Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

On the Limitations of Force

Traci^ing Control for Hydraulic


Andrew Alleyne
Mem. ASME
alleyne@uiuc,edu
Servosystems^
This paper presents analysis of a particular force tracking control problem for recti-
Rui Liu linear hydraulic actuators governed by a servovalve. It presents no new theory, but
rather uses a revealing model reduction insight coupled with Classical analysis to
explain a physical phenomenon. As such, this work is an attempt to explain why a
Department of Mechanical and
Industrial Engineering,
seemingly innocuous problem is more subtle than initially believed. A motivation for
University of Illinois, Urbana-Champaign, this problem is given along with prior attempts at a simple solution. It is shown that
1206 West Green Street, simple controller solutions are quite adequate for other types of control objectives
Urbana, IL 61801 such as force regulation or position tracking. However, most simple solution methods
are shown to be inadequate for force tracking due to fundamental limitations of the
problem formulation. Due to an inherent feedback mechanism, the poles of the plant
being forced by the hydraulic actuator become zeros of the open loop force transfer
function. Therefore, more advanced control algorithms are shown to be a necessity
rather than a luxury.

Introduction Similar results were also confirmed by Ulsoy (1997) for a quar-
ter car active suspension test rig.
The motivation for this problem stems from the field of Ac- The main contribution of this paper is to explain, using simple
tive Suspension (Hrovat, 1997). For the purposes of this inves- analytical tools, the fundamental limitation imposed by the for-
tigation, a simplified version of the single degree of freedom mulation of the problem. This limitation dictates that simple
problem is given in Fig. 1.1. This problem has received signifi- feedback laws will be incapable of producing adequate force
cant attention (Yue et al., 1989), with the bulk of the effort tracking control results thereby necessitating alternative laws
going to determining the appropriate input, u, to provide to the (Alleyne and Hedrick, 1995; Vossoughi and Donath, 1995; Lee
system. The most common approach was the use of Linear and Srinivasan, 1991). Such a conclusion has been drawn em-
Quadratic Optimal Regulation. Since the road input was treated pirically by individual investigators (Ulsoy, 1997) but an ade-
as a disturbance, and the goal was to minimize the sprung mass quate explanation has not been formally presented in the litera-
velocity subject to rattlespace constraints, the system was a ture. This exposition of the problem will not consider the ap-
stationary, infinite horizon problem that led to an easy solution proaches of force regulation/tracking from the robotic literature
with fixed gain state feedback. Given the cost function (Patarinski and Botev, 1993). In most of these applications,

/ =
the optimal solution is
r {q\x'^ + q^v'^ + ru'^)dt
the environment is very stiff, with relatively low compliance.
(1.1) Additionally, the end effector has a contacting and free region
of operation which can lead to contact instability. In this investi-
gation, the actuator is always in contact with the load. The rest
of the paper is organized as follows. In Section 2, a model of
= —k^x — feu = —Kx (1.2) a hydraulic system will be introduced along with model valida-
tion. Section 2 also illustrates an effective model reduction
where K is the solution to an Algebraic Ricatti Equation. There which provides the key to a straightforward analysis of system
have been several variations on the basic method; however, in performance. Section 3 presents the performance analysis for
all of these approaches the desired system input as given in Eq. the force control problem and points out the inherent limitations.
(1.2) is & force. The assumptions were that either (i) there The analysis techniques are simple Classical Control ap-
existed an ideal actuator capable of producing the necessary proaches. Section 4 introduces the position control problem and
force instantaneously or (ii) any real actuator could easily be demonstrates that the performance barriers of the force control
controlled to track the necessary force. In Alleyne (1994), the problem are not present. Therefore, the position control problem
second assumption was examined for an electrohydraulic sys- can be easily satisfied by a simple controller, within the accu-
tem. It was assumed that a simple Proportional-Integral-Deriva- racy limits of the model reduction. Section 5 offers some conclu-
tive would be sufficient for the necessary inner force loop clo- sions. In particular, the successful experience of many research-
sure. Figure 1.2 shows the resultant experimental force tracking ers with position control of hydraulic systems led to erroneous
for a 1 Hz road disturbance. The results in Fig. 1.2 were obtained assumptions about the force control problem.
on the U.C. Berkeley Half Car Test Rig described in Yi and
Hedrick (1995). Despite a significant amount of effort, the PID
controller was incapable of producing the appropriate force and 2 System Model
had a significant phase lag which increased with frequency. Figure 2.1 depicts the experimental apparatus representative
of the problem to be studied in this work. This particular system
was designed and built by Moog Inc. The system consists of a
' This work was supported by NSF DMI96-24837CAREER and ONR N00014- servovalve mounted on a double ended cylinder. The cylinder
96-1-0754. is connected to a slide mass which then reacts against a spring
Contributed by the Dynamic Systems and Control Division for publication in
the JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND CONTROI,. Manuscript
load. The goal is to have the actuator track a specified reference
received by the Dynamic Systems and Control Division March 25, 1997; revised force trajectory. Ideally, the actuator's output of interest would
manuscript received April 1998. Associate Technical Editor; R. S. Chandran. be measured by a load cell attached to the piston cylinder rod.

184 / Vol. 121, JUNE 1999 Transactions of the ASME


Copyright © 1999 by ASME
Downloaded From: https://dynamicsystems.asmedigitalcollection.asme.org on 06/19/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use
v: inertial
velocity

x: relative
displacement

Fig. 1.1 One degree of freedom active suspension 0 0.5 1 1.5 1 2.5 3 3.5 4 4.5 5

time(a)

Fig. 1.2 1 Hz actuator force tracl(ing w/PID. --- Desired Force


The apparatus of Fig. 2.1 is equipped with cylinder differential Actuai Output Force
pressure sensors instead of a load cell. As will be detailed in
Section 3, however, it is possible to generate an accurate esti-
mated value of the force from the available measurements. mand with the system under proportional feedback. For the
A model for a servovalve controlled cylinder was developed actual system there is a slight deadzone in the servovalve that
in Alleyne (1996) and results in the following state equations shows in the figure but will be neglected in the analysis to
for the system, where the parameters are defined in the nomen- follow without detracting from the main focus. As is evidenced
clature. The reader is referred to the reference for details on the by the figure, the model is quite accurate for predicting experi-
model development. mental behavior.
For the purpose of this analysis, the model in Eq. (2.1) can
X{ — X2 be greatly simplified. Assuming the dynamics of the servovalve
load are sufficiently fast, the input variable becomes proportional to
X2 = — (—kxi — bx2 + Axi — Ft) the servovalve spool displacement from null. Assuming that the
actuator is sized properly for the load and the supply pressure
is high enough, the velocity of the actuator is proportional to a
. _ A/3, C,„A
first-order filter on the input. For the system in the nomenclature,
the corresponding filter is:
actuator .
CdPeW
H P-X4VP, - sgn {x^)x^ Velocity(.9) = — ^ Input(.v) = Input(.s) (2.2)
s +a s + 200
JC4 — X5 Figure 2.3 shows the output velocity versus input current for a

Xs = X^

UJIX4
2
2^a;„A:6 (2.1)
servovalve Xe = U)„X5
KfA,

The friction term in Eq. (2.1) is usually a stiction (Armstrong-


Helouvry et al, 1994) type of phenomenon due to the cylinder
seals. Depending on the type of actuator, the amplitude of the
friction can be quite small (Thompson and Chaplin, 1996) or
very significant. The model representing the friction was identi- spring load
fied in Liu (1998). Figure 2.2 shows the experimental and
simulation results for a square wave position reference com- Fig. 2.1 Experimentai apparatus

Nomenclature
A = piston area = 3.35e-4 m^ Ff = Coulomb friction = 120 Newtons Ky^ = feedback wire stiffness = 7.83 in-
a = hydraulic coefficient = (4A/3JV,) y = hydraulic coefficient = {AC^wPel lbs/in
= 1.513elO N / m ' V,) 'fUp = 1.545e9 N/(m"kg°-') m = load mass = 24 kg
A„ = spool end area = 0.0765 in^ k = load spring stiffnesses = 16,010 N/ P, = supply pressure = 1500 psi =
b = load viscous damping = 310 N/ m 10,342,500 Pa
(m/s) Kc — dQIdPi^ = pressure flow gain p = hydraulic fluid density
pe = effective bulk modulus Kf = net stiffness on armature/flapper = V, = total actuator volume
/3 = hydraulic coefficient = {AC„„PJ 115.0 in-lbs/in to,, = natural frequency of 1st stage =
V,) = 1.0 K,, = Hydraulic Amplification gain = 814 Hz
C,i = valve discharge coefficient 173.0 inVsec/in X = cylinder displacement
C„„ = total internal leakage coefficient K,, = dQ/dx^^i^^ = valve flow gain ^ = damping ratio of 1 st stage = 0.36
K, = torque motor gain = 0.010 in-lbs/
mA

Journal of Dynamic Systems, Measurement, and Control JUNE 1999, Vol. 121 / 185

Downloaded From: https://dynamicsystems.asmedigitalcollection.asme.org on 06/19/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


i: o reduced model

expeiimenlal data

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1


time isec)

Fig. 2.2(a) Spool position Fig. 2.4 Experimental model reduction

3 Force Control Analysis


Consider the LTI system shown below in Fig, 3.1 that repre-
sents a system in which the controller input signal is sent to an
actuator which then affects the plant. However, the output of
the plant is fed back to the actuator through some feedback
mechanism H(s). The transfer function from control input (u)
to actuator force ( / ) is not the same as the actuator transfer
function, G„(s).

Actuator F o r c e ( s ) G„(i)
Gfu(s) ^ (3.1)
Input Signal(s) + G„is)G,(s)His)
Fig. 2.2(b) Actuator position
Defining the actuator, plant and feedback numerator polynimals
Fig. 2.2 IVIodel validation. - - - (model) (experiment) as [Na{s), Npis), N,,{s)], and the denominator polynomials as
[Dais), Dpis), jDft(j)], the transfer function of Eq. (3.1) can
be rewritten as
5 Hz sinusoidal input signal simulation. Figure 2.4 shows the
experimental verification of the model reduction. The reduced N,{s)Dp(s)D„is)
Gf,(s) (3.2)
model is superimposed on the experimental data as well. Note D,{s)D„{s)D,(s) + N„(s)Np(s)N,(s)
that in Fig. 2.4, the scales are different from the simulation
results since the input signal is a voltage to an amplifier, not Equation (3.2) shows that the poles of the plant become the
the current. Additionally, there is a bias voltage required to keep zeros of the open-loop force transfer function irrespective of
the output of the mass position oscillating about an equilibrium the actuator transfer function. The plant poles would manifest
position. This explains the slight asymmetry in the normalized themselves as force loop zeros even if the actuator transfer
input about zero. The simplified model results in Fig. (2.4) are function was a pure gain, G^s) = 1, which indicates a perfect
also normalized and bias shifted to allow for a clear comparison. actuator. The relevance of Eq. (3.2) for hydraulic servosystems
Figures 2.3 and 2.4 justify the model reduction given in Eq. can be clearly seen by examining the actuator dynamics of Eq.
( 2 . 2 ) . The frequency range in which the reduced model retains ( 2 . 1 ) . The cylinder velocity acts as a feedback term from the
fidelity depends greatly on the the physical parameters of the position output of the cylinder to the pressure differential across
actual system. Based on observations, and given the parameters the piston. This is further illustrated in Fig. 3.2 which shows a
of the nomenclature, the model reduction of Eq. (2.1) into Eq. block diagram of Eq. (2.1) hnearized without valve dynamics
(2.2) is valid for frequency ranges up to 1 0 - 1 5 Hertz. and treating friction as a disturbance. Therefore, for a hydraulic
servosystem, the poles of the load will manifest themselves as
the zeros of the open loop force transfer function. This phenom-
enon can also be seen in the excellent work of Dyke et al.
(1995).
For the system given in Fig. 2.1, the force from the cylinder
is given as

actuator plant
u: control force y: output
input • O " Ga(s) Gp(s)
- ,

H(s)
•0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.
Input Current (A) feedback

Fig. 2.3 "ooo" Simplified model full model Fig. 3.1 Actuator-plant interaction

186 / Vol. 121, JUNE 1999 Transactions of the ASME

Downloaded From: https://dynamicsystems.asmedigitalcollection.asme.org on 06/19/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


leaku^c system which correlates well with the simulation parameters in
the Nomenclature.
cylinder As indicated in Fig. 3.3, the two open-loop zeros of the input-
"T displaccmcnl
ms^ +bs+k to-force transfer function are the poles of the system the actuator
iressihilily ,^ • ^ - • , * , , . is attached to. Due to pole-zero cancellation, the transfer func-
cylmdcr tion from input-to-output position, Gj.„ = X{s)/U(s), does not
velocity contain these zeros. As will be detailed shortly, these two open-
/ loop zeros limit the possible bandwidth of the closed-loop force
Feedback: H(s) system. If there is no damping in the system and the open loop
Fig. 3.2 Linearized dynamics transfer function of the actuator from input to position is defined
as Gact(^) then

mx + bx + kx * loacl^piston ^f (3.3) •^open loop (.V) = [ms^ + k]G^a(s) (3.6)


U(s)
where x is the displacement of the mass from equilibrium and
m includes the inertia of the actuator piston. The transfer func- Equation (3.6) implies that if the controller transfer function is
tion from force to position then becomes defined as Gds), the closed loop transfer function from a refer-
ence trajectory, R(s), to output force is
F(s) = [ms^ + hs + k]X{s) (3.4)
F{s) ^ [ms^ + k]G,,,(s)G,(s)
Using Eq. (2.2), and performing the integration from position -'closed loop (S)^ (3.7)
to velocity in the Laplace domain gives the transfer function R(s) 1 + [ms^ + k]G^,t(s)G,(s)
from input to output force for the reduced hydraulic system If the desired reference trajectory is A sin (a;*r) where uj* =
model. VA:/m, the closed-loop transfer function becomes invariant with
respect to any controller, provided (ms^ + k) and Gaci(i)Gc(i)
F{s) bs + k
(3.5) have no common factors. The corresponding transfer function
U(s) s{s + a) is Gcioscdioop(jw*) = 0. Essentially, due to the blocking zeros,
Figure 3.3 shows an experimental transfer function of the sys- none of the reference command information gets passed to the
tem in Fig. 2.1. Due to the sensing limitation of the system output. Equation (3.7) points out the importance of having some
(pressure instead of force), the data in Fig. 3.3 gives the fre- damping in the plant to avoid invariant frequencies.
quency response from input voltage to the right-hand side of Eq. While Eq. (3.7) points out a particular limitation in force
(3.3):P,„„,Ap Ff. To accomplish the frequency response, control of the system in Fig. 2.1, this work is interested in the
pressure and positions measurements were run through an ana- performance of a class of simple control strategies. Perhaps the
log signal conditioner supplied by Moog Inc. The conditioned simplest control strategy is a Proportional-Integral-Derivative
readings were then input to a 200 MHz Pentium PC via the A/ controller:
D input on an Analog Devices RTI 815 board. Using real-
time algorithms written in C, the PC used the pressure signals, U(s) K„ 1+IL + K^ E{s) (3.8)
multiplied by the piston area, and superimposed an estimated K,s K„ S + Td

friction force based on the velocity of the cylinder. The cylinder


velocity was determined via filtered numerical differentiation where E{s) is the force tracking error. There are two reasons
of the LVDT cylinder position measurement. For a sinusoidal for using a fihered derivative term in Eq. (3.8). First, the re-
input signal, the corresponding raw output "force" signal being duced open-loop system model would be otherwise noncausal
sent to the Hewlett-Packard Dynamic Signal Analyzer (DSA) and hence unrealizable. Second, the hydraulic system's pressure
via the PC D/A was initially too noisy to perform an accurate state variable has a very stiff dynamics due to the relatively
swept sine frequency response. Therefore, the PC performed low compressibility of the oil. Noise amplification caused by
additional digital signal processing on PioadAp Ff to smooth numerical differentiation of a force feedback signal can lead to
this out and give a relatively clean output to the DSA. It is instability. Figure 1.2 shows some of the effects of a derivative
possible that the on-line signal processing may influence the gain at the peaks of the force output. The derivative gain excites
"force" signal used in Fig. (3.3). However, the filters used higher frequency dynamics causing a slight chatter in the force
were all of a sufficiently high bandwidth ( > 3 0 Hz) so that output. If the gain is increased, or the filter removed, it is very
the low frequency ( < 1 0 Hz) portion of the DSA response is easy to destabilize the system.
sufficiently accurate. The figure indicates the open loop zeros Using Eqs. (3.5) and (3.8), the open-loop transfer function
at the damped natural frequency ( ~ 4 Hz) of the mechanical becomes

^ ^ V
• ! X ;:nK^-
__|y^ . : _ ^ \ Olu

PKU r\\

Fig. 3.3 Experimental transfer function

Journal of Dynamic Systems, Measurement, and Control JUNE 1999, Vol. 121 / 187

Downloaded From: https://dynamicsystems.asmedigitalcollection.asme.org on 06/19/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


O.I., zaro
; o

.0

: \ : J
10 : I
l. O.L pole ; Kp»9a-5 :;

• 10

•20
:1
; o

-200 -ISO -160 -140 -120 -100 -60 -40 -20


llme(sec)

Fig. 3.4 Root locus for P-control Fig. 3.6(a) Kp = 4.0e-5

ms + bs + k
-'Open loop (.s)
s{s + a)
Gfuis)

kp
If, ^A 2 , ( Ki\ ^i 1
[l + -~]s^ + Td + — \S + — Ti
L\ KpJ \ Kp) Kp J (3.9)
S{S + Trf)

G,{s)
The closed-loop performance of this system can be readily
checked by an Evans Locus analysis with varying values of Kp
using the parameters given in the nomenclature. For the case Fig. 3.6(b) K^ = 9.0e-5
of proportional control, the locus is given in Fig. 3.4. As can
be seen in the figure, the two open-loop zeros at —bl2m ± Fig. 3.6 Experimental output force tracking
j^\b'^ — 4mk\ 12m cause a lightly damped pole pair as the gain
is increased. If there is no damping in the system of Eq. (2.1),
the open-loop zeros lead the reduced order system to be only Figure 3.6 experimentally verifies the conclusions obtained
marginally stable at higher gains. Figure 3.5 shows a simulated by analysis and simulation. Since the actual system shown in
time response for force control of the full nonlinear system in Fig. 2.1 can provide only a compressive spring load, the data
Eq. (2.1) at two different gains: less than and greater than the shown in Fig. 3.6 has a 1000 N offset in addition to the 100 N
predicted breakaway point {Kp = 9.0e-5) on the locus of Fig. square wave. As with Fig. 3.3, the force output used in the
3.4. The low feedback gain value is Kp = 5.0e-5, corresponding feedback was observed from other data rather than directly
to the slower response; the high value is Kp = 15.0e-5 corre- measured. Unlike the frequency response of Fig. 3.3, the step
sponding to the faster response with a slight overshoot. The responses of Fig. 3.6 resulted in very low velocities close to
dashed line reference signal for the output forces of Fig. 3.5 is steady state. Therefore, the use of an estimated friction in a
a 6 rad/s, 100 Newton square wave that is passed through a force calculation scheme was not acceptable. Instead, the left-
first order filter {50/(j' + 5 0 ) } . The filtering is included to hand side of Eq. (3.3) was used to estimate force.
prevent the excitation of higher frequency dynamics of the sys-
tem caused by the pressure variation in the cylinder. Some of mx + bx + kx (3.10)
these can be seen at the beginning of the transients in Fig. 3.5.
As evidenced in the figure, a simplified analysis predicts the The acceleration was first determined via a Sensotec JTF 3629
overshoot behavior in the complete nonlinear system caused by accelerometer and bandpass filtered. However, the signal to
the higher gain. noise ratio was too low to be useful. Therefore, both the acceler-
ation and velocity terms in Eq. (3.10) were determined by stably
filtering the measured position. The critical gain required for a
step response overshoot is not exactly the same as the model
due to some slight model uncertainty and there are also some
unmodeled dynamics that show up in the higher gain case.
However, the main characteristics are quite clear.
Figure 3.7 shows the locus for a PID controller. The gains
were chosen such that K/Kp = 10, K^/Kp = 0.1, and T^ =
100.0. The gains could be chosen to affect the locus shape and
achieve some measure of pole placement but arbitrary pole
placement is not possible. Using Eq. (3.9) and defining G^s) =
N^{s)/Dc{s) and Gp{s) = Np(s)/Dp(s) the closed-loop transfer
function is written as

Npis)N,(s)
•^closed loop (s)=- (3.11)
tim6(S9C) Dp(s)DAs) + Npis)NAs)

Fig. 3.5 Simulated output force tracldng where the characteristic polynomial is given as

188 / Vol. 121, JUNE 1999 Transactions of the ASME

Downloaded From: https://dynamicsystems.asmedigitalcollection.asme.org on 06/19/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


X(s) _ a
O.L zero
o
(4.1)
U(s) s{s + a)
The open-loop transfer function of the system with a simple
( PID controller is
\
O.L. pole

K.,\ ^ s ' + s +
1
^openloop (.5) (4.2)

: H.) { a
s(s + a)
In contrast to Eq. (3.8), there is no need to filter the derivative
term here for either causality or stability since the position
dynamics aren't as stiff as the pressure dynamics. As is evident,
the choice of the gains of the PID controller determines the
Fig. 3.7 Root locus for PID control open loop zeros for the position control problem. Within the
accuracy of the model reduction, they can be chosen to be as
far in the open left half plane as desired to achieve necessary
bandwidth. In fact, arbitrary pole placement can be achieved
D,{s)DXs) + N,(,s)NAs) = A(^) (3.12) with the open-loop system of Eq. (4.2). In this case, the desired
For arbitrary closed loop pole placement of Eq. (3.9), the de- characteristic polynomial is a third-order Hurwitz polynomial.
sired characteristic polynomial is a fourth-order Hurwitz poly-
A(s) = .y' + 6,s- + 62S + 63 (4.3)
nomial.
A(.«) = s^ ^ i5,j^ -I- bts'^ + S^s + 6^ (3.13) Absorbing the high frequency gain, a, into the controller gains,
the closed-loop denominator polynomial is
Substituting Eq. (3.9) into the right-hand side of Eq. (3.12),
and absorbing the high frequency gain a into the controller D„{s)DAs) + N,(s)N,(s)
gains, gives = .?' + s^a + K,,) + sK„ + K, (4.4)
D„(s)D,(s) + N,{s)N,(s) = s\\ + m{K„ + K,)) Equating the right-hand sides of Eqs. (4.3) and (4.4) gives a
+ s\{a + T,,) + m{K,,T,, ^ Kt) + b(K„ + K,,)) set of three equations with three unknowns which can easily be
solved for K^, /f,, and K^i. This type of feedback algorithm is
+ s\aTa + mKiT,, + b{K„T^ + K,) + k(K,, + K^)) therefore quite suitable for basic position tracking objectives.
Additionally, similar conclusions can be reached with simple
+ s{bK,Ta + k{K„Ta + K,)) + kK,Ta (3.14) lead-lag controllers. Naturally, more advanced techniques
Equating the coefficients of Eqns. (3.13) and (3.14) gives a would be required to compensate for some of the nonlinearities
system of five equations with four unknowns: K,,, K/, K,i, and and dynamics that are neglected in the model reduction: e.g.,
T,i. Therefore, a solution does not exist. In general, a controller seal friction or the deadzone shown in Fig. 2.2. This would
with second-order numerator and denominator polynomials be particularly true for very precise position control at higher
could theoretically be chosen to arbitrarily place the poles of bandwidths (Lee and Srinivasan, 1991; Tsao and Tomizuka,
Eq. (3.11). However, the constraint of having one of the con- 1994). However, a vast majority of position tracking objectives
troller poles at the origin (integrator) and the other on the can be met with simple controller algorithms.
real axis (derivative filter) removes this ability for the PID.
Additionally, Eqs. (3.9) and (3.14) are based on the reduced 5 Conclusions and Discussions
order model of Eq. (2.2). There is a pole at the origin associated
with the natural integration of the hydraulic cylinder from flow This investigation develops no new theory. However, it does
to position. To move this pole far into the OLHP in the presence use common analytical tools and an effective model reduction
of the two lightly damped zeros would be very difficult and to gain new insight into the formulation of a particular problem.
may requires a large loop gain for any simple transfer function From the results presented here, it is quite clear that simple
based controller. Experimentally, and even in simulation, large algorithms are an inappropriate choice for a system such as in
gains excite the unmodeled dynamics of the full system repre- Fig. 2.1 if the desired objective is to track a specific force
sented by (2.1) and easily result in force loop instability. profile containing significant frequency components. Due to the
unavoidable feedback of piston velocity to actuator pressure,
The system of Eq. (3.9) and Fig. 3.7 is obviously stable since the force transfer function open loop zeros are a result of the
all the poles remain in the open left half plane and will therefore system attached to the actuator. These can be lightly damped
be able to regulate a constant reference force. However, there or even nonminimum phase zeros if the actuator is connected
are still fundamental bandwidth limitations. The force transfer to an unstable system. These zeros can't be changed by feedback
function open loop zeros dictated by the plant poles limit the and the simple algorithms are severely bandwidth limited. The
effectiveness of simple feedback controllers. Moreover, these problem becomes compounded if there exist multiple degrees
zeros can't be moved via feedback. Therefore, for tracking pur- of freedom in the plant attached to the hydraulic servosystem;
poses beyond the bandwidth limitation a simple type of algo- for example a 2 DOF | car active suspension (Hrovat, 1997).
rithm (i.e., PID, lead-lag) is not suitable. Each new mode adds an additional pair of zeros that may be
lightly damped. Therefore, more advanced algorithms are a ne-
4 Position Control Analysis cessity, not just a luxury. The position tracking problem does
not suffer from such ill-posedness since stable pole-zero cancel-
The bandwidth limitations caused by open-loop zeros for the lation eliminates the plant poles from appearing in the position
force tracking problem are not applicable to the position open loop transfer function. Since a servovalve/cylinder is ap-
tracking problem. As shown in Fig. 3.3, stable pole-zero cancel- proximately flow causal it can behave as a velocity source.
lation occurs in the input to position transfer function. Using Therefore, its position can be controlled independent of the
Eq. (2.2) the reduced order transfer function from input to dynamics of its environment; not so with force control. For
position is

Journal of Dynamic Systems, Measurement, and Control JUNE 1999, Vol. 121 / 189

Downloaded From: https://dynamicsystems.asmedigitalcollection.asme.org on 06/19/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use


many position tracking problems with a hydraulic servosystem, Dyke, S. J., Spencer, B. F., Quast, P., and Sain, M. K., 1995, "Role of Control-
Structure Interaction in Protective System Design," Journal of Engineering Me-
the issue of bandwidth is not a significant concern with simple chanics, ASCE, Vol. 121, No. 2, pp. 322-338, Feb.
feedback controllers within the physical limitations of the actua- Hrovat, D., 1997, "Survey of Advanced Suspension Developments and Related
tor/servo valve and the accuracy of the reduced model. It is this Optimal Control Applications," Automatica, Vol. 33, No. 10, pp. 1781-1817,
success with position tracking of hydraulic servosystems that Lee, S. R., and Srinivasan, K., 1991, "Self-Tuning Control Application to
Closed-Loop Servo-Hydraulic Material Testing," ASME JOURNAL OF DYNAMIC
led some previous efforts into force tracking to start with errone- SYSTEMS, MEASUREMENT, AND CONTROL, Vol. 112, No. 4, pp. 680-689, Dec.
ous assumptions. Liu, R., 1998, "Nonlinear Control of Electrohydraulic Servosystems: Theory
and Experiment," M.S. thesis, Dept. of Mech. & Ind. Eng., Univ. of Illinois,
Urbana-Champaign, Sept.
Patarinski, S., and Botev, R., 1993, "Robot Force Control, a Review," Mecha-
References tronics. Vol. 3, No. 4, pp. 377-398.
Thompson, A. G., and Chaplin, P. M., 1996, "Force Control in Electrohydraulic
Alleyne, A., 1994, "Nonlinear and Adaptive Control of Active Suspensions," Active Suspensions" Vehicle System Dynamics, Vol. 25, pp. 185-202.
Ph.D. dissertation. University of California, Berkeley. Ulsoy, G., 1997, Personal Communication, Seattle, WA, Jan.
Alleyne, A., 1996, "Nonlinear Force Control of an Electro-Hydraulic Actua- Vossoughi, R., and Donath, M., 1995, "Dynamic Feedback Linearizadon for
tor," Proceedin/^s of the 1996 Japan/USA Symposium on Flexible Automation, Electro-hydraulically Actuated Control Systems," ASME JOURNAL OF DYNAMIC
pp. 193-200, Boston, MA, June. SYSTEMS, MEASUREMENT, AND CONTROL, Vol. 117, No. 4, pp. 468-477, Dec.
Alleyne, A., and Hedrick, J. K., 1995, "Nonlinear Adaptive Control of Active Yi, K., and Hedrick, J. K., 1995, "Observer-Based Identification of Nonlinear
Suspensions," IEEE Transactions on Control Systems Technology, Vol. 3, No. System Parameters," ASME JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT,
1, Mar., pp. 94-102. AND CONTROL, Vol. 117, No. 2, pp. 175-182, June.
Armstrong-Helouvry, B., Dupont, P., and Canudas de Wit, C , 1994, "Friction Yue, C , Butsuen, T., and Hedrick, J. K., 1989, "Alternative Control Laws
in servo machines: Analysis and control methods," Applied Mechanics Reviews, for Automotive Active Suspensions," ASME JOURNAL OF DYNAMIC SYSTEMS,
Vol. 47, No. 7, pp. 275-305, July. MEASUREMENT, AND CONTROL, Vol. H I , pp. 286-291, June.

190 / Vol. 121, JUNE 1999 Transactions of the ASME

Downloaded From: https://dynamicsystems.asmedigitalcollection.asme.org on 06/19/2019 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like