Classical Cascade and Sliding Mode Control Tracking Performances For A Xy Feed Table of A High-Speed Machine Tool

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Int. J. Precision Technology, Vol. 1, No.

1, 2007 65

Classical cascade and sliding mode control tracking


performances for a xy feed table of a high-speed
machine tool

Z. Jamaludin,* H.V. Brussel and J. Swevers


Department of Mechanical Engineering,
Katholieke Universiteit Leuven,
Celestijnenlaan 300B, Heverlee B3001, Belgium
E-mail: zamberi.jamaludin@student.kuleuven.be
E-mail: hendrik.vanbrussel@mech.kuleuven.be
E-mail: Jan.Swevers@mech.kuleuven.be
*Corresponding author

Abstract: The development of robust tracking controllers for linear feed drive,
high speed machine tools is the primary objective of this work. Robustness
implies the controller ability to preserve tracking accuracy in the presence of
cutting forces and friction forces. This paper discusses the design and tracking
performances of traditional cascade and sliding mode controllers for a xy feed
table. Tracking performances are analysed through circular tests that are
performed at selected tracking speeds and circle radius. For cascade P/PI
controller, tracking error is proportional to the reference speed and is inversely
proportional to the velocity gain value of the position loop controller. Speed
and acceleration feed forward further reduce this tracking error. Quadrant
glitches, a product of non-linear friction phenomena at the point of velocity
reversal, are observed. Sliding mode controller, however, significantly reduces
quadrant glitches and exhibits greater dynamic stiffness while its tracking
accuracy is comparable to the cascade controller.

Keywords: linear drive; robustness; cascade controller; sliding mode;


quadrant glitch.

Reference to this paper should be made as follows: Jamaludin, Z.,


Brussel, H.V. and Swevers, J. (2007) ‘Classical cascade and sliding mode
control tracking performances for a xy feed table of a high-speed machine
tool’, Int. J. Precision Technology, Vol. 1, No. 1, pp.65–74.

Biographical notes: Z. Jamaludin is a PhD student in the Department of


Mechanical Engineering, Division Production, Machining and Automation
(PMA) at Katholieke Universiteit Leuven, Belgium. He received an ME from
National University of Malaysia (UKM).

H.V. Brussel received a PhD from Katholieke Universiteit Leuven, Belgium.


He works as a Professor at the Department of Mechanical Engineering,
Division Production, Machining and Automation (PMA) at Katholieke
Universiteit Leuven, Belgium. His current research interests are in the areas of
mechatronics system (motion, mechatronic compiler), robotics (medical,
mobile manipulation) and holonic manufacturing systems.

J. Swevers received a PhD from Katholieke Universiteit Leuven, Belgium.


He works as a Professor at the Department of Mechanical Engineering,
Division Production, Machining and Automation (PMA) at Katholieke

Copyright © 2007 Inderscience Enterprises Ltd.


66 Z. Jamaludin, H.V. Brussel and J. Swevers

Universiteit Leuven, Belgium. His research interests are in the fields of


modelling and identification of mechatronic systems, optimisation and
advanced control of complex systems.

1 Introduction

Intense global competition has increased pressure on manufacturers to attain ever higher
productivity and produce quality products at competitive prices. High-speed machining is
attractive in a sense that it creates good potential for higher productivity with its shorter
machining time. The overall machining time is reduced with increases in both table
positioning speed and acceleration. However, high-speed machining creates another
challenge, that is, machining process requires the feed drive to be robust against fast
acting disturbances (Pritschow, 1998).
Motion tracking controllers are designed with the fundamental objective of attaining
the best possible position tracking accuracy. This, however, is significantly influenced by
friction forces acting closely at the load and cutting forces from the machining process.
One prominent tracking controller that exists in majority of servo motion control systems
is classical cascade controller. Various improvements and modifications to this controller
have been suggested in the literature with the aim of improving the tracking
performance. For example, Doenitz (1999) analysed the tracking error of combined
cascade controller with a proposed disturbance observer. Pritschow (1996) on the other
hand, presented different positioning behaviours between classical P/PI cascade
controller and reduced state space controller. In addition to classical cascade controller,
sliding mode controller, a well known non-linear controller, is highly regarded for its
disturbance rejection properties. Altintas (2000) has demonstrated the contouring
performance superiority of this controller against pole placement controller with feed
forward friction and servo dynamics compensation.
The primary objective of this work is to analyse motion tracking performance of
linear feed drive high speed machine tools of an xy milling table. Tracking performance
of a non-linear sliding mode controller is analysed and compared to the traditional and
more popular cascade P/PI controller. Comparison in tracking performance is based on
axial tracking errors, contour tracking errors, dynamic stiffness and quadrant glitches.
Series of circular tests were performed at various tracking speeds and circle radii.

2 Basic design of tracking controller

2.1 Cascade controller


Classical cascade controller (see Figure 1) consists of two distinct loops; an inner speed
loop and an outer position loop.
The closed-loop bandwidth of the inner speed loop is higher than the outer position
loop and this ensures that the dynamics delay of the inner loop is ignored by the outer
position loop (Wang, 2004). The controller is designed-based on traditional open-loop
shaping and is constructed and tuned from the speed loop to the position loop.
A proportional (P) plus integral (I) control forms the speed loop. The controller
Classical cascade and sliding mode control tracking performances 67

parameters are designed from the knowledge of gain margin and phase margin of the
resulting speed open-loop transfer function that is based on the actual measurement of
open-loop frequency response function of the system (see Figure 2).

Figure 1 Schematic diagram of P/PI cascade controller

Figure 2 Theoretical (a) open-loop and (b) closed-loop Bode diagram of the y-axis speed loop
with PI controller

The phase margin and the gain margin indicate the system stability margin and its
transient response (Franklin et al., 2002). A proportional controller completes the
position loop. The proportional gain was selected based on the gain margin and the phase
margin of the position open-loop transfer function. The stability of each loop is
confirmed by the Nyquist plot of the open-loop transfer function. The following transfer
functions relate input disturbance, d(t), to output position, y(t) and reference signal, r(t),
to tracking error, e(t). kf is the motor constant, kv, kp and ti (see Figure 1) are the controller
parameters and m is the mass of the system.
d (s ) mti s + k f k p ti s + (1 + kv ti ) k f k p s + k f k p kv
3 2

= (1)
y( s ) ti s

e(s ) ms 2 + k f k p s
= (2)
r (s ) ms 2 + k f k p s + k f k p kv

The transfer function given in Equation (1) defines the dynamic stiffness of a system.
It indicates the controller disturbance rejection capacity. For a constant velocity
reference input signal, r(t) = v(t), tracking error is proportional to the tracking velocity
but is inversely proportional to the velocity gain factor, kv;
v
e(∞) = (3)
kv
68 Z. Jamaludin, H.V. Brussel and J. Swevers

2.2 Sliding mode controller


Sliding mode control is a form of Variable Structure Control (VSC). There are two
design components of sliding mode control; namely, switching function and control law.
Switching function determines the dynamics transient response of a system. For a
second order system, the sliding surface, s, is a function of the tracking error, e(t) and its
time derivative
n −1
⎛d ⎞
s(e, e) = ⎜ + λ ⎟ e; e(t ) = y(t ) − r (t ) (4)
⎝ dt ⎠

For a second order system, n = 2


s(e, e) = e + λ e (5)

where, y(t) is the actual position, r(t) the desired position and λ a positive constant.
The choice of λ determines the transient response characteristics of the system.
Control law, on the other hand, consists of a signum function and the equivalent
control which includes speed and acceleration feed forward:

u(t ) = uequivalent − K ⋅ sign(s) (6)

At the moment of sliding, the switching function is equivalent to zero and hence;
s(e, e) = 0 (7)

Therefore,
s(e,  e + λ e = 0
e ) =  (8)

with
e = y − r
(9)
e = 
 y − r

Hence, for a simple mass-damper-spring system,


my + cy + ky = u (10)

without the disturbance function, d(s), and by combining Equations (8)–(10), the control
input becomes:

uequivalent = m(r − λ e) + cy + ky (11)

Equivalent control maintains the sliding motion. The discontinuous signum function
(see Equation 6) with proportional gain K ensures that the states are attracted to
and remain in the sliding surface in finite time. This discontinuous signum function,
however, leads to high frequency switching. The states are oscillating about the surface
instead of remaining on them. This is widely known as chattering (Young et al., 1999)
and is a major disadvantage to sliding mode control application. Several techniques have
been suggested to reduce the effect of chattering (Utkin and Guldner, 1999).
A continuous approximation of the discontinuous signum function is among the options.
Classical cascade and sliding mode control tracking performances 69

A sigmoid-like function of the following form replaces the discontinuous signum


function (Spurgeon and Edwards, 1998)
s
Vδ (s ) = (12)
| s | +δ

The positive constant, δ illustrates the degree of the continuous approximation.


A comparison between the discontinuous signum function and its continuous
approximation is shown in Figure 3. Instead of having an ideal sliding motion, the states
lie within a specified boundary of the sliding surface resulting in non-ideal sliding
motion, also known as pseudo-sliding.

Figure 3 Continuous smooth approximation of the signum function

3 Experimental setup

The controllers’ performance were validated on a xy feed table of a high-speed milling


machine (see Figure 4). The upper top stage (y-axis) is driven by a single ETEL linear
motor while the bottom stage x-axis, is driven by two ETEL parallel linear motors. Two
Heidenhain linear encoders provide the table two axes relative position measurements.
In the absence of a speed sensor, the speed signal is generated from the derivative of the
encoder position measurements. The derivative is combined with a first-order low-pass
filter to attenuate amplified measurement noise associated with the derivative action.
The controllers of both axes were implemented on a dSPACE 1102 DSP controller board
linking the host computer to the three ETEL drives.

Figure 4 A xy feed table with three linear drives for high speed milling application

Three different controllers were implemented:


1 a traditional cascade P/PI controller
2 a cascade P/PI controller with speed and acceleration feed forward
3 a sliding mode controller.
70 Z. Jamaludin, H.V. Brussel and J. Swevers

A series of circular test was performed to validate and compare tracking performance of
these controllers. The controllers were designed to track circles of 15 and 30 mm radii,
each at constant tracking speed of 0.001 m/s, 0.005 m/s and 0.01 m/s. Each axis
position tracking error was recorded. The actual trajectory was compared to the reference
trajectory. The system was not subjected to any external disturbance signal. In addition to
tracking, the dynamic stiffness of the cascade and sliding mode controllers were
investigated and compared. A band-limited white noise signal was introduced into the
system as an input disturbance and the position error was recorded.

4 Experimental results

The controllers’ performance was assessed through circular tests and dynamic stiffness
measurements. The circular tests reveal the axial and contour tracking performance while
dynamics stiffness measures the disturbance rejection property of the controllers.

4.1 Dynamic stiffness


Dynamic stiffness is defined as the transfer function from output position to disturbance
force. For high-speed milling, high machine dynamic stiffness is desired. The milling
table experiences minimum position tracking error from the influence of the cutting
force. The dynamic stiffness measurements of the P/PI cascade controller and the
sliding mode control are presented in Figure 5. The position tracking error was recorded
as the system was subjected to a band-limited white noise signal. Figure 5 indicates that
for both axes, sliding mode controller provides better stiffness than cascade controller
over the range of the system’s bandwidth, which is 40 Hz. However, cascade controller
displays higher stiffness (until about 4 Hz) at much lower frequency range. This is due to
the integrator within the cascade speed loop controller. The results show that sliding
mode has better disturbance rejection capacity than cascade controller from the
frequency range of 4 Hz until the bandwidth of the system.

Figure 5 Dynamics stiffness comparison of cascade controller and sliding mode


controller for (a) x-axis and (b) y-axis
Classical cascade and sliding mode control tracking performances 71

4.2 Tracking error


A motion controller with high tracking accuracy has its output position closely tracks the
input reference position. Tracking error is categorised into axial tracking error and
contour tracking error. Axial tracking error indicates the individual axis tracking
performance while contour tracking error describes both the x and y axes tracking
performance. Figure 6 illustrates the axial tracking error of the controllers for circular
tests with a radius of 15 mm, executed at different tracking speeds. As indicated in
Equation (3), the tracking error increases with increases in tracking speed. Speed and
acceleration feed forward reduce significantly these errors. For example, at 0.005 and
0.01 m/s, feed forwards reduce the tracking error with nearly a factor of two. Sliding
mode yields the smallest tracking errors. The tracking errors are nearly half of that of the
cascade controller with speed and acceleration feed forward. Similar observations and
conclusions are made for circular tests with other radii. The tracking errors for a circle
with a 30 mm radius are summarised in Table 1.

Figure 6 Axial tracking error with (a) cascade, (b) cascade with speed and acceleration feed
forward and (c) sliding mode for a 15 mm radius circle tracking test at
different speeds
72 Z. Jamaludin, H.V. Brussel and J. Swevers

Table 1 Maximum tracking error for circle of radius 30 mm with varying tracking speed for
(a) cascade, (b) cascade with speed and acceleration feed forward and (c) sliding
mode

Tracking speed/controller 0.001 m/s 0.005 m/s 0.01 m/s


Cascade 15 µm 45 µm 80 µm
Cascade + speed + acceleration feed forward 15 µm 25 µm 40 µm
Sliding mode 10 µm 15 µm 20 µm

In comparison to Figure 6, it is observed that there was no significant difference in the


magnitude of the tracking errors with doubled circle radius. Tracking error depends on
the tangential velocity of the reference trajectory and not the dimension of the circle.

4.3 Quadrant glitch


Quadrant glitch is a special form of contour tracking error. It is characterised by the
presence of ‘spikes’ in every quadrant of a circle, as illustrated in Figure 7. This
phenomenon arises from complex non-linear friction occurring at the point of motor
reversal motion. Sliding mode control is widely known for its robustness properties
against matched uncertainties and disturbance (Wang et al., 2003). Quadrant glitches can
also be clearly recognised from the axial tracking error measurements. Figure 8 compares
the axial tracking error of cascade controller, cascade with speed and acceleration feed
forward and sliding mode control.

Figure 7 Quadrant glitches as seen from circle contour test

Figure 8 Spikes as seen from axial tracking error for (a) cascade, (b) cascade with speed
and acceleration feed forward and (c) sliding mode controller
Classical cascade and sliding mode control tracking performances 73

Spikes are clearly visible from the axial tracking errors of both the cascade controllers.
They are significantly reduced with the sliding mode controller. The axial tracking errors
indicate the advantage of sliding mode in disturbance rejection of friction forces. Similar
observation and conclusion can be derived from the contours of the circular tests as
shown in Figure 9. The actual and the reference contours of the cascade controllers and
the sliding mode controller, for a circle test of radius 30 mm at a tracking speed of
0.001 m/s, are compared. The actual contours were reconstructed with amplified radial
error (200 times) for graphical presentation purposes. Spikes are clearly visible with both
the cascade controllers.

Figure 9 Quadrant glitches for circle test of radius 30 mm and tracking speed of 0.001 m/s for
(a) cascade, (b) cascade with speed and acceleration feed forward, and (c) sliding
mode control
74 Z. Jamaludin, H.V. Brussel and J. Swevers

The radial errors at the quadrant positions for the cascade controllers is within
±10 micrometre while the sliding mode controller has radial error within a range of
±5 micrometre.

5 Conclusions

• sliding mode controller exhibits higher dynamic stiffness and better tracking
accuracy than the two classical cascade controllers
• speed and acceleration feed forward improve tracking accuracy of classical
cascade controller
• the classical cascade controllers clearly exhibit quadrant glitch phenomenon as
observed from the axial tracking errors and the contour tracking errors from the
circular tests
• sliding mode controller removes spikes associated with quadrant glitches.

References
Altintas, Y. (2000) ‘Sliding mode controller design for high-speed feed drives’, Annals of the
CIRP, Vol. 49, No. 1, pp.265–270.
Doenitz, S. (1999) ‘Comparison of disturbance suppression for servo drives’, Eighth European
Conference on Power Electronics and Applications, Lausanne.
Franklin, G.F., Powell, J.D. and Naeni, A.E. (2002) Feedback Control of Dynamic System,
4th edition, New Jersey: Addison-Wesley.
Pritschow, G. (1996) ‘Control of high dynamic servo axes for milling machines’, Production
Engineering, Vol. III, No. 1, New York: Carl Hanser Verlag, pp.63–68.
Pritschow, G. (1998) ‘A comparison of linear and conventional electromechanical drives’,
Annals of the CIRP, Vol. 47, pp.541–548.
Spurgeon, S.K. and Edwards, C. (1998) Sliding Mode Control, London: Taylor & France,
pp.15–16.
Utkin, V.I. and Guldner, J.S. (1999) Sliding Mode Control in Electro-mechanical Systems, London:
Taylor and Francis.
Wang, J. (2004) ‘Robust tracking controller design with application to the motion control of an x-y
feed table for high-speed machining’, PhD Dissertation, Faculty of Engineering, Department
of Mechanical, Division of Production, Machine Design and Automation (P.M.A), Katholieke
Universiteit Leuven, Belgium.
Wang, J., Brussel, H.V. and Swevers, J. (2003) ‘Positioning and tracking control of an x-y table
with sliding mode control’, Proceedings of the 4th IFAC Symposium on Robust Control
Design, Milan.
Young, K.D., Utkin, V.I. and Ozguner, U. (1999) ‘A control engineer’s guide to sliding mode
control’, IEEE Transactions on Control Systems Technology, Vol. 7, No. 3.

You might also like