A Time Delay Controller for

Kamal Youcef-Toumi
Associate Professor.
Systems With Unknown Dynamics
This paper focuses on the control of systems with unknown dynamics and deals with
Osamu Ito the class of systems described by x=f(x,t) +h(x,t) + B(x,t)u + d(t) where h(x,t)
Graduate Student. and d(t) are unknown dynamics and unexpected disturbances, respectively. A new
control method, Time Delay Control (TDC), is proposed for such systems. Under
Department of Mechanical Engineering,
the assumption of accessibility to all the state variables and estimates of their
Laboratory for Manufacturing and delayed derivatives, the TDC is characterized by a simple estimation technique that
Productivity, evaluates a function representing the effect of uncertainties. This is accomplished
Massachusetts Institute of Technology, using time delay. The control system's structure, stability and design issues are
Cambridge, MA 02139 discussed for linear time-invariant and single-input-single-output systems. Finally,
the control performance was evaluated through both simulations and experiments.
The theoretical and experimental results indicate that this control method shows ex-
cellent robustness properties to unknown dynamics and disturbances.

1 Introduction
Most of the well-developed control theory, either in the fre- ance is obtained. Therefore, this approach is restricted to
quency domain or in the time domain, deals with systems repetitive tasks only.
whose mathematical representations are completely known. Other control systems have also been proposed for
However, in many practical situations, the parameters of the dynamical systems with norm bounded uncertainty in system
system are either poorly known or operate in environments parameters and input disturbances. Selected references in this
where unpredictable large system parameter variations and area are [7] and [8]. Reference [12] extends the fundamentals
unexpected disturbances are possible. Underwater vehicles, of robust nonlinear control strategies with applications to
robot manipulators, and autonomous systems are a few ex- robot manipulators.
amples. In such situations, the usual fixed-gain controller will This paper proposes another control method, Time Delay
be inadequate to achieve satisfactory performance in the entire Control (TDC) [17, 18], which depends on neither estimation
range over which the characteristics of the system may vary of specific parameters, discontinuous control, nor repetitive
[10]. actions. Rather, it depends on the direct estimation of a func-
Several advanced control techniques have been developed tion representing the effect of uncertainties. This is ac-
for such systems. One of the primary methods is Adaptive complished using time delay. The gathered information is used
Control [3, 6, 11, 14], In Adaptive Control, the structure of to cancel the unknown dynamics and the unexpected distur-
the controller is selected a priori, usually PD or PID type con- bances simultaneously. Then, the controller inserts the desired
troller. The controller gains are then updated using a recur- dynamics into the plant. In other words, the TDC uses past
sively estimated parameters of the plant so that the plant out- observation of the system's response and the control input to
put closely follows the desired response. As stated in [5, 9], directly modify the control actions rather than adjusting the
this method considers slowly varying parameters, linear controller gains or identifying system parameters thereby
dynamics equations and/or bounded uncertainty. leading to a model independent controller. This algorithm can
Sliding Mode Control [13, 16, 22] is another powerful deal with large unpredictable system parameter variations and
method which can deal with nonlinear systems. Based on disturbances. Yet, the system's performance is very
Lyapunov's method, the control scheme is characterized by satisfactory.
discontinuous function with high frequency chattering. The In Section 2, the control problem is defined and the Time
plant parameter variations and disturbances are assumed Delay Control algorithm is presented. Section 3 discusses the
bounded. system's structure, stability analysis and design issues for
Learning Control [1, 15] is an approach which is based on linear time-invariant and single-input-single-output systems.
trial and error. Each time the system performs the same task, Section 4 is an evaluation part of the paper. Effectiveness of
data is collected and used to update the control action. By the TDC system is evaluated through both simulations and ex-
repeating this process several times, betterment in perform- periments. Finally, in Section 5, the results are summarized
and the directions of the future research are suggested.
Contributed by the Dynamic Systems and Control Division of THE AMERICAN
SOCIETY OF MECHANICAL ENGINEERS and presented at the American Control
Conference, 1987, Minneapolis, Minn. Manuscript received by the Dynamic 2 Derivation of the Control Law
Systems and Control Division July 1987; revised manuscript received July 1988.
Associate Editor: G. E. Young 2.1 Error Dynamics and Structural Constraint. The

Journal of Dynamic Systems, Measurement, and Control MARCH 1990, Vol. 112/133

Copyright © 1990 by ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 nonlinear systems considered in this paper are described by the and through some algebraic manipulation, given in Appendix
following dynamic equations. I, the error dynamics is given by
x = f(x,0 + h(x,0 + B(x,Ou + d(0 (2.1) e = (A r a +K)e+ ( I - B B + ) ( - f - h - d + A m x + B m r - K e )
where x is an «xl plant state vector, u is an rxl control vector, (2.10)
B(x,/) is an nxr control distribution matrix with rank r, f(x,0
and h(x,0 are nxl nonlinear vectors representing respectively In order to obtain the desired error dynamics given by equa-
known and unknown part of the plant dynamics s and d(0 is an tion (2.7), the following structural constraint must then be met
unknown disturbance vector. The variable t represents time. so that the error vanishes as time goes to infinity.
This expression is common in many applications. For exam- j I - B B + | { r f - h - d + Amx + B m r - K e ) = 0 (2.11)
ple, in robot manipulators, h(x,0 can correspond to nonlinear
torques caused by Coriolis -and centrifugal effects or If B is nxn and B ' exists, the structural constraint of equa-
nonlinearities such as dry friction at each joint. The vector d(0 tion (2.11) is always satisfied since I - B B + = I - B B _ 1 is a
represents any kind of disturbance such as external torques. zero matrix. If not, the choice of the reference model and the
The objective of this research is to be able to control such error feedback gain matrix is somehow restricted since the
systems and guarantee performance despite the presence of above constraint equation contains A,„, B,„ and K. Moreover,
large dynamic variations in h(\,t) and large unexpected distur- some elements of unknown dynamics vector h(x,0 and unex-
bances in d(0- In this paper, it is assumed that all the state pected disturbances vector d(0 should be known in order to
variables and their derivatives are accessible or estimated. ensure the system to satisfy the above constraint. It is shown in
Also, this paper will address issues when the control distribu- Appendix II that this condition is always satisfied for systems
tion matrix B is known. expressed in canonical form. It must be noted that I - B B +
Let us define the reference model that generates the desired has rank n — r and thus equation (2.11) effectively consists of
trajectory as a linear time-invariant system given by n—r constraint equations instead of n constraints. On the
other hand we would like to completely control n states using r
x m = A r a x m + Bmr (2.2) controls. So, the structural constraint simply indicates that r
where xm is an nxl model state vector, A ra is an nxn constant inputs can really control only r states and the rest n-r states
stable system matrix, B m is the nxr constant command should automatically be controlled under the constraint.
distribution matrix, and r is an rxl command vector. The error
vector, e, is defined as the difference between the plant and the 2.2 The Time Delay Estimation and Control Action. A
model state vectors: structural constraint for this scheme has been derived and a
class of system that satisfies it has been identified. Now it is of
e=xm-x (2.3) interest to determine the control action u that will force the
plant to follow the reference model in the face of unknown
The control objective is to force the error to vanish with a dynamics h(x,/) and unexpected disturbance A{t) which appear
desired dynamics: in the right-hand side of equation (2.8). These two terms ap-
pear as a sum and their effect can be determined from the
e = Aee (2.4) plant dynamic equation equation (2.1):
where A e is an nxn error system matrix, which defines some
h(x,0 + d(0 = x(0 - f(x,0 - B(x,0u (2.12)
desirable dynamics. By combining equations (2.1) through
(2.3), one obtains an equation that governs the error In order to obtain an estimate of the effect of the term
dynamics, h(x,0 + d(0, it is considered that the value of the function
h(x,/) + d(0 at the present time t is very close to that at time
e = A m e + { - f - h - d + Amx + B m r - B u ) (2.5) t—L'm the past for a small time delay L,
If it is possible to determine a control u of equation (2.5) such
h(x,0 + d(0 = h(x,?-L) + d(t-L) (2.13)
that the following equation is always met
Now, combining equations (2.12) and (2.13), the effect of the
- f - h - d + A m x + B r a r - B u = Ke (2.6)
function h(x,0 + d(0 is estimated by:
then, substitution of equation (2.6) into equation (2.5) leads to h(x,t) + d(t) = x(t-L)-t(x,t-L)-B(x,t-L)u(t-L) (2.14)
e = (A m +K)e = A e e (2.7) A fundamental step in this derivation is to delay time by an
where K is an nxn error feedback matrix. In the above equa- amount of L to estimate the system's unknown behavior,
tion, the error system matrix A e can be arbitrarily determined namely h(x,/) + A(t). The TDC control law is then obtained by
through proper choice of the error feedback gain matrix K. substituting equation (2.14) into equation (2.8) and is given by
The control signal u, that satisfies equation (2.6), must then be
u(t) = B + (t){ -f(t)-x(t-L) + i(t-L)
selected in order to obtain a desired error dynamics. However,
equation (2.6) cannot always be satisfied because the number
of controls is generally smaller than the number of the states. + B a - L ) n ( ? - L ) + A m x(0 + B m r ( 0 - K e ( 0 ) (2.15)
Thus, a best approximate solution of the equation is adopted In the above equation, each term has the following meaning:
to determine the control u: (1) B + (0, a pseudo inverse matrix cancels the control
matrix B(f),
u = B + { - f - h - d + Amx + B m r - K e ) (2.8)
(2) the term, -f(t)-x(t-L) + i(t-L) + B(t-L)u(t-L),
attempts to cancel the undesired known nonlinear dynamics
where B + = ( B r B ) _ 1 B r and is known as a pseudo-inverse
f(0, unknown nonlinear dynamics h(0 and the unexpected
matrix. Note that the matrix B T B is an rxr nonsingular matrix
disturbance d(0,
since B is of rank r. The condition for which equation (2.8) ex-
actly satisfies equation (2.6) must be determined, since this is (3) the term A„,x + B m r, inserts the desired dynamics of
not always the case. To find this condition, equation (2.8) is the reference model, and
substituted into equation (2.1) to obtain (4) the error feedback term, - K e , adjusts the error
dynamics.
x = f + h + B B + ( - f - h - d + A„,x + B m r - K e ) + d (2.9) Thus this controller observes the states and the inputs of the

134/Vol. 112, MARCH 1990 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 system at time t — L, one step into the past, and determines the where x = [x1( x 2 , .
control action that should be commanded at time t. It is im-
portant to stress that this controller identifies the unknown
dynamics by evaluating a function directly every L seconds. 0, 0, 0,
Therefore, any changes in dynamics will be detected within the 0, 1,
time period L. A=
2.3 Control Law for Systems in Canonical Form. As
mentioned earlier, one class of systems that satisfy the struc-
tural constraint are those expressed in canonical form. Thus, -a*
the objective of this section is to derive the Time Delay Con-
trol law for this specific class of systems. Physical systems can and a,, a2, . . . ,an axe unknown constants, and b is a known
often be described using this representation. Consider a constant. This formulation corresponds to a canonical form
nonlinear plant with n states and r inputs where each term of representation of equation (2.16) in the following manner:
equation (2.1) can be partitioned as follows:

9
0
x= ; f(x,/) = h(x,0 =
Xr fr(x,0 h r (x,0

f r (x,0 = 0, h r (x,0 = - fljx, - a2x2 - Ct„Xn

B(x,/) = ; d(0 = (2.16)
Br(x,0 d r (0 d r (0 = 0,B(x,0 = b (3.2)
The reference model and error feedback gain matrix K can be
where xq and 0 are (« — r)xl vectors, x i = [x r+1 , . . . , x„]T is chosen based on equation (2.17) in order to satisfy the struc-
also an (n-r)xl vector, xr, f,, h r and d r are rx\ vectors, and tural constraint of equation (2.11).
B r is an rxr nonsingular matrix. The reference model and an
error feedback gain matrix can also be partitioned in a similar
fashion such that the total controlled system satisfies the struc-
Jm -*ml (3-3)
tural constraint of equation (2.11) (see Appendix II).

OIL 0 0
B,„ = K=

(2.17)

where, in this case, 0 is an (n — r)x(n — r), (n — r)xr or (n — r)xn wherex,„ = [ x m l , x m 2 ,

matrix, lq is an (n — r)x(n — r) matrix, Amr is an rxn matrix,
B mr is an rxr matrix and Kr is an rxn matrix. Substituting 0, 1,
equations, (2.16) and (2.17) into equation (2.15), the TDC
o,
control law becomes: 0, 0, i,
Am = b,„ =

+ Br(t-L)u(t-L) + A mr x(0 + Bmrr(t) - Kre(t)} (2.18) ~am\> ~am2' a

ml>

Note that in this canonical system formulation, the above con-

trol law includes an inverse of the control distribution matrix 0, 0,
B r instead of a pseudo-inverse matrix as seen in the general
0, 0,
TDC control law in the previous section. To this end, the fun- K (3.4)
damentals of this approach have been defined. These include a
structural constraint condition, a time delay scheme and the
time delay control action. — ki, —k2, . . . , — kn

3 Analysis and Design for LTI-SISO Systems Substituting equations (3.2) through (3.4) into equation
(2.18), the TDC control law can be obtained as follows.
If the plant is a linear time-invariant (LTI) and single-input-
single-output (SISO) system, discussion of a Time Delay Con- u(t) = u(t-L) + (l/b){-xn(t-L)-nmnx„(t)-
trol system can be considerably simpler and more intuitive us-
ing classic frequency domain technique. In order to describe - a m l x , ( 0 + b,„r(0 + k „ ( x m „ - x „ ) + . . . + k , ( x , „ , - x , ) )
the main feature of this controller, this section will focus on
(3.5)
the analysis and design for such systems.
From the above equations (3.1), (3.3) and (3.5), the following
3.1 Control System Structure. Consider the following Laplace transformations are obtained for the plant, the model
LTI-SISO plant with unknown system matrix A.
and the control action as G(s), G,„{s), and U{s), respectively.
x = Ax + bw G{s)=Y{s)/U(s) = b/P(s) (3.6)
i
y=xi (3.1) v/hereP(s) = s"+a„s"- + . . .+a,

Journal of Dynamic Systems, Measurement, and Control MARCH 1990, Vol. 112/135

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Time D<alay Controller
b y e- P(s)
m m U(l): b
Pis)
m * ?T P(s)
-Ls
u(l u
-L)
Fig. 1 Block diagram of TDC for a LTI-SISO plant

Time Delay Controller

plant
b Pis)
m m I u • b

f
fr
b Pis)
P(s)
m ^\ Ls

Fig. 2 Reduced block diagram #1 of TDC for a LTI-SISO plant

model
e: P(s)
m : y^
P(s)
?: L 3 Pis) Fig. 4 Root locus of TDC for a LTI-SISO plant
model
Fig. 3 Reduced block diagram #2 of TDC for a LTI-SISO plant r b 1
y y
m m
P(s) 6 s* 1
Gm(s)=Y,„(s)/R(s) = bm/Pm(s) (3.7) m

where Pm(s) = s" +a,ms"~'+ . . .+aml Fig. 5 Reduced block diagram #3 of TDC for a LTI-SISO plant
Ls
U(s)= {Pk(s)Ym(s) + b„Ms)-Pmk(s)Y(.s)}/{b(l-e~ )} order system with a large bandwidth. The resultant block
(3.8) diagram is shown in Fig. 5. Since the second block in Fig. 5
can be regarded as almost unity, the whole system behaves as
where Pk(s) = k„s"~1 +. . . +k1s + kl the reference model. Performance properties of such control
systems for SISO systems which include stability, disturbance
Pmk(s) = e-Lss" + (a,„„+ kn)s"-> + . . . +(amX+k{)
rejection properties, and uncertainties in the control distribu-
Now we can describe the main feature of the TDC system in tion matrix could be found in reference [21].
frequency domain by analyzing the LTI-SISO systems defined From the above analysis, a Time Delay Control system can
above. be interpreted as one where the command r is prefiltered by
Suppose that the desired error dynamics is governed by the reference model into ym and then the error ym — y is forced
to zero by the high gain integrator l/Ls with pole/zero
e = A„,e (3.9) cancellation.
which simplifies the algebra by making all the error feedback 3.2 Stability Analysis. It is also possible to discuss the
gains to be zero. stability issues of this controller using the above formulation.
*,=. =0 (3.10) By combining equations (3.6) through (3.8), one obtains

In addition, for a small time delay L, the following approx- Y(s) _ b„ Pm(.s) + Pk(s)
L
imation holds. R(s) Pm(.s)\l-e- s)P(.s) + Pmk(.s)
x„(/-L)sx„(0 (3.11) Therefore, the characteristic equation of the closed loop
system is
Now combining equations (3.8), (3.10) and (3.11) leads to
Pm(s) {(1 - e-L°)P(,s) + Pmk(s) 1 = 0 (3.17)
U(s)= ibmR(s)-Pm(s)Y(s)}/{b(l-e-L°)} (3.12)
Usually, the model's characteristic equation Pm(s) is chosen to
or using equation (3.7), be stable, thus one is to examine:
U(s)= {Pm(s)Yn!(s)-Pm(s)Y(s))/{b(l-e-^)} (3.13) (l-e~L°)P(s) + Pmk(s) = 0 (3.18)
This equation can be reduced to or equivalently,
Ls
U(s) = e- U(s)+ [Pm(?)/b) I Ym(s)- Y(s)} (3.14) l + {-e-Ls + (Pmk(s)/P(s))}=0 (3.19)
A block diagram for the whole system can then be obtained as The stability condition of equation (3.18) or (3.19) can be
shown in Fig. 1. With the following approximation of solved by using existing techniques such as the Nyquist stabili-
ty criterion, or Routh-Hurwitz criterion using the Pade ap-
A-Ls (3.15)
proximation, and it gives the upper limit of the time delay L to
when L is sufficiently small, the block diagram of Fig. 1 can maintain the stability. Strictly speaking, the above stability
first be reduced to the one of Fig. 2 and then to the one of Fig. test. should be done for the entire possible range of the
3. unknown polynominal P(s). However, a good result can
A typical root locus of the closed loop part of the system usually be obtained using the "worst" P(s), which can be
(inside the dotted line of Fig. 3) is shown in Fig. 4. If the delay defined as the polynominal that corresponds to the "fastest"
time L is small enough, in other words, if the effective con- plant. Stability can also be interpreted intuitively using equa-
troller gain \/L is sufficiently large, one closed loop pole goes tion (3.18). If the time delay is infinitely small, L —0, then the
to negative infinity whereas all the other poles approach the overall stability will depend only on Pmk(s) and P(s).
fixed closed loop zeros. Therefore, pole/zero cancellation oc- As seen in equation (3.16), the stability condition depends
curs, which makes the whole closed loop part look like a first on the time delay, the error feedback gains, and the

136/Vol. 112, MARCH 1990 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 equation, in order to meet model following specifications,
which must be defined in terms of gain, phase and frequency
range. Again, strictly speaking, the above transfer function
should be evaluated for the entire possible range of the
unknown polynominal P(s) but a good result can usually be
obtained using the "worst" P(s). A simple example which il-
lustrates the design procedure can be found in [20].

4 Evaluation of Control System Performance

4.1 Simulations
lime(sec)
4.1.1 First Order Nonlinear System. Simulation was
Fig. 6(a) Response of the plant and the model done for the following first order plant to demonstrate the ef-
fectiveness of the Time Delay Controller.
x = cos(x) + u + d (4.1)
where cos(x) is treated to be the unknown dynamics of the
plant, and d = 0.5*l(/-0.5) is the unexpected disturbances
with magnitude 0.5 and acts at ^ = 0.5 s. Note that 1(0
represents a unit step function. The reference model to be
followed was selected to be a first order system with a time
constant of 1 second.
xm=-xm+r (4.2)
where r = l(t). The desired error dynamics was defined to be
Ume(3«
Fig. 6(b) Response of the error e=-aee (4.3)
where ae = 1. The TDC control law is obtained simply by com-
bining equations (4.1) through (4.3) and (2.18).
u(t) = u(t~L) + { -x(t-L)- x + r + kpe] (4.4)
where kp = ae-\. As seen in Fig. 6(a), the response of the
plant and that of the model match almost perfectly. In fact,
the error in Fig. 6(b) confirms this result. As seen in Fig. 6(c)
the Time Delay Controller completely cancels the step dis-
turbance right after it is applied at time / = 0.5 s.
Additional simulations were done for different error feed-
back gain kp's. The relationship between the error dynamics
liine(»«l and the error feedback gain is clearly seen in Fig. 7(a). The
Fig. 6(c) Response of the control input larger the gain is, the quicker the error dynamics is as
predicted by equation (2.7). The error dynamics actually has
characteristic equations of both the plant and the model. For the time constant of (\/ae)= 1, 1/2 or 1/6 second, which is
example, for a first order system, the stability condition with a shown in Fig. 1(b). However, as seen in Fig. 7(c), a larger con-
small time delay becomes as follows using digital implementa- trol effort is of course necessary for quicker convergence.
tion technique [17, 19].
4.1.2 Higher Order Nonlinear System. A robot
L<\/{(am + k)a] (3.20) manipulator, shown in Fig. 8, was chosen as a nonlinear
This stability condition states that a smaller time delay is re- system to control. This system can be described by the follow-
quired for a faster plant, a faster model and/or a larger error ing equations,
feedback gain. This general idea usually holds and intuitively
0 "
makes sense.
\6[] r<*ii
3.3 Design Procedure. As shown in equation (3.5), h h 'n'
design parameters to be chosen are (1) error feedback gains &, d_
through kn, which adjust error dynamics and (2) time delay L, — = — + (4.5)
which is related to model following performance. Substituting ~di
equations (3.3) and (3.4) into equation (2.7), the error system e\ h H ' j -r2 .
matrix can be obtained and its characteristic equation can be J2]
calculated as
s" + (am„+k„)s"-i+ . . . + (aml+kl) = 0 (3.21) where
Using the above equation, error feedback gains, k{ through
k„, can be chosen to give the desired poles of the error 2(5/3 + cos02), 2/3 + cos02,
dynamics. From equations (3.6) through (3.8), the following H = (l/2)*
transfer function is obtained and it is clear that we would like 2/3 + cos02, 2/3
to have it as close to unity as possible.
Y(s)/Ym(s) = {Pm(s) + Pk(s)}/{(l-e-L°)P(s) + Pmk(s)} (3.22) -smd2*e2*(2e\ + d\y
h = (l/2)*H-
A proper time delay L can be chosen by evaluating the above sin0,*0,2

Journal of Dynamic Systems, Measurement, and Control MARCH 1990, Vol. 112/137

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 //

unit mass and

unit length

Motor 2 .
Motor 1 *

limelKC)
Fig. 7(a) Response of the plant and the model

Fig. 8 Two degree-of-freedom robot manipulator

°l'°ml i. 8, IPDI
(rad)
9
* : (TDC), e m l

D
2'0ra2 l
B
(rad) 2 , T D U ' um2 v ^ - ^ ~ ~
timelKCl ^ ^ ** 82(PD)
Fig. 7(b) Response of the error
2'. 4'. 6'. •3

time (sec)
Fig. 9(a) Response of the manipulator and the model

e, ITDC)
(rad)

* e, IPD)
-«.»4
9'

e
2
(rad) e 2 (PD)
timelsei
Fig. 7(c) Response of the control input

time (set
In order to determine the effectiveness of the controller, the Fig. 9(b) Response of the error
Coriolis and centrifugal force vector h is treated to be com-
pletely unknown. The reference model was selected as a set of T
' ,
second order systems described by

0:10" ®m\ " 0 "

: 0 1 #M2 " 0 lime (sec)
+ 8.3
It "N
A
IrK B r «l
/ \ t 2 (TDC)
"ml mr 0/nl mr . 2 _
M 2 (PD)
e
0m2 . m2 . » 1
8 2'. 4'. 6'. a
where lime (sec)
Fig. 9(c) Response of the torque input
A = 2
-co„ 0 -2fw„ 0
(1/2)* ( l - c o s ( f (TT/4))) (rad), 0<t<4s
2
0 -co„ 0 -2faH
J (rad), ^4s
+ o,„ 2 0
For simplicity, let the desired dynamics be
:
0 +co„ e = Ame (4.7)
with a natural frequency w„ and damping ratio £ of 2.5 rad/s which corresponds to
and 0.707, respectively. The input commands r{ and r2 were
chosen to be Kr = 0 (4.8)

138/Vol. 112, MARCH 1990 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 °l°ml
(rad/sec2) (rad)

^8,(TDC).8 m l
^^ ^^Sw^ ^^^***^M(v

time (sec)
8
2- e .n2 i
(rad/secN
(rad)
• i * ^ ^ " ^ ^ W .^jy****^^ *• e2 (PD)

time (seel time (sec)

Fig. 10 Noisy acceleration measurement signal Fig. 11(a) Response of the manipulator and the model

, e , (TDC)

From, equations (4.5) through (4.8) and (2.18), the Time (rad)
Delay Control law becomes: /
/« e( IPD)
-9.14
T
i = H,*{H II
a. 12
e
2
T2 t-L t-L (rad)
« .as / \ *• e 2 <PD)
' e2(TDC) \
S,
+ A„ + B„

Fig. 11(b) Response of the error

'i i
The following PD controller was derived and used to compare
the results. Note that this PD controller does not compensate
for the Coriolis centrifugal force. *»MV4,+i"A'

= H/*{A„ + B„ (4.10) T
2 «,

7"2 'A

The first simulation was done under the above formulation

and with a delay of L = 0.01 second. With the Time Delay Fig. 11(c) Response of the torque input
Controller, the manipulator followed the desired model very
well as shown in Figs. 9(a). In fact, the maximum position
tracking error is considerably reduced to only some 1/50 of
that with the PD controller as shown in Fig. 9(b). Even though
the response chosen is slow, the nonlinear effects are signifi- Eiternal spring as
simulated unknown dynamics
cant with the PD controller [4]. The required torques for both
motors are satisfactory. Their magnitudes are not excessive
and the signals are very smooth as seen in Fig. 9(c).
In many practical situations, the available acceleration
signal could be very noisy. Such a situation is simulated by ap-
plying white noise with <j = 0.03 ( = about 10 percent of the
maximum signal) to both signals. The acceleration measure-
ment signals then became very noisy as shown in Figs. 10.
However, the robot manipulator still followed the reference
model fairly well as seen in Figs. 11(a) through 11(c). In fact,
the position tracking errors remained much less than that with
the PD controller. This is because the plant acts as a low pass
filter and considerably reduces the high frequency components
of the white measurement noise.
i t
Many physical plants, especially mechanical systems, can be Controller
considered as low pass filters similar to this case and then
noise problem may not be significant in the TDC systems. If ( AT&T PC 6300
this is not the case, however, some efforts to eliminate the Fig. 12 Experimental hardware
measurement noise are necessary in order to ensure good
tracking performance. Also, this noise insensitivity property
was confirmed by simulation with white noise in tachometers used to position a servo motor system as shown in Fig. 12. The
[18]. inherent unknown dynamics here are viscous and dry friction.
Also, additional uncertainty was simulated by an elastic spring
4.2 Experiments. In order to see the effectiveness of the that was physically attached to the motor load. Then, the
Time Delay Controller in a real system, the control law was dynamic equations for this system are

Journal of Dynamic Systems, Measurement, and Control MARCH 1990, Vol. 112/139

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 2' settling lime
- 0.8 sec

r - n/8 rad
e 8 r • ,,/8 rad

t
ro com pensated
spring torque
Top lrace: e <0.22 rad/div. ) Top trace: 8 ( 022 rad/div. )
Bouom lrace;t m< 0.068 Nm/div. )
Bottom lrace' ta/, 0.068 Nm/div )
HoriZ.lrace: lime ( O.S sec.ldiv. )
HorLt. trace: time ( O.S sec./div. )
Flg.13 Experiment #1 (PO controller)
Fig. 15 Experiment #3 (TOC with a soft spring)

2' sellling ti me
- 0.8 sec 2' sell ling lime
- 0.8 sec

r.,,/8rad 8 r - ,,/8 rad.'

compensated
spring torque
Fig.14 Experiment #2 (Time Delay Controller)
Flg.16 Experiment #4 (TOC with a hard spring)

d
dt
o
- (k s / J)*f) - (b/J)*O - dssgn(O)
J In the first experiment, the following PD controller was
used in order to see how large the unidentified viscous damp-

+ [0 J
1/J
7
m
ing coefficient bs and dry friction d s are,
7 m(t)=J{ -w,,2f}(t)-2sw ll 0(t) + w 1l 2r(t)} (4.16)
(4.11)
If the external spring is not attached and both b s and d s are
where f} is motor rotation angle, 7 m is motor torque, and k s is small enough, the above controller will make the plant behave
an unknown spring constant, bs is an unknown viscous fric- the same as the model. However, the actual response showed a
tion coefficient, d s is unknown dry friction and J( = 4.0* 10 large steady state error as seen in Fig. 13, which implies that
-4kg_m 2 = 0.53 oz-in-sec 2 ) is the total inertia. The reference the unknown viscous damping coefficient and/or dry friction
model was chosen to be a second order system defined by cannot be ignored and some compensation technique is
necessary. In order to compensate for these undesired and
d
unknown dynamics, the Time Delay Control law of equation
dt [ :: J [ (4.15) was used. This time, as shown in Fig. 14, the plant
showed no steady state error, 0 percent overshoot and 0.8 s for
where the natural frequency W n , damping ratio Sand reference the 2 percent settling time, which means that the plant closely
command rare 5 rad/s, 1 and 7l'/8 rad = 22.5 deg, respectively. followed the model. An external spring was then attached to
Let the error dynamics to be e=Ame thus making error feed- simulate additional unknown dynamics. Figure 15 (with a soft
back gains to be zero. Then, combining equations (4.11), spring of k s = 0.17 Nm/rad) and Fig. 16 (with a hard spring of
(4.12) and (2.18), one obtains the TDC control law: k s = 0.23 Nm/rad) show that the response remained almost the
same despite the unexpected external spring torque. Note that
7 m(t) =7 m(t- L) + J{ - O(t- L) - w2nf}(t) - 2sw lI O(t) + w2lI r(t) J
the controller does not depend on the unknown system
(4.13) parameters of k s ' d s and b s ' Thus, even if they change over
time, the same performance will be guaranteed. Additional ex-
Since an angular accleration signal was not available in the ex- periments with a reference model having a damping ratio of
perimental hardware, the following approximation was used, 0.5 can be found in [20]. All of these responses verify the good
O(t - L) =. (O(t) - O(t - L) J/ L (4.14) mocj.el following property of this controller. Since the
distinguishing feature of the Time Delay Controller is the
The final control law can be obtained by substituting equation angular acceleration term, the last experiment was done to see
(4.14) into equation (4.13). what would happen when the acceleration term is taken out of
the control law
7 m(t)=7 m(t-L)+J{ - [O(t)-O(t-L)]lL
(4.17)
- w2nf}(t) - 2sw n0(t) + w211 r(t) J
The above PID controller was applied to the system without
where the delay time Lis 5 ms. an external spring. As seen in Fig. 17, the system went into an

140/Vo1.112, MARCH 1990 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 References
I Arimoto, S., Kawamura., S., and Miyazaki, F., "Can Mechanical Robots
Learn by Themselves?" Proceeding of Second International Symposium on
Robotics Research, Kyoto, Japan, Aug. 1984.
9 r. ,,18 rad 2 Asada, H., and Youcef-Toumi, K., "Analysis and Design of a Direct-
Drive Arm with A Five-Bar-Link Parallel Drive Mechanism," ASME JOURNAL
OF DYNAMIC SYSTEMS, MEASUREMENT AND CONTROL, Vol. 106, No.3, Sept.
1984, pp. 225-230.
3 Astrom, K. J., and Wittenmark, B., "On Self-Tuning Regulators,"
Automatica, Vol. 9, 1973, pp. 185-199.
4 Brady, M., et al. Robot Motion: Planning and Control, M.I.T. Press,
1983.
5 Craig, J. J., Hsu, P., and Sastry, S. S., "Adaptive Control of Mechanical
Fig. 17 Ex'perlment #5 (PIO controller) Manipulators," Proceeding of the IEEE International Conference on Robotics
and Automation, Apr. 7-10, 1986.
6 Dubowsky, S., and DesForges, D. T., "The Application of Model
Referenced Adaptive Control to Robotic Manipulators," ASME JOURNAL OF
oscillatory mode due to the high gain integrator part of DYNAMIC SYSTEMS, MEASUREMENT AND CONTROL, Vol. 101, 1979, pp. 193-200.
Tm(t)=Tm{t-L). This result can be interpreted as that the ac- 7 Gutman, S., "Uncertain Dynamical Systems," IEEE Trans. Automatic
celeration term, or the higher derivative term in general, in the Control, June 1979, pp. 437-443.
8 Gutman, S., and Leitmann, G., "Stabilizing Control for Linear Systems
Time Delay Controller stabilizes the system with a high gain with Bounded Parameter and Input Uncertainty," Proc. 7th IFIP Conference
integrator. on Optimization Techniques, Nice, France 1975.
Overall, the control signals of the TDC were very noisy due 9 Hsia, T. c., "Adaptive Control of Robot Manipulators-A Review,"
Proceeding of the IEEE International Conference on Robotics and Automation,
to the numerical differentiation in equation (4.14). However, Apr. 7-10, 1986.
the system showed very good performance and this indicates 10 Ih, C. C., and Wang, S. J., "Dynamic Modelling and Adaptive Control
that use of higher derivative signals, with proper hardware, for Space Stations" JPL Publication, Vol. 85-57, July 1985.
should not be a very serious problem as also demonstrated in II Kalman, R. E., "Design of Self-optimizing Control," ASME Transac-
tion, Vol. 80, No.2, 1958, pp. 468-478.
[2]. Again this is because this servomotor worked as a lowpass 12 Spong, M. W., Ford, J. S., and Klenwaks, J., "The Control of Robot
filter and reduces the noise effect. Manipulators with Bounded Input, Part 2: Robustness and Disturbance Rejec-
tion," Proc. 23rd IEEE CDC Las Vegas, Nevada, Dec. 1984.
13 Siotine, J.-J.E., and Sastry, S. S., "Tracking Control of Nonlinear
Systems Using Sliding Surfaces with Applications to Robot Manipulators," In-
5 Conclusion ternational Journal of Control, Vol. 38-2, 1983, pp. 465-492.
14 Tomizuka, M., and Horowitz, R., "Model Reference Adaptive Control of
This paper discussed the control of systems with unknown Mechanical Manipulators," IFAC Adaptive Systems in Control and Signal Pro-
dynamics. A new method, Time Delay Control (TDC) , was cessing, San Francisco, CA, 1983.
proposed. The simple control algorithm does not require 15 Uchiyama, M., "Formulation of High-Speed Motion Pattern of a
Mechanical Arm by Trial," Transaction of Society of Instrument and Control
parameter identification, infinite frequency switching or
Engineering of Japan, Vol. 14, No.6, Dec, 1978, pp. 706-712.
repetitive action. Rather, it estimates the unknown dynamics 16 Utkin, V. I., "Equations of Sliding Mode in Discontinuous Systems,"
by performing a function evaluation every sampling time. It Automation and Remote Control 1(11), 1972.
was shown that the choice of the reference model and the error 17 Youcef-Toumi, K., and Ito, 0., "On Model Reference Control Using
Time Delay for Nonlinear Plants with Unknown Dynamics," M.I.T. Report,
feedback gain matrix is restricted depending on size of a con- LMPIRBT 86-06, June 1986.
trol distribution matrix. The simplified TDC control law was 18 Youcef-Toumi, K., and Ito, 0., "Controller Design for Systems with
given for systems in a canonical form. Unknown Dynamics," Proceeding of American Control Conference, Min-
Linear time-invariant (LTI) and single-input-single-output neapolis, MN, June 1987.
19 Youcef-Toumi, K., and Ito, 0., "Model Reference Control Using Time
(SISO) plants were discussed. System's structure was ex- Delay for Nonlinear Plants with Unknown Dynamics," Proceeding of Interna-
plained as the combination of (1) a prefilter of the model, (2) a tional Federation of Automatic Control World Congress, Munich, Federal
high gain integrator and (3) pole/zero cancellation. Stability Republic of Germany, July, 1987.
condition was given. In addition, a design procedure was ex- 20 Youcef-Toumi, K., Ito, 0., And Leung, Y. F., "Controller Design for
Systems with Unknown Nonlinear Dynamics," Final report, NSF/ENG-8801O.
plained on how to choose proper error feedback gains and 21 Youcef-Toumi, K., and Leung, Y.F., "Analysis of Time Delay Con-
time delay. trollers for SISO Systems," Proc. USA-Japan Symposium on Flexible Automa-
Performance of the control system was evaluated through tion, Minneapolis, Minn., July 1988.
both simulations and experiments. Through simulations for a 22 Young, K.-K.D., "Controller Design for a Manipulator Using Theory of
Variable Structure Systems," IEEE Transaction, SMC-8-2, 1978, pp. 101-109.
first order plant, good model following and very quick dis-
turbance rejection property were demonstrated. It was also
demonstrated that the error dynamics can be adjusted by
choosing a proper error feedback gain. The good model APPENDIX I
following property was confirmed through simulations for a
Algebraic Manipulation Leading to Equation (2.10)
more complicated nonlinear system-a robot manipulator. It
turned out that measurement noise did not cause a serious Subtracting equation (2.9) from equation (2.2),
problem. Then, the validity of the controller was verified
through experiments using a servomotor positioning system.
The focus of the current research include stability analysis + {I-BB+)(-f-h-d+Bmr} (Ll)
for nonlinear plants, discussion for cases with an unknown
control distribution matrix, and more detailed experimental Adding - Amx + Amx( = 0) to the right-hand side of equation
evaluation. (1.1 ),

Acknowledgments +BB+Ke+ {I-BB+} {-f-h-d+Bmr}

The authors gratefully acknowledge a fellowship support
from Matsushita Electric Industrial Co., Ltd. This research = Am(x m-x) + Ke +
was partly supported by the National Science Foundation
under contract No. MSM-8702839. {I-BB+} [-f-h-d+Amx+Bmr-Ke} (1.2)

Journal of Dynamic Systems, Measurement, and Control MARCH 1990, Vol. 1121141

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 By using the definition of the error in equation (2.3), equation [0 I] (H.2)
(2.10) is obtained:
e = (A,„e + K)e + ( I - B B + ) ( - f - h - d + A m x + B m r-Ke) I 0
0 0 (113)
(1.3)
or From equation (2.16),
e = (A m +K)e + ( I - B B + ) { - f - h - d + A m x + B„,r-Ke)
- f - h - d - A„,x + B m r - Ke
(2.10)
xs-[ 0 I„]x
- f r - h r - d - A m r x + B,„ r r-K r e.
A P P E N D I X II 0
fr-hr-d-A,
1
Canonical System and the Structural Constraint
rx + B m ,r-K,eJ
From equation (2.16) Substituting equations (II.2) and (II.3) into equation (2.11),
r the structural constraint becomes:
(B B)- [0 B/] 0 1
( I - B B + ) ( - f - h - d - A m r x + BmrKe)
B,
= (B/Br)-'=Br-'(B/)- (II.1) = I 0 (II-4)
0 0 • f r - h r - d - A m r x + Bmrr
Then,
=0
I-BB+=I-B(BrB) 'Br
1 B.-^B/)-1 [0 B/] Thus, the structural constraint is always met for systems ex-
pressed in canonical form.

142/Vol. 112, MARCH 1990 Transactions of the ASME

Downloaded From: http://dynamicsystems.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use