converges to this ellipsoid (asymptotically or in nite time). This is done with the aid of the second method of Lyapunov. This invariant ellipsoid can be considered as a characteristic of the inuence of the uncertainties in a system. So, the minimum (in some sense) invariant ellipsoid corresponds to a robust feedback control. Hence, the main problem is to design the observer-based linear feedback control providing the convergence of any trajectory of the system (3),(4),(7) to the minimum invariant ellipsoid. Here we will use the trace as the criterion of the ellipsoids minimality. III. MAIN RESULTS Dene the state estimation error as e := x x . Then its time derivative satises e = (A F C )e + wx F wy e )T where z Introduce the extended vector z := ( x R2n . Then it follows +F w z = Az := where A and w := A + BK 0 FC A FC := , F 0 I F F (8)

where x Rn is the state vector, u Rm is the control input, y Rk is the system output, wy Rk is the output perturbation, f : Rn R Rn is an unknown (from a class given below) nonlinear function, B Rnm , C Rkn are the system matrices. The following assumptions will be in force throughout: the output disturbances wy are inside of a bounded ellipsoid, that is, wy
2 K T = wy K wy 1

(2)

where K > 0 is given; the nonlinear function f (x, t) is quasi-Lipschitz, namely, it belongs to the class of functions satisfying f (x, t) Ax
2 Kf

+ x

2 Kx

where 0 < Kf Rnn , 0 < Kx Rnn , A Rnn are known matrices, 0; without loss of generality, after the normalization of this inequality it is sucient to consider only two-valued case: = (0; 1). The pair (A, B ) is controllable and the pair (A, C ) is observable. Using denotation wx := f (x, t) Ax the system (1) can be rewritten in the form x = Ax + Bu + wx y = Cx + wy with wx
2 Kf

(3) (4)

+ x

2 Kx

(5)

Here we only consider the linear feedback controls u = Kx , K Rmn (6)

wx . wy Our aim here is to nd the control gain matrix K and the observer gain matrix F providing a good enough stabilization as well as state estimation of the system (8), or more exactly, to design K and F such that the corresponding invariant ellipsoid, called below quasi-minimal, would contain the minimal one. Theorem 1. If the following optimization problem tr(X1 ) + tr(H ) min subject to R1 C T Y2 0 Y2 X1 X 2 Y2T C X2 -Y2 I Y2T -Y2T 0 -3 K 0 X2 X1 I 0 0 1 K - 1 x 2 0 (10) (9)

with respect to observed state x Rn which are obtained by the classical Luenberger observer having the structure x = Ax + Bu + F (y C x ),

F Rnp

(7)

The robust stabilization of the system (3),(4) will be realized using the method of Invariant Ellipsoids (see, for

0 X2 -2 Kf 0 0

IEEE Catalog Number: CFP08827-CDR ISBN: 978-1-4244-2499-3 Library of Congress: 2008903800 978-1-4244-2499-3/08/$25.00 2008 IEEE

2008 5th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE 2008)

-R1 -2X2 I H I with I X2

I X1 AT +AX1 +Y1 B T +BY1T +1 X1 0,

Representing (16) in the form wx Kf wx + z T Kz z, one has Kz = Kx Kx Kx Kx

1 2 + 3 , 2 0, 3 0

X1 0, X2 0, H 0 := AT X2 +X2 A-Y2T C -C T Y2 +1 X2 to the matrix variables Rnm , Y2 Rkn and 3 , then the ellipsoid with 0 1 X2 (11)

has a solution with respect H, X1 , X2 , R1 , R2 Rnn , Y1 the scalar variables 1 , 2 and the matrix X1 P = 0

T z z wx A2 wx := 2 , wy wy Kz 0 0 Kf 0 A2 := 0 0 0 0 T z z 1 wx A3 wx 1 := 3 w w y y 0 0 0 A3 = 0 0 0 0 0 K

And analogously, wy
2 K

is quasi-minimal invariant ellipsoid of the system (8) with (2), (5) with the feedback control gain matrix
1 Y1 K = X1 T

(12)

and the observer gain matrix

1 F = Y2 X2 T

(13) P1 0 0 P2

Proof. Dene the Lyapunov function as V (z ) : =(z, P 1 z ), P 1 =

Finally, by the S -procedure if there exist 1 0 , 2 0 and 3 0 such that 0 1 1 + 2 2 + 3 3 (17) A0 1 A1 + 2 A2 + 3 A3 then the conditions (14), (15) and (16) hold and the ellipsoid with the matrix P is an invariant ellipsoid of our system. The rst condition in (17) can be represented as 1 2 + 3 , i 0, i = 1, 2, 3 For the second one, dene Q := A0 1 A1 2 A2 3 A3 0 or Q= P1 F C +2 Kx 2 P2 -F T P2 P1 F -P2 F 0 -3 K (18)

where P > 0 is the matrix of an invariant ellipsoid should be minimized. Then 1 = (z, P 1 z V )+(z, P z) 1 w))+((Az +F w), P 1 z ) = (z, P (Az +F P = z [A
T T 1

+P

]z +w F P A

z +z P

w F

z wx 0 wy (14) This ellipsoid (z, P 1 z ) 1 will be invariant if and only 0, that is, for z satisfying if outside of it we have V z P
T 1

or expressing in its quadratic form T z T P 1 +P 1 A P 1 F A = wx V T 1 P F 0 wy

1 C T F T P1 +2 Kx 0 F T P1 with 1

0 P2 -2 Kf 0

z1 wy
2 K

(15) 1

Together with (14) and (15) we also have wx Kf wx + xT Kx x and (16)

To fullll all this constraints let us apply the, so-called, S -procedure (see, for example, [17]). Dene T P 1 + P 1 A P 1 F A , 0 := 0 A0 := T 1 P F 0 Since z z T P 1 z 1 wx A1 w y P 1 0 0 0 A1 := 0 0 T z wx 1 := 1 wy 0 0 0

1 I : =AT K P1 +P1 AK +2 Kx , AK = A + BK + 2 1 2 : =AT I F P2 +P2 AF +2 Kx , AF = A F C + 2 Applying the quadratic non-singular transformation 1 0 0 0 P1 0 I 0 0 T1 = 0 0 I 0 0 0 0 I to the matrix Q we get 1 C F + 1 2 K x P1 0 FT
T T T = Q1 = T1 QT1 F C+ 0 1 2 P1 Kx

-P2 F 0 0 -3 K F

2
-F T P2

P2
-2 Kf

P2

IEEE Catalog Number: CFP08827-CDR ISBN: 978-1-4244-2499-3 Library of Congress: 2008903800 978-1-4244-2499-3/08/$25.00 2008 IEEE

2008 5th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE 2008)

1 T 1 1 1 1 1 := P1 AK + AK P1 + 1 P1 + 2 P1 K x P1 T 2 := AF P2 + P2 AF + 1 P2 + 2 Kx

implies
1 1 P2 (P 1 AT K + AK P1 + 1 P1 )P 2 1 T 1 1 1 -P2 -P2 -(P 1 AK +AK P1 +1 P1 ) 1

Obviously that 1 P1 I + Q1 = Q 0 0 where = Q

(2 Kx )
1 P1

(22)

0 0

0 (19)

Applying then the Schur complement to the matrix inequality

1 T 1 1 1 AK + AK P1 + 1 P1 ) R1 2P2 (P1

1 CT F T 0 FT

FC 2 P2 F T P2

0 P2 2 Kf 0

F P2 F 0 3 K

(23)

we get R1 2P2 I Dening

1 T 1 K , X2 := P2 , Y2 := F T P2 , Y1 := P1 X1 := P1

1 T 1 1 AK + AK P1 + 1 P1 1 := P1

I 1 T 1 1 P1 AK + AK P1 + 1 P1

0 (24)

2 := AT F P2 + P2 AF + 1 P2 Q2 = 0 F P2 -P2 F -2 Kf 0 0 -3 K 0 0

Using the Schur complement to (19) we obtain 1 CT F T 0 FT 1 P1 FC 2 P2 F T P2 I

1 P1 I 0 0 1 - 1 K x 2

using (22), (23), (24) and the matrix inequality X Y YT Z X Y YT Z X , then 0

1 T 1 1 AK +AK P1 +1 P1 1 := P1 T 2 := AF P2 +P2 AF +1 P2

Analogously, apply the transformation P2 0 0 0 0 0 I 0 0 0 T2 = 0 0 I 0 0 0 0 0 I 0 0 0 0 0 I to Q2 we obtain Q3 = P2 F C 0 2 P2 -2 Kf P2 F T P2 0 I 0

T T2 Q2 T2

valid for any X = X T , Z = Z T and X = X T (20) can be restricted by Y2T C 0 Y2T X2 X1 R1 T C T Y2 X Y I 1 2 2 K 0 0 0 X 2 2 f Y2 Y2 0 -3 K 0 1 X1 X 2 I 0 0 - 1 K x 2

1 := AT X2 +X2 A-Y2T C -C T Y2 +1 X2 -R1 -2X2 I I X1 AT +AX1 +Y1 B T +BY1T +1 X1 0

1 C T F T P2 0 F T P2 1 P1 P2 1 2 : :

= P2 F -P2 F 0 -3 K 0

1 P 2 P1 I 0 0 1 - 1 K x 2

0 (20)

The minimization problem to be solved is

1 ) min tr(X1 ) + tr(X2

(25)

subject to the constraints (10). Introduce the following additional constraint

1 H X2

1 T 1 1 = P2 (P1 AK +AK P1 +1 P1 )P2

= AT F P2 +P2 AF +1 P2
T T T

H I

I X2

By -inequality (see [17]) XY + Y X X X + Y

nk nk 1

(21)

This allows to reduce the optimization problem (25) to a linear one tr(X1 ) + tr(H ) min Remark 1. Since the proof of the theorem 1 is based on the S -procedure, Schur complement and -inequality [17], which give only the upper estimates of the matrix inequalities, we can guarantee that the obtained solution gives only the quasi-minimal invariant ellipsoid.

, Y R and any 0 < = valid for any X R T Rkk being applied for X = P2 and Y = I it follows X + X T X X T + 1 that for
1 T 1 1 AK + AK P1 + 1 P1 ) := (P1

IEEE Catalog Number: CFP08827-CDR ISBN: 978-1-4244-2499-3 Library of Congress: 2008903800 978-1-4244-2499-3/08/$25.00 2008 IEEE

10

2008 5th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE 2008)

IV. ROBUST FEEDBACK SYNTHESIS: NUMERICAL ASPECTS The problem (9) is, in fact, a bilinear optimization problem because of the term X1 X2 participating in the matrix constraint (10). The numerical solution of such problems is usually a non trivial task. The feasible set for a BMI is nonconvex in general case. Semi-denite relaxations (as in [14] [15]) and the solution through nonlinear programming methods (e.g. branch-bound algorithm) can be considered as two alternatives to solve a given BMI. The eciency of the Matlab-Tools PENBMI application is highly sensitive to the initial point selection which is desired to be close to a solution [16]. It is based on the penalty function approach. The following lemma simplies the problem of a feasible starting point nding. Lemma 1. The set of variables satisfying (10) contains the set of ones satisfying Y2T C 0 Y2T 0 R1 C T Y2 X2 -Y2T I 0 -2 Kf 0 0 X2 (26) 0 Y2 Y2 0 -3 K 0 0 I 0 0 R2 -R1 -2X2 I I X1 A +AX1 +Y1 B T +BY1T +1 X1 +
T

The term 3 in (27) can be bounded by R2 as

1 1 1 P1 P1

1 1 K R2 2 x

(28)

Applying again the Schur complement we can express (28) as 1 1 Kx P1 R2 1 2 0 1 P1 Remark 2. Notice that for the xed scalar parameters 1 and 2 the matrix inequalities (26) become LMIs. They can be solved using packages such as SeDuMi Toolbox, YALMIP Toolbox and the standard MATLAB LMItoolbox. Remark 3. The main optimization problem (9) can also be solved using the following two-steps procedure: 1) rst, we x the scalar parameters 1 , 2 and solve the problem (9) with respect to the matrix variables which satisfy LMI- constraints (26); 2) second, we solve the optimization problem (9) only with respect to scalar parameters 1 , 2 (under the found matrix variables). V. NUMERICAL EXAMPLE Consider the nonlinear model of a single link exible joint manipulator. The equations of motion for this system taken from [18] are: Iq 1 + mgl sin(q1 ) + K (q1 q2 ) = 0 Jq 2 + Bq2 K (q1 q2 ) = u where q1 and q2 are the link and motor angles, I is the link inertia, J is the inertia of the motor, B is the actuator damping, K is the spring stiness, m and l are the mass and length of the link respectively and u is the control 1 , input. Dene the state variables x1 = q1 , x2 = q 2 to obtain the following state space x3 = q2 and x4 = q representation: 0 1 0 0 0 0 -0.2 0 0.2 0 x+ 0 u+ mgl sin (q 1 ) x = 0 0 0 0 1 0 9 9 0 -9 0 0 with the following state-output mapping x1 x2 y= 1 0 0 0 x3 + wy x4 where wy R2 is an output noise which can be estimated as follows T K wy 1, K = 500 wy Assume also that system nonlinearities are bounded as

1 - 1 Kx -R2 2

X1

X1 -

0,

H I

I X2

1 2 + 3 , i 0, i = 1, 2, 3 with := AT X2 +X2 A-Y2T C -C T Y2 +1 X2 0 0 0 0 X1

and Y :=

Proof. By the -inequality (21) with X T := X2 0 0 0 0 the matrix inequality (20) containing Q3 can be estimated as Q3 Q3 0 where 1 C T F T P2 0 F T P2 0 P2 F C 2 P2 F T P2 I Q3 := 0 P2 2 Kf 0 0 P2 F P2 F 0 3 K 0 0 I 0 0 3 0 (27)

with: : =AT F P2 + P2 AF + 1 P2 1 1 1 1 1 3 : = Kx + P1 P1 2 that, in view of (24), -R1 -2X 2 I 1 2

1 T 1 1 : =P2 (P1 AK + AK P1 + 1 P1 + )P2

I T X1 A +AX 1 +Y1 B T +BY T 1 +1 X1 +

f (x, t) Ax

x1

2 1.5

IEEE Catalog Number: CFP08827-CDR ISBN: 978-1-4244-2499-3 Library of Congress: 2008903800 978-1-4244-2499-3/08/$25.00 2008 IEEE

11

2008 5th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE 2008)

0.3 0.25 0.2 0.15 0.1 Real Observed

0.05 0 0.05 0.1 0.15 0.2 0 10 20 30 Time 40 50 60

the system. Due to nonlinearities and perturbations the constraints involved into this optimization problem turns out to be bilinear matrix inequalities. Using appropriate transformations and xing the scalar parameters (1 and 2 ), the bilinear constraints become LMI. The robustness property of this scheme is conrmed by the application to the single link exible joint robotic manipulator. References
[1] D.P. Bertsekas and I.B. Rhodes, On the minmax reachability of target set and target tubes, Automatica, vol. 7, pp. 233-247, 1971. [2] D. Glover and F. Schweppe, Control of linear dynamic systems with set constrained disturbances. IEEE Trans. Automat Control, vol. 16, no. 5, pp. 411-423, 1971. [3] M.J. Corless and G. Leitmann, Continuous state-feedback guaranteeing uniform ultimate boundedness for uncertain dynamic systems, IEEE Trans. Automat. Contr., vol. 26, no. 5, pp. 1139-1144, 1981. [4] J.C. Doyle, Synthesis of robust controllers and lters, Proceedings of the 22th IEEE Conference on Decision and Control, pp. 109-114, 1983. [5] B.A. Francis, J.W. Helton and G. Zames, H -optimal feedback controllers for linear multivariable systems, IEEE Trans. Automat. Contr., vol. 29, pp. 888-900, 1984. [6] M.A. Dahleh and J.B. Pearson, l1 optimal feedback controllers for MIMO discrete-time systems, IEEE Trans. Automat. Contr., vol. 32, pp. 314-322, 1987. [7] M.W. Spong and M. Vidyasagar, Robust Linear Compensator Design for Nonlinear Robotic Control, IEEE Journal of Robotics and Automation, vol RA-3, no. 4, 1987. [8] H.K. Khalil, Nonlinear Systems, Prentice-Hall, Englewood Clis, NJ, 1992. [9] J. Abedor, K. Nagpal and K. Poola, A Linear Matrix Inequality Approach to Peak-to-Peak Gain Minimization, Int. J. Robust Nonlinear Control, vol. 6, pp. 899-927, 1996. [10] S. Boyd, E.G. Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix Inequalities in System and Control Theory, SIAM, Philadelphia, 1994. [11] F. Blanchini, Set invariance in control-a survey, Automatica, vol. 35, no. 11, pp.1747-1767, 1999. [12] B.T. Polyak and M.V. Topunov, Suppression of bounded exogenous disturbances: output feedback, Automation and Remote Control, vol. 69, no. 5, pp. 801-818, 2008. [13] M.G. Safonov, K.C. Goh and J.H. Ly, Control system synthesis via bilinear matrix inequalities, in Proceedings of the 1994 American Control Conference, Baltimore, MD, USA. [14] J. Loefberg, Minimax Approaches to Robust Model Predictive Control, Ph.D. Thesis, Linkoeping University, 2003. [15] S. Boyd and L. Vandenberghe, Semidenite programming relaxations of non-convex problems in control and combinatorial optimization, In: Communications, Computation, Control and Signal Processing: A Tribute to Thomas Kailath, Kluwer, 1997. [16] D. Henrion, J. Loefberg, M. Kocvara and M. Stingl, Solving polynomial static output feedback problems with PENBMI, In Proceedings of the 44th IEEE Conference on Decision and Control and the European Control Conference, pp. 7581-7586, Seville, Spain, 2005. [17] A.S. Poznyak, Advanced mathematical tools for automatic control engineers: Deterministic techniques, vol. 1, Elsevier, New York, NY, 2008. [18] M.W. Spong and M. Vidyasagar, Robot Dynamics and Control, John Wiley, New York, 1989.

q1

Fig. 1: Angular position of the link.

One can rearrange them to the form (5) with 1e-20 0 0 0 0 1 0 0 , =0 Kf = 0 0 1e-20 0 0 0 0 1e-20 1.5 0 0 0 0 1e 20 0 0 Kx = 0 0 1e 20 0 0 0 0 1e-20 A linear feedback is designed according to the main results of the previous section. The obtained parameters K and F realizing the robust output linear controller are as follows K F = = 28.27 0.701 946.11 0.173 616.23 0.764 488.42
T

0.539

The obtained performance index (9) is 0.0287. The gure 1 show the angular position of the controlled single link manipulator. As it can be seen from the above numerical simulation the closed loop system has a good stabilization performance, but the convergence processes in general are slow. This situation seems to be usual for robust linear controllers being applied to nonlinear models containing external noise or disturbances. Remark 4. For illustrative purposes the system parameters of the single link exible joint manipulator were assumed to be known. If they were unknown but bounded, the parameters Kf and Kx must be changed according to the corresponding uncertainties. VI. CONCLUSIONS Here the stabilization scheme for a class of nonlinear systems under the presence of uncertainties and disturbances is presented. As a principle part it contains the linear output controller using the current state estimates and designed based on the invariant ellipsoid technique. It minimizes the eect of the perturbations and nonlinearities in

IEEE Catalog Number: CFP08827-CDR ISBN: 978-1-4244-2499-3 Library of Congress: 2008903800 978-1-4244-2499-3/08/$25.00 2008 IEEE

12