Professional Documents
Culture Documents
Nonlinear Model Linear Robust Control Overhead Traveling Cranes
Nonlinear Model Linear Robust Control Overhead Traveling Cranes
Nonlinear Model Linear Robust Control Overhead Traveling Cranes
2197-2207, 1997
Proc. 2nd World Congress ofNonlinear Annlysts
Pergamon 0 1997 Elsevier Science Ltd
hinted in Great Britain. All rights reserved
0362-546X/97 $17.00 + 0.00
PII: SO362-546X(97)00107-7
Keywords and phrrrsea: Overhead traveling crane, Nonlinear model, Robust conrtol.
1. INTRODUCTION
Overhead traveling cranes are widely used in factories for the transfer of heavy loads. A truck which
bears a load is driven by a motor Mx, and it moves on a rail in the transversal direction. The load is
hoisted by a hoisting motor Mz installed on the truck. Furthermore, the whole device which contains
the truck, the hoisting motor Me, and the driving motor Mx is driven by another motor My, and it
moves on an another rail in the longitudinal direction. Thus, the truck can freely move on a overhead
horizontal plane.
From the control point of view, it is desired to transfer a load to a specified place as quickly as
possible while minimizing the swing of the load during transfer as well as the swing at the end of
transfer. For designing a proper controller of the crane, we need a mathematical model of the whole
crane system. On the basis of Newton’s law of motion, a system of nonlinear ordinary differential
equations can be derived as state equations, where torques of three motors are taken as controls. We
manipulate the state equations and obtain a simple bilinear system. During the repeat of operations
of load transfer, controller does not know each value of load mass, but only knows its nominal value.
On the basis of given initial and final conditions of the dynamical system, an open-loop control
input to each motor can be calculated, such that the state of the system can be transferred to a
neighborhood of the equilibrium state corresponding to the given final condition. At the end of the
open-loop control, we apply feedback control inputs to the motors so that the state of the system
approaches the equilibrium state as quickly as possible. Thus our control scheme here is open-loop
plus feedback control, which is similar to our previous control scheme for a rotary crane [l].
2. DERIVATION OF DYNAMICAL MODEL
We consider an overhead traveling crane as shown in Fig. 1 and Fig. 2. For simplicity, we make the
following assumptions.
Assumption 1. The load can be regarded as a point mass.
Assumption 2. Frictional torques which may exist in torque-transfer mechanisms can be neglected.
Assumption 3. The ropes have no mass and no elasticity.
The following notations which are shown in Fig. 1 and Fig. 2 will be used.
(t,d : coordinates of the root of the hoisting rope in the horizontal plane
(%iLC) : three-dimensional coordinates of the load when the origin is taken at the
2197
2198 Second World Congress of Nonlinear Analysts
We see that.
5 = lsin$cos$, jj = lsin$sin$, C = Icost&
(3)
(4)
On the other hand, the truck moving in the x-direction is subjected to the forces F and ?‘,, and
the girder moving in the y-direction is subjected to the forces F and FYI respectively. Hence, the
equations of motion are written as
In the same way, the equations of motion of the driving motors Mx, My, and MZ are given as
J,ii, = --Fs, t ~z, J,&, = -Fp,, t ry, J,e’z = Fr, t Tz. (‘3)
It is clear that
(= Trbz, rj = 7yoy, I= r,8c. (7)
Eliminating F, and F, from (5) and (6) and using the relation (7) yields
where
1 1
a= = m, + (J&g’ ay = my t (J&)’
It follows from (3) and (8) that
Equations (4), (8) and (10) are the fundamental equations of the crane system. The state equations
can be derived from these equations by taking f, f, [, 7, C and i, $, i, rj, i as state variables. Note
that the variables t, 7, and 1 and their time derivatives can be observed by measuring the rotation
angles and the rotation speeds of the three motors Mx, My, and Mz, respectively, and that measuring
the variables 2, 5, [ and their time derivatives is not easy.
Now the problem is to express the rope tension F in terms of the state variables and the control
variables r2, rv, and rz. It is clear that
l2 = I2 + f2 + (2. (11)
11
2
i2 + S” t (2 _ i2
F/ml = (1 + ?)I” + ma,52 + mavg2
L
In (14), if we use (11) for I and (12) for i, then F/ml can be expressed in terms of the state variables
- LL’
Z, y, [, z, y, [ and the control variables.
Let us define the new control variables
where cf is the z-coordinate of the final position of the load. Because 1 # 0, given the state variables,
the pair (~r,us,‘u) gives a unique pair of (rz,rr,,rz), and vice versa. In the final equilibrium state,
we see from (4), (9) and (10) that ~1 = us = v = 0, rr = -mgrl, and F = mg. From physical point
of view, the tension F should be nonnegative. As a sufficient condition for F 2 0, we assume that
Let the final position of the load be [ = [r, 17 = of, and C = [f, and let us define another state
variables by
i=t-tr, f=v-w, t=c+.
Then (4) is rewritten as
Second World Congress of Nonlinear Analysts 2201
(17)
Furthermore, if we define the first state vector
For the second state vector x2 = [c, i, ii, $I’, we obtain the same equation as (18):
where the matrix Arv:i; obtained from Al, b!y replacing p2= in Al, by p2v. Let us define the third
state vector x3 = [<, C] . From (16) the following state equation is obtained:
If we define the state vector x = [XT, XT, xTIT and the control vector v = [ul, 2~2, vlT, the state
equation of the system can be written as
k=Ax+B,v+~xv. (21)
It is clear that Equation (21) has the bilinear form. It can be easily seen that the pair (A, B,) in
(21) is controllable.
We assume that the dynamical system is initially in the equilibrium state corresponding to t(O) =
to, ~(0) = 1)0, C(O) = (a. Then th e initial condition is given by
where
In the same way, we define the functions ~~(1) = a(Tf - w)[l - cos(nt/tl)] + ~1~. The standard
crane operation is as follows: The load is hoisted up from the initial height 6 to a higher position
Cm(O<Ln<C) 0 c1oser to the ceiling in order to avoid obstacles on the earth, then the load is kept to
the position c,,, for the time interval t,l 5 t 5 tm2, and lastly it is hoisted down to the final position
Cf ([f > <,,,) near the earth level. Consequently, we assume the following nominal trajectory:
where c,,, , t,l , and t,.,,2 should be chosen properly. Further, if we define the functions
w = %(x,U1,u2,7z;m).
The inverse relation of the above equation will be denoted by 7Z = +-‘(x, x1,74, w; m). We assume
that all the parameter values except the value of m are known. Once the initial state XI, the heights
[,,, and 0, and the times t,l , t,2 and tl are given, the open-loop control torque inputs @(t),u;(t),
and T:(t), that almost realize the trajectories p(t), f(t), and C”(t), can be calculated via the
following procedure:
Second World Congress of Nonlinear Analysts 2203
l Given the functions p(t), q”(t), and C”(t) > 0, first calculate (F/ml)O(t) using the equation
(FlmvYt) = (9 - i;“WlC”(t),
which is obtained from (4). To ensure the relation F 2 0, the times tml,tmz and tl should
satisfy the inequality
l Solve the differential equation (3) for g(t), i(t) and t(t), i(t) corresponding to the p(t) and
v”(t), respectively.
which is obtained from (8) by using fi in place of m, calculate u:(t) and u:(t) corresponding
to the p(t) and v”(t), respectively.
. Since 1 and i can be calculated from (1.1) and (12), respectively, 7;(t) can be calculated by
using 7,” = 9 -l(xo,u~,u;,wo;fiz).
It is clear that, if m = 7fi, among the ten state variables, six state variables i, i, +, $, I, and 1
satisfy the final conditions. Since the value of m is in a neighborhood of %, at the time tl all the
state variables are supposed to be in a neighborhood of the specified final state. Therefore, at the
final time tl we switch the control scheme frsom the open-loop control to the feedback control. In
the following, we linearize the state equations and the output equations about the final equilibrium
state xF, and we design a robust feedback controller on the basis of the linearized equations.
Now, let us define a new control variable 113 by
Thus, the linearized state equations and output equations around x = xF are obtained as follows:
where
On the basis of the linearized equations, linear robust feedback controllers will be designed. Since
(27) and (28) have the same form, we only consider (27). It is easily seen that the linear system (27)
is controllable and observable.
Since the mass m and the final height <f of the load are usually variable at each repeat of operation,
a robust feedback controller is required for the repeated operation of the crane. All parameters of
the system are pr and pzr defined by (17). If the observer is designed for a fixed set of parameter
values of the system, and variation of the parameter values is large, the linear closed-loop system
incorporating the observer may become unstable [2]. When the system is minimum phase, high gain
observer can stabilize the system [3]. H owever, because our linearized system is not minimum phase,
using observer deteriorates the robustness of the linearized system. Fortunately, it happens that
the linear system can be stabilized by the output feedback. Let Ki = -[ki, kz]. Then the output
feedback is given by
u1 = Klyl = -k& - kz;. (30)
It is easy to see that
Id- AI, - BlKrCll = s’ + kas3 + (PI + par + kl)s’ + plkas + plkl. (31)
Because pl > 0, pz2 > 0, by using the Hurwitz stability criterion, the linear feedback system is stable
if and only if
ICI > 0, kz > 0.
Note that the above stability condition also ensures the robust stability of the linearized system,
because it is independent of the system parameters p1 and ~2~. In the same way, the linear system (29)
can be robustly stabilized by the output feedback uz = Kzy2 with K2 = -[k3, k4] (k3 > 0, kq > 0).
Let K3 = -[kS, kg] (kg > 0, kc > 0). The third system governed by (29), which is equivalent to
(26), can also be stabilized by the output feedback
f + k6t + k5[ = 0.
The control scheme here is the open-loop plus feedback control. Also, it is the linear control
of the nonlinear system. Clearly the open-loop control torque u”(t) is applied to the system until
t = tr. At t = ti the system is switched to a feedback control system. If ti is too small, the swing
of the load during the transfer will be too large. Therefore, we should choose as small value of tr
as possible, such that the swing of the load during the transfer can be kept small. The closed-loop
transfer function from the external input to the output yr is given by
G,(s) = n(s)/d(s)
[I ; =
s’ t kas3 t (PI t
where X; are the roots of the characteristic equation d(s) = 0. The gain values kg and ,bs for us were
chosen as
k5=$, b=2w,.,,
where w, is a proper constant.
We are trying to prove the global stability of the original nonlinear system (21). However, because
of the strong nonlinearity (14), we have not succeeded yet.
4. SIMULATION
Computer simulations were carried out for planar motion of the crane, where ~(1) E 0. The data
used for simulation were as follows: m, = 8OOkg, J=/ r: = 200kg, J,/T~ = 200kg. Consequently,
CY - 0.001. Nominal values of the variable parameters were taken as 7fi. = l,OOOkg, cf = 4.9m.
Tieiefore, Fr = g/cf = 2, and ~2~ = jjra,ii~ = 2. The initial and final conditions were taken as
[IJ = 0, If = lOm, <u = <f = 4.9m. Also %,,,I = 49, t,,,2 = 6s, tl = lOs, <,,, = 2m. By using
the Optimization Toolbox of MATLAB [4], f,he gain values that minimize (34) were calculated as
kl = 2.0, k2 = 2.83. Also, w,, was chosen as w, = 2.
In the first case, the values of the variable parameters were assumed to coincide with the nominal
values. Fig. 3 shows the time responses of the truck position t(t) and its speed i(t), the swing S(t)
and its speed i(t), and the negative height C(t) and its speed i(t), respectively, when m = fi and
[f = cf. Fig. 4 shows the time responses of the control torques q(t) and us(t), and the variable
v(t) = (l/pr)(F/ml) - 1 that represents the tension of the rope. Up to t = tr = 10s the open-loop
control inputs are given, and after t = tl the feedback control inputs are given to the nonlinear
system. Fig. 3 shows that after t = 51the swing Z(t) decays quickly by the feedback. From Fig. 4
we see that v(t) 2 -1. This means that the tension of the rope is always nonnegative.
In the second case, we calculated the time responses when m = 800kg and f,f = 5.8m, and the
feedback gains were fixed. Fig. 5 and Fig. 6 show the same time responses as in Fig. 3 and Fig. 4,
respectively. In this case, c(t) approaches the final value [f oscillating during the open-loop control,
and the swing Z(t) decays but not so quickly as in Fig. 3. As far as the simulation results are
concerned, the linear output feedback stabilizes the nonlinear system and works satisfactorily even
the parameters vary.
REFERENCES
1. SAKAWA Y. & NAKAZUMI A., Modeling and control of a rotary crane, ASME Journal of Dynamic
Systems, Measurement, and Control, 107, ZOO-206 (1985).
2. DOYLE J. C. & STEIN G., Robustness with observers, IEEE ‘Bans. Automatic Conbol, AC-24, 607-611
(1979).
3. JABBARI F. & SCHMITENDOHF W. E., Effects of using observers on stabilisation of uncertain linear
systems, IEEE Thw. Automatic Control, 38, 266-271 (1993).
4. GRACE A., Optimiration Toolbox User’s Guide, The MathWorks Inc. (1994).
2206 Second World Congress of Nonlinear Analysts
-5
0 5 10 15 20 25
Time(s)
Time(s)
4 ;--\,
,---I _____--_______-____---.
OC \ ,
._-’
-2L 0 5 10 15 20 25
Time(s)
Time(s)
Time(s)
15.
-5
0 5 10 15 20 25
Time(s)
0 5 10 15 20 25
Time(s)
i
c
0 5 10 15 20 25
Time(s)
10
0 5 10 15 20 25
Time(s)
0 5 10 15 20 25
Time(s)