Eigenstructure Assignment for Design of Multimode Flight Control Systems

Kenneth M. Sobel and EliezerY. Shapiro

ABSTRACT: Advancedaircraftsuchas control configured vehicles (CCV) provide the capability for implementing multimode control laws, which allow the aircraft performance to be tailored to match the characteristics of a specific task or mission. This is accomplished by generating decoupled aircraft motions, which can be used to improve aircraft effectiveness. In this article, we describe a task-tailored multimode flight control system, which was designed by using eigenstructure assignment. motions include yaw pointing, lateral translation, and direct sideforce. Yaw pointing is characterized by heading command without a change in lateral directional flight path angle. Lateral translation is characterized by lateral directional flight path command without a change in heading. Direct sideforce is characterized by lateral acceleration command without a change in sideslip angle. All three lateral modes also require that there be no change in bank angle. The application of eigenstructure assignment to conventional flight control design hasbeen described by Shapiroet al. in Ref. [l]. A design methodology that uses eigenstructure assignment to obtain decoupled aircraft motions has been described by Sobel et al. in Refs. [2]-[5].In this article, we use eigenstructure assignment to design a task-tailored multimode flight control system. The longitudinal design is illustrated by using the unstable dynamics of an advanced fighter aircraft, and the lateral design is illustrated by using the dynamics of the flight propulsion control coupling (FPCC) vehicle. eigenvectors of A - BFCareclose to the respective members of the set (v?). The feedback gain matrix F will exactly assign r eigenvalues. It will also assign the correspondingeigenvectors, provided that they were chosen to be inthe subspace spanned by the columns of ( A i l - A ) - B . This subspace is of dimension m , which is the number of independent control variables. In general, a chosen or desired eigenvector v f will not reside in the prescribed subspace and, hence, cannot be achieved.Instead, a bestpossible choicefor an achievable eigenvector is made. This best possible eigenvector is the projection of I,? onto the subspacespannedbythecolumnsof (Ail - A)-B. We summarize with the following: The matrix F will exactly assign r eigenvalues. It will also exactly assign each of thecorresponding r eigenvectorsto m -dimensional subspaces, which are constrained by A:, A , and E . If more than 172 elements are specified for a particular eigenvector, then an achievable eigenvector is computed by projecting the desired eigenvector onto the allowable subspace. This is the subspace spanned by the columns of ( A i l - A ) - B . If control over a larger number of eigenvalues is required, then additional independent sensors must be added.

Introduction
Advanced aircraft such as controlconfigured vehicles (CCV) provide the capability to control the aircraft in unconventional ways. One such approach is to generate decoupled motions, which can be used to improve tracking and accuracy. The decoupled motions are obtained by utilizing a tasktailored multimodeflightcontrol system, which implements feedback gains not only as a function of flight condition but also as a function of the mode selected. The aircraft performance can then be tailored to match the desired characteristics of a specific task or mission. For the longitudinal dynamics of a control configured vehicle, the flaperons and elevator form a set of redundant control surfaces capable of decoupling normal control forces and pitching moments. The decoupled motions include pitch pointing, vertical translation, and direct lift control. Pitch pointing is characterized by pitch attitudecommand without a change in flight path angle. Vertical translationischaracterized by flight path command without a change in pitch attitude. Direct lift control is characterized by normal accelerationcommandwithout a change in the angle of attack. For the lateral dynamics of a control configured vehicle, the vertical canard and rudder form a set of redundant surfaces that is capable of producing lateral forces and yawing moments independently. The decoupled
This work was done while Kenneth M. Sobel andEliezer Y . ShapirowerewithLockheed91520. CaliforniaCompany,inBurbank,CA Eliezer Y . Shapirois now with HR Textron, Valencia, CA 91355.

Eigenstructure Assignment Basics

Consider an aircraft modeled by the linear time-invariant matrix differentialequation given by

P=Ax+Bu
y =C X

(1)

(2)

where x is the state vector (n x l), u is the control vector (rn x l), and y is the output vector ( r x 1). Without loss of generality, we assume that the m inputs are independent and the r outputs are independent. Also, as is usually the case in aircraft problems, we assume that rn, the number of inputs, is less than r , the number of outputs. If there are no pilot commands, the control vector u equals a matrix times the output vector J.
= -Fy

If improved eigenvector assignability is required, then additional independent control surfaces must be added.
Now suppose that in addition to transient shaping, we desire the controlled (or tracked) aircraft variables yr to follow the command vector u, with zero steady-state error where

v, = Hx

(3)

The feedback problem can be stated as follows [l]: Given a set of desired eigenvalues, (A?), i = 1,2, . . . , r and a corresponding set of desired eigenvectors, (vf), i = 1, 2 , . . . , r , find a real rn x r matrix F such that the eigenvalues of A - BFCcontain (A:) as a subset,andthecorresponding
0272-1708/85/0500-0009$0l .OO 0 1985 IEEE

The complete control law is derived by Broussard [6] and Davison [ 7 ] . If the command inputs u, are constant, and if the tracking objective is to have the aircraft variables y, approach the command inputs in the limit, then the control input vector is given by
u = (flz

-+ FC&)uc
-

Fv

(4)

feedforward

feedback

May 7985

where

implement modes 1and 2, using the same gain matrix.

Further details of the eigenstructure assignment algorithm may be found in [ 11.

Pitch Pointing (Mode 1) and Vertical Translation (Mode 2)

The objective in pitch pointing control is to command the pitch attitude while maintaining zero perturbation in the flight path angle. The measurements are chosen to be pitch rate, normal acceleration, altitude rate. and control surface deflections. The altitude i r data computer, rate is obtained from the a and it is used to obtain the flight path angle via the relationship

Longitudinal Multimode Flight Control Design

The model of the advanced fighter aircraft [8] will be described by the short period approximation equations augmented by control actuator dynamics(elevator and flaperons).The equations of motion are described by Eqs. (1) and (2) where the state x has five components and the control u has two components. -pitch attitude -pitch rate -angle of attack -elevator deflection -flaperon deflection -elevator deflection command -flaperon deflection command

y =(6) h/TAS
where TAS is true airspeed. The surface deflections are measured by using linear variable differential transformers (LVDT) . We include y as a state because this is the variable whose perturbation we require to remain zero. Thus, we replace 0 by y + a in the state equations and obtain an equation for y. The resulting state-space model is given by Eqs. (1) and (2) with
x = [Y. 9 , a,s e ,
24

the projection of the desired eigenvector onto the allowable subspace. We choose the desired eigenvectors to decouple pitch rate and flight path angle. Such a choiceshould prevent an attitude command from causing significant flight path change. The desired eigenvectors and achievable eigenvectors are shown in Table 1, from which we observe that we have achieved an exact decoupling between pitch rate and flight path angle. The"X" elements in the desired eigenvectors represent elements that are not specifiedkcause they are not directly related to the decoupling objective. We now compute the feedforward gains by using Eq. (4). For the pitch pointing problem
J,

Hx =

[e, y ] '

(loa) (lob)

U, =

[e,,

yclr

where
0, = pilot's pitch attitude command

x =

yc = pilot's flight path angle command

H = [

A =

S f ] '

-0

(7)

1 0 0 0 -0.8693 43.223 -17.251 -1.5766 0 0.9933 -1.341 -0.1689 -0.2518 0 0 0 -20 0 -0 0 0 0 -20

= [ L ,6 , J

(8)

y = [q. nvP,y, 6,,

S f ] '

(9)

The feedforward gain matrix consists of four gains, which couple the commands 0, and yc to the actuator inputs. The control law is described by

B=[;

j]
I
unstable short period mode pitch attitude mode actuator mode mode 0.0

Theeigenvalues of theopen-loopsystem from matrix A are given by A1 = -7.662 h z = 5.452

h3 =

h4 = -20 elevator h5 = -20 flaperon actuator

Our fiist step in the design is to compute the feedback matrix F. The desired short period frequency and damping are chosen to be 6 = 0.8 and on= 7 rad/sec. These values were chosen to meet MIL-F-8785C specifications for category A, level 1 flight. Category A includes nonterminal flight phases that require rapid maneuvering. precision tracking, or preciseflightpathcontrol. Level 1 flying qualities are those that are clearly adequate for the mission objectives. We can arbitrarily place all five eigenvalues because we have five measurements. We can also arbitrarily assign two entries in each eigenvector because we have two inputs. Alternatively, we can specify more than two entries in a particular eigenvector, and then the algorithm will computeacorresponding achievable eigenvector by taking

When yc = 0. we can command pitch attitude without a change in flight path angle , = (pitch pointing). Alternatively, when 8 0, we can command flight path angle without a change in pitch attitude(vertical translation).' The feedback and feedforward gains are shown in Table 2. The pitch pointing and vertical translation responses are shown in Fig. 1. We observe that both responses exhibit excellentdecoupling betweenpitch attitude and flight path angle. An additional feature of thedesign is that the aircraft is stable with goodhandlingqualities in the event of a flaperon failure. Of course, decoupled modecontrol would no longer be possible.

The normal acceleration at the pilot's station nrp is used as a controlled aircraft variable for pitch pointing.

Table 1 Eigenvectors for Pitch PointinglVertical Translation

n ,

= [-0.268,47.76, -4.56,4.45]

Achievable Eigenvectors Desired - 9 - Eigenvectors a

6,

(5)

-Sfwhere nrp is in g's and 9. a, &, and 8 , are in radians or rad/sec. In what follows, we
10
~

/E Control Systems Magazine

~~

Table 2 Pitch PointingNertical Translation Control Law Feedforward Desired Gains Eigenvalues
A;'.2

Mode 3: Direct Lift Control

The objective in direct lift control (DLC) is to command normal acceleration (or equivalently flight path anglerate)without a change in angle of attack. To achieve acceleration command following, we include integrated normal acceleration in the state vector. Thus, for the DLC problem, we choose the state vector to be
x = [q. a, a, ,

= -5.6 5 j4.2
= =

n,

6,

Sf

A: = -1.0 -2.88 -0.367 -0.931 -0.149 -3.25 -0.153 0.747

Ai
A<

-19.0 2.02 -19.5

4.08

][

0.954 0.210 6.10 0.537 -1.04

Sf,4 '

1.5

where

nd=

integral of normal acceleration at the pilot's station

"
w

1.0

The measurement vector is chosen to be

+ .
m

Y
0.5

[ q , a , nd: 6,,

' I %

-I
0.0
1.o

0.0
2.0

3.0

4.0

5.0

(A) PITCH POINTING RESPONSE

The desired eigenvalues and desired eigenvectors are shown in Table 3. Observe that are the zeros in thedesiredeigenvectors chosen to decouple the short.period motion from the normal acceleration. The feedback gain - matrix is also shown in Table 3. To obtain normal acceleration command following, we feedback the integral of the error between measured nzpand commanded nip. The control law is described by

.=[
where

-9-f11
-f21 -f12 -h2 -f,4 -f24

-f15

-fis

6, 6f-

4 )
- 0.5

= rlq -

(nrp)command

0.0

1 .o

2.0

3.0

4.0

5.0

(6) VERTICAL TRANSLATION RESPONSE

Fig. 1. Longitudinal decoupled responses.

The DLC responses to a lg normal acceleration command are shown in Fig. 2. Observe that we achieve a large change in flight path angle with an insignificant deviation in angle of attack. Thus, the aircraft is climbing with almost no change in angle of attack.

Table 3 Direct Lift Control Summary Desired Eigenvalues Desired Eigenvectors Feedback Gains
-0.722-6.70-0.220 -0.301 0.994 5.51

0.468 0.0587 0.223 0.187

4 a short period

May 1985

11

ANGLE O F ATTACK,

a (DEGREES1

-0.086"

cy

0.0

I 0.5

I 1.0

I
1.5

I
2.0

I
2.5

I 3.0

I 3.5

I I 4.0 5.0 4.5

'I

FLIGHT PATH ANGLE, 7 (DEGREES)

13.0"

The couplings that we wanted to be zero are now greater than they were for design # 1. Also, the eigenvalues are not quite where we had specified that they should be. The responses for design #2 are shownin Figs. 3 and 4. Figure 3 showsthe lateral pointing response to a unit step heading command. The change in flight path angle is less than 0.01 degree, and the change in bank angle is less than 0.25 degree. Figure 4 showsthe lateral translationresponse to aunit step flight path command. The change in heading is less than 0.012 degree, and the change in bank angle is less than 0.14 degree. Both designs are considered to achieve acceptable performance.

' 5 0.0 0.5 -1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Mode 6: Direct Sideforce Control
TIME(SECONDS)

Fig. 2. DLC responses.

Lateral Multimode Control Law Design

The model of the flight propulsion control coupling(FPCC)lateraldynamicsis described by eight state variables x , three control variables u, and five measurement variables J. The eight state variablesare sideslip angle (p), bank angle (4).roll rate ( p ) , yaw rate ( r ) , lateral directional flight path ( y = $ + p ) , rudder deflection ( c S r ) . aileron deflection (ijO), and canard deflection (6<). Thethreecontrolvariablesare rudder aileron command (60c.), and command (6rc), canard command ( t j C c ) . The five measurement variables are r-. p , p . &, y. Because of space limitations, the detailed numerical results are not presented, but the general approach is outlined. Further details maybe found in [9]. In what follows, we implement modes 4 and 5 using the same gain matrix.

We computethefeedforwardgains by using Eq. (4). Thetracked variables are given by

?'I

[4,y. $IT

The objective in direct sideforce control (DSC) is to command lateral acceleration (orequivalentlylateraldirectionalflight path) without a change in sideslip angle. To achieveacceleration commandfollowing, we include integrated lateral acceleration in the state vector. Thus, for the DSC problem, we choose the state vector to be
.x = M , 4 , p - r, Y. 6,,

6.. S,, n,,,lT

and the pilot commands are given by

I(,

[&.y<.,&IT

where integral of lateral acceleration n p , = at pilot's station

where
&.

= commanded heading

yc = commanded lateral directional

The measurement vector is chosen to be

flight path
$<

.v

[ r , P , p . 6. n,,lr

=0

Yaw Pointing (Mode 4) and Lateral Translation (Mode 5)

We desire to decouple the lateral directional flight path responsefrom the bank angle, roll rate, and yawrate responses. Thus, the desired eigenvectors are chosen such that the flight path mode will not affect the bank angle, roll rate, or yaw rate responses and so that the flight path response will consist only of the flight path mode. Fordesign # I , which corresponds to the full feedback gain matrix. the achievable eigenvectors are very close to those that were desired. The control law gives achievable eigenvalues almost exactly those that were desired.
12

Since bank angle is commanded to be zero. we neednotimplementthegainsthat multiply dc.Itis included in the numerical computations only to avoid the need fora pseudo-inxrekion. The time histories for design # I are not shown; however. the yaw pointing responses to a unit step heading command are such that I y ( 2 0.0004 degreeand 14, 5 0.0031 degree.Thelateraltranslation responses to a unit step lateral flight path command 14; 5 0.008 degreeand aresuchthat : b ,5 0.004 degree. Design #2 is characterized by an additional specification that seven of thefeed.back gains be constrained to be zero. The zero elements are chosen based upon the physical insight that the roll autopilot should be able to operate somewhat independently of the lateral directional control system. Of course. some degradation will result, but the responses will still be acceptable from a practical point of view. Furthermore, by reducing the number of gains, we have increased the reliability of the control system.

Thedesiredeigenvectors were chosen such that the lateral acceleration mode would be decoupled from both the dutch roll mode and the roll mode.This choice yields the following:

-_-_

x 1

0 0
0 0

1x

x x
X J X

x x x x

:y,o,--

0 _- 0x 1

dutchrollrollmode

To obtain lateral acceleration command following. we feed back the integral of the error between measured n,, and commanded nyp.This approach is similar to that used for direct lift control. Several designs are investigated. Design #1 is characterized by an output feedback
F E E Control Systems Mogozine

-:"
0 0

1 0 :

x 1

10;

xx x x

x x
-1_
acc mode mode deGoupling

1.20

0.00

-0.04

-0.08

0.00

1.oo

2.00

3.00
HEAOING

4.00

5.00

-0.12

I
0.00 2.00

I HEADING

1.00

I 3.00

I 4.00

5.00

0.08 r

1.20

0.00

-0.04

I
0.00

1.00

2.00

I 3.00

4.00

I 5.00

0.00

1.00 3.00

2.00
LATERALFLIGHT PATH

4.00

5.00

LATERAL FLIGHT PATH

0.00

1 .oo

2.00

3.00

4.00

5.00

0.00

1 .oo

2.00

3.00 5.00

4.00

BANK ANGLE

BANK ANGLE

Fig. 3. Yaw pointing response.

Fig. 4. Lateral translation response.

gain matrix without anygainconstraints. The time histories (not shown) exhibit a lateralacceleration,whichachievesits lg commandedvaluewhilethesteady-state sideslip angle and bank angle are described = by Ip,.I = 0 . 4 4 d e g r e e a n d 145rl 0.014 degree, respectively. Design #2 is characterized by the constraints that bank angle and roll rate shall not be fed back to either the rudder or canard. In this design, the number of gains is reduced by more than 25 percent as compared to
May 7985

design #l. The time histories (not shown) exhibitalmostthesamebehavioras in design # l . Design #21 is characterized by feeding back proportional plus integral sideslip angle described by

Thetimehistoriesforthisdesignare shown in Fig. 5. Sideslipangleattains

maximum of 0.42 degree, but it approaches zero as the time into the maneuver increases. The bank angle attains a maximum value of 0.18 degree, which is acceptable although it is larger than for designs #1 and #2. Further, we observe that both heading and lateral flight path have achieved a change in excess of 17 degrees during the 10-sec maneuver. We conclude that design #21 is acceptable because it achieves heading and flight path changes with insignificant variations in sideslip angle and bank angle.
73

Fighter Aircraft," Proceedings of the AIAA 9th Atmospheric Flight Mechanics Conference. San Diego. California. Aug. 1982. [9] K. M . Sobel. "Application of Eigenstructure AssignmenttoTask-TailoredMultimode Flight Control System Design." Lockheed California Company Report LR-30852, Feb. 1985.
1.20 r

o * o .r -t

LATERAL ACC ATPILOT(G'sl

BANK ANGLE IDEGI

-0.120 -0.178

0.69
0.46

r c
I
I I

0.00 1.00 2.00

3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00

TIME(SECONDS)

Fig. 5.

Directsideforce responses.
ceedings of the 1984 Sational Aerospace ar7d Electronics Conference, Dayton. Ohio.

Conclusion
A task-tailoredmultimodeflightcontrol system was designed by using eigenstructure assignment.Thepilot can choosebetween sixdifferentmodesinordertomatchthe a specific task or misaircraft performance to sion. The design methodologies for the different modes have been described including an explanation of the choices forthe desired eigenvectors. Aircraft responses were shown to demonstrate the effectiveness of the flight control system.

pp. 570-575. May 1984.

141 K. M . Sobel and E. Y. Shapiro. "Application

Kenneth M. Sobel was born in Brooklyn, New York. in 1954. He rec e i v e dt h eB . S . E . E . degree from the City College of New York in 1976 andtheM.Eng. andPh.D. degrees in 1978 and 1980, respectively,fromRensselaerPolytechnicInstitute. Troy, N e w York. From 1976 to 1980, he was a Research Assistant with the Department of Electrical and Systems Engineering. Rensselaer Polytechnic Institute. There he performed research in the areas of reduced state optimal stochastic feedback control and model referenceadaptive control for multi-input multi-output systems. Since 1980. he has been with Lockheed California Company where he has published numerous papers on the application of modern control theory to aerospace problems.His wsork on adaptivecontrol has been chosen to appear in Academic Press' Advances in Control a17d D y a m i c Sxstems. Dr. Sobel has taught courses in lumped parameter systems, feedback systems.and digital systems at both Rensselaer Polytechnic Institute and California State University, Northridge. He currently holds the position of Adjunct Assistant Professor at the University of Southern California, \vherehe teaches courses in the Electrical Engineering Department. Dr. Sobel is a Senior Member of LEEE and a member of Eta Kappa Nu and Sigma Xi. E l i e z e r Y. S h a p i r o receivedtheB.Sc.and M . S c .d e g r e e sf r o m Technion-Israel in 1962 and 3965, respectively, theSc.D.degreefrom ColumbiaUniversity, New York. in 1972, and theMBAdegreefrom UCLA in 1984. From 1962 to 1966, he was with the Armament Development .4uthority, Israeli Ministry of Defense, where he was responsible for the analysis, design. and evaluation of high performance analog and digital systems. From 1968 to 1972, he developed computer-controlled typesetting equipment for Harris Intertype Corporation and was a Consultant to theVarityperDivisionof Addressograph-Multigraph Corporation. During 1973-1974.he was a Senior Staff Engineer for PRD Electronics. Inc. In 1974, he joined Lockheed California Company as an Advanced Systems Engineer involved in applyingmoderncontrol theory to the SR-71 and other advanced aircraft.
IEEE Control Systems Mogozine

References
A. N. .4ndq. E. Y. Shapiro, and J. C. Chung. "Eigenstructure Assignment for Linear Systems." IEEE Trans. Aerosp. Electron. S w . , vol. AES-19, pp. 711-729, Sept. 1983. K. M . Sobel and E. Y. Shapiro. "A Design Methodology for Pitch Pointing Flight Control Systems,"J. Guid. Contr..and Dpnam., vol. 8, no. 2.Mar.-Apr. 1985. K . M . Sobel. E. Y. Shapiro.and R. H. Rooney. "Synthesis of Direct Lift Control Laws Via Eigenstructure Assignment." Pro-

of Eigensystem Assignment to Lateral Translation and Yaw Pointing Flight Control.'' Proceedings of rite '3rd IEEE Conferenceon Decision and Control. Las Vegas. Nevada, pp. 143-1428.Dec. 1984. [ 5 ] K . M . Sobel. E. Y. Shapiro.and A . N. Xndq. "Flat Turn Control Law Design Using Eigenstructure Assignment." Proceedings of 7th International S~rnposiunl017 theMarhernatical Theon of .vCht'OrkS and S.vsteins. Stockholm. Sweden. June 3985. [6] M. J. O'Brien and J . R. Broussard. "Feed Forward Control to Track the Output of a ForcedModel." Proceedings of the 17th IEEE Conference on Decision and Control. San Diego, California. Jan. 1979. [7]E. J . Davison. "The Steadystate Invertibility andFeedforwardControl of Linear Time-Invariant Systems." IEEE Trans. Auto. Contr.. vol. AC-?I. no. 4. pp.529-534. Aug. 1976. [8] D.B. Ridgely. J . T. Silverthorn. and S. S. Banda. "Design andAnalysis of a Multivariable Control Systemfora CCV Type

He was responsible for applying observer theory to thedesign,development,andsuccessfulflight test on a Lockheed L-1011 aircraft of a state reconstructorthatgenerated a yawratesignal throughanalyticalmeans.Subsequently,he formed a team of highly qualified specialists who focus on developing new techniques for control of high performance aircraft and applying them in actual hardware. This team has been formalized into the FlightControlResearchDepartment,

which, under Dr. Shapiros direction, is responsible for company-wide research and development of advanced flight control systems analysis, design, and evaluation methodology. Since 1985, Dr. Shapiro has been General Manager of the Special Products Division at HR Textron, Valencia, California. He has published over 70 papers on the application of modem control theory to aerospace problems. Dr. Shapiro is cocreator of a course entitled,

Analysis and Design of Flight Control Systems Using Modem Control Theory, which he teaches at the University of California, Los Angeles, and the University of Maryland. Through this course, he shares his experience in using modem control theory to successfully design advanced flight control systems. Dr. Shapiro is an Associate Fellow of AIAA, IEEE, and a member of SeniorMemberof Sigma Xi.

1985 American Control Conference

TheFourthAnnualAmericanControl Conference will take place in the historic and dynamic setting of the Boston Maniott Hotel Copley Place, Boston, Massachusetts, from June 19 to 21, 1985. A total of 40 invited sessions and 18 contributed sessions will be presented on all aspects of control theory and practice. The scheduled plenary speakers include Dr. Nam P. Suh, Assistant Director forEngineeringattheNationalScience Foundation; Dr.DonaldC.Fraser, Vice President of the Charles Stark Draper Laboratory; and Dr. Arthur Gelb, Founder and President of TheAnalyticSciencesCorporation (TASC). TheGeneral Conference Chairman is Professor Yaakov Bar-Shalom. For questions regarding the technical prog a m , contact the Program Chairman: Professor David Wormley, MIT, Department of Mechanical Engineering, Cambridge, MA 02139, phone: (617) 253-2246. See the February 1985 issue of IEEE Control Systems Magazine (p. 46) for the Hotel Reservation Form and the Advance Registration Form.

Paul Revere Statue

May 1985

U .S .S . Constitution Old Ironsides

Minuteman Statue in Lexington

15