Modeling and Control of Linear Motor Feed Drives for Grinding Machines

Qiulin Xie and Steven Y. Liang

George W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology, Atlanta, GA 30332-0405, U.S.A.

Abstract

This paper presents a dynamic model of linear motor feed drives for grinding machines. In this
work, analytical or empirical models for functionality-critical features of linear motor, including
friction, grinding force, and force ripple will be discussed. First, a dynamic friction model, based
on the LuGre approach, will be used in the interest of describing not only observed static but also
dynamic friction phenomena such as presliding displacement, which is the prevailing friction
phenomenon for high precision application. In order to study the influence of grinding force on
linear motor feed drive positioning performance, an analytical grinding force model proposed by
Hahn and Lindsay is employed. In addition to friction and grinding force, force ripple is taken
into account, in view of its importance to machine tool applications, based on empirical
modeling. To validate the force ripple model, the simulation result has been compared with the
experimentally measured open velocity response of a linear motor motion system. The
effectiveness of the model has been shown by the agreement between the simulation and
experimental results. The friction model has also been validated through the comparison of
simulation results to experimentally measured response obtained by frequency domain
identification on an electromechanic motion system.

For the controller part, this report investigates the effectiveness of a model-based sliding mode
control for precision tracking in linear motor feed drives for grinding machines. The model based
sliding mode control makes use of authentic linear motor feed drives modes with the
considerations of friction force, force ripple and grinding force, which are functionally critically
for the implementation of linear motor feed drives in grinding machines. The simulations results
show that the system can achieve submicron positioning performance.For the low velocity
tracking, an accuracy of 4nm/s can be achieved after 0.25 second duration in the presence of
friction, force ripple and grinding force.

1. Introduction

One of the most ever-present goals in manufacturing is that of increasing the levels of quality
and accuracy of the parts being fabricated. Grinding, as an important finishing process, is of
great practical relevance with respect to this objective (Toenshoff, Karpuschewski et al. 1998).
To establish a precision grinding process, precision motion control of machine tool feed drive is
very crucial. With the continuing stringent demand on high performance, the tolerance on the
feed drive system, in terms of precision, is also ever more tightened. Currently, ball screw is the
dominant setup for the grinding machine tool feed drive (Slocum 1992). The application of
conventional feed drive for high speed and high accuracy machining is very limited (Pritschow
1998; Denkena, Tonshoff et al. 2004), because of its inherent deficiency such as dead zone,
friction induced backlash, transmission errors existing due to pitch tolerance of the leadscrew,

1
and mechanical limitations on position, velocity, and accelerations. As a promising technology, a
linear motor direct feed drive discards the transmission system required by a conventional feed
drive. Therefore, there exists no transmission associated error such as backlash, pitch error, etc.
Also, the friction problem is greatly alleviated by the application of direct drive. As a result,
direct linear drives can achieve high accuracy. Machine tools equipped with linear direct drive
have emerged and been displayed at the 2002 Exposition Mondiale de la Machine Outil (EMO)
in Hanover, Germany. In 2000, among the 25,000 total machining centers manufactured by
global manufacturers, 1,100 applied linear motor technology. In 2001, this figure has more than
doubled to reach 3,000 (Byrne, Dornfeld et al. 2003). Linear motors have been successfully
applied to Landis LTI and Toyoda GC32M, both of which are cam shaft grinders (Landis;
Toyoda).

When linear motor feed drives are implemented in grinding machines, the following disturbances
will arise: (1) force ripple, which originates from cogging, reluctance force, and commutation
error; (2) grinding force disturbances directly acting on linear motor; and (3) friction caused by
the relative movement between the slider and the supporting guideway, which continues to be a
problem even though it is already reduced when compared to conventional drive. Linear motor
feed drives gain high speed and high accuracy potential by discarding transmission systems.
However, because of the elimination of transmission systems, the effect of system disturbances
on the positioning performance will be much more significant due to the direct contact between
the moving mass and the linear motor. To become a viable feed drive technology for grinding
machines, all of these factors should be overcome by explicitly taking them into account in
controller design. One prerequisite for controller implementation is a good modeling of linear
motor feed drive for grinding machines. Unfortunately, the modeling of linear motor feed drives
for grinding machines has not been found in literature. To compensate for the gap in technique,
modeling of linear motor feed drives for grinding machines will be presented in this work, which
can address friction, grinding force, and force ripple.

For a linear motor feed drive system, the relative movement between the slider and guided way
incurs friction. Uncompensated friction in machine tool feed drive system causes static state
error, limit cycle, and stick-slip which imposes constraints on the positioning and tracking
performance of the machine, which in turn limits the product quality that can be achieved
(Armstrong-Helouvry, Dupont et al. 1994). To make valid friction compensation possible, an
authentic friction model is often indispensable. Compared to ball screw feed drive system, the
friction exhibited in linear motor feed drive is alleviated but not completely eliminated. Thus,
friction still remains as a major factor influencing the precision of the linear motor system,
particularly when the precision requirement on the feed drive system reaches to submicron
regime.

As an important physical phenomenon, friction has been intensively researched by doing

experiments, modeling, and simulation studies. By experimental studies, many friction properties
have been observed and can be classified into two categories: static characteristics, which
includes the kinetic and viscous force and the Stribeck effect; and dynamic characteristics, which
comprises presliding displacement, varying breakaway force, Dahl effect, and a frictional lag
(Armstrong-Hâelouvry 1991; Armstrong-Helouvry, Dupont et al. 1994).

2
To interpret those observed friction properties, many models have been proposed. All of the
existing models can be boiled down to static and dynamic models that target to explain the
observed friction characteristics. In static models, friction is modeled as a static map between
velocity and friction force. Due to their simplicity, these models have been extensively used for
both the ball screw feed drives and the linear motor feed drives control (Yang and Tomizuka
1988; Tung, Urushisaki et al. 1993; Tan, Huang et al. 2002; Yao and Xu 2002; Elfizy, Bone et al.
2004). Static models assume there is no motion while sliding. However, Dahl (1976) has
observed that there is a presliding displacement on the order of 2-5 microns in steel junctions
which is approximately a linear function of the applied force until breakaway occurs. Presliding
displacement is believed to be dominant in extremely high precision positioning applications and
of significant interest to the control community (Dahl 1976; Futami, Furutani et al. 1990; Ro and
Hubbel 1993). To capture static and dynamic characteristics such as presliding displacement, the
Dahl effect and friction lag, Dahl (1976) developed a comparatively simple model that was used
extensively to simulate systems with ball bearing friction. The friction model is an extension of
Coulomb friction, but it produces a smooth transition. One of the shortcomings of the Dahl
model lies in its incapability to model the steady state friction characteristic in sliding regime
such as the Stribeck curve. To overcome this deficiency, a joint effort between Lund and
Grenoble has been made which led to the derivation of a new nonlinear analytic friction model,
i.e., the LuGre model (Canudas de Wit, Olsson et al. 1995; De Wit and Lischinsky 1997). The
LuGre model combines the presliding behavior of the Dahl model with the steady state friction
characteristic in sliding regime, such as the Stribeck curve. The strength of the dynamic LuGre
friction model is the ability to capture a large number of practically observed friction phenomena
as described in (Canudas de Wit, Olsson et al. 1995; De Wit and Lischinsky 1997). Therefore,
the LuGre model serves as a good friction model for machine tool feed drives, especially for the
application where position accuracy requirements may be down to submicron regime.

The application of the LuGre model for friction modeling and compensation in conventional ball
screw feed drive system has been demonstrated by Ro et al. (2000). However, static friction
models which cannot capture the presliding effect are still widely used in the control community
to model the friction for linear motor feed drives (Tan, Huang et al. 2002; Yao and Xu 2002).
Since it is common that the precision requirements on linear motor feed drives often is in
submicron regime, a more accurate dynamic friction such as the LuGre model should be a better
choice when modeling friction for linear motor feed drives. Based on this consideration, the
LuGre model will be utilized to model friction for linear motor feed drives in this work.

When linear motors are used in machine tool feed drives, cutting forces will have severe effects
on the positioning and tracking performance of linear motor feed dives because they are directly
reflected to linear motors. To reduce positioning and tracking errors induced by cutting force is a
challenge when implementing linear motor feed drives for machine tools. Much effort has been
devoted to meet this challenge. For single point turning process, an H ∞ optimal feedback control
approach was proposed by Alter and Tsao (1996) to increase the dynamic stiffness for the force
loop so as to minimize the effect of cutting force on the positioning and tracking performance of
a linear motor feed drive system. Later on, Choi et al. (2005) designed a MIMO H ∞ controller
and implemented it successfully on a linear motor X-Y table control of linear motor machine tool
feed drives for end milling process aimed at improved tracking and contouring performances.
Unfortunately, although grinding process usually demands higher feed drives positioning

3
performances when compared to single point cutting and deterministic end milling process, no
research has been done to investigate the grinding force effects on linear motors feed drives for
grinding machines in the current literature.

Among all the available linear motors, synchronous permanent linear motors (PMLMs) are
probably the most closely related to applications involving high speed and/or high precision
motion control owing to their benefits such as high force density available, low thermal loss, etc.
Due to these advantages, PMLM is a very good candidate for machine tool feed drives. However,
in addition to thrust force, PMLMs generate undesired force ripple which causes position
dependence of thrust that must be compensated to achieve high positioning and tracking
performance (Otten, de Vries et al. 1997). One of the sources of force ripple is cogging force
which results from the mutual attraction between the magnets and iron cores of the translator
(Van Den Braembussche, Swevers et al. 1996). It is present even when there is no motor current.
Another source of force ripple is reluctance force, which is caused by the variation of the self
inductance of the windings with respect to the related position between the translator and the
magnets. The modeling force ripple for linear motor can be found in literature (Van Den
Braembussche, Swevers et al. 1996).

Although several approaches have been presented in literature to modeling linear motor stages
for high precision motion control applications, they cannot be readily applied to modeling linear
motor feed drives for grinding machines due to the following: (1) Although linear motor is
targeted to high precision motion, the conventional static friction models, which are not capable
of handling presliding displacement, were used. This is not satisfying because presliding
displacement is essential for a precision system like a linear motor feed drive for grinding
machine to meet its accuracy which may be down to submicron level; (2) The controller
algorithms used for linear motor feed drives often lump the cutting force as a part of external
disturbance, and do not consider the coupling nature between the grinding force and the feed
drive output. As a result, the bigger range of uncertainty caused by modeled grinding force will
generate a bigger positioning tolerance. To solve these problems, first, a dynamic friction model
will be presented. This model can capture not only observed static but also dynamic friction
phenomena such as presliding displacement, which is the dominant friction contribution for high
precision application. Then, a grinding force model based on the relation between the radial
infeed velocity and grinding force is employed. In addition to friction and grinding force, force
ripple is also modeled. Moreover, all of the three factors that have influences on the direct feed
drive positioning and tracking have been taken into account under one framework.

In this report, a modeled based sliding mode controller will designed according to the established
model of the linear motor feed drives for grinding machines. The merits of the model-based
sliding mode controller are two folds: (1) By using the authentic model of the linear motor feed
drive, the uncertainty intervals associated with the system model can be reduced as small as
possible. (2) The unaccounted model uncertainties and external disturbances, although already
minimized, can be handled by the sliding mode controller. Because of these merits, the controller
can achieve the robustness in presences of model uncertainties and external disturbances with the
smallest loss of tracking performance.

4
2. System Modeling
A linear motor can be envisioned as a rotary motor cut axially and unrolled flat. It actually
consists only of the primary part "stator" and secondary part “rotor" as illustrated in Figure 1.
The thrust is directly applied to the slide or to the object to be moved. For almost every kind of
rotary motor, there is a counterpart in linear motor. The same basic technologies used to produce
torque in rotary motors are used to produce force in linear motors. Similar to its rotary
counterpart, a linear motor can be classified as either a DC or AC motor which can then be
further classified as induction motors, linear synchronous motors, or linear variable reluctance
motors (Boldea and Nasar 1997). Among all the available linear motors, synchronous permanent
linear motors (PMLMs) are probably the most naturally related to applications involving high
speed and/or high precision motion control due to their benefits such as high force density
available, low thermal loss, etc. Due to these advantages, only PMLM will be investigated in this
research.

2.1 System Modeling

The mathematical model for motor terminal voltage can be expressed as

. di
u (t ) = K e x + Ri (t ) + L (1)
dt
where u(t ) is time-varying motor terminal voltage, K e is back EMF voltage constant, L is
inductance, and R is resistance. Usually, resistance is very small compared to inductance, so the
electrical dynamics can be ignored. According to Newton’s law, the following equation can be
obtained
..
F (t ) = m x + f friction + f ripple + f grinding (2)
where m is the moving thrust mass, f ripple , f friction , and f grinding are force ripple, friction force and
grinding force, respectively. The thrust force F (t ) is expressed as
F (t ) = K f i (t ) (3)
where K f is the force constant determined by motor poles and magnetic flux, etc. If we
combine the above equations and ignore the electrical dynamics, a general model for linear
motor feed drive system can be described by the following differential equation
.. K f Ke . K f
x=( x+ u (t ) − f friction − f ripple − f grinding ) / m (4)
R R

2.2 Friction Modeling

The LuGre model combines the presliding behavior of the Dahl model with the steady state
friction characteristic in sliding regime such as the Stribeck effect as shown in figure 2. The
strength of the dynamic LuGre friction model is its ability to describe a large number of
practically observed friction phenomena as described in (Armstrong-Helouvry, Dupont et al.
1994). For high precision motion, a friction model, which can address both presliding and sliding
behavior, will be desirable.

5
The LuGre friction model (Canudas de Wit, Olsson et al. 1995) will be adopted since it combines
the presliding behavior and the steady state friction characteristic in sliding regime such as the
Stribeck curve which plays an important role in describing the lower velocity motion of linear
motor According to the LuGre friction model, the friction force can be modeled as a function of
the state variable z as illustrated in figure 3 and velocity υ :
dz
F f = σ 0 z + σ 1 + σ 2υ (5)
dt
where the parameters σ 0 , σ 1 and σ 2 are the asperity stiffness, the micro-viscous friction
coefficient, and the viscous friction coefficient, respectively. The interpretation of the internal
state is linked to the bristle friction model; viz. the state variable z represents the average
deflection of the contacting asperities:
dz υ
= υ −σ0 z (6)
dt g (υ )
where g (υ ) is a decreasing function for increasing velocity with an upper limit equal to the
stiction force Fs and a lower limit equal to the Coulomb friction force Fc :
δ vs
− v / vs
g (υ ) = sgn(v)( Fc + ( Fs − Fc )e ) (7)
where vs is the Stribeck velocity and δ vs is the Stribeck shape factor.

2.3 Grinding Force Modeling

Basically, grinding force models consist of a physical part which includes the speed ratio, the
depth of cut, and the equivalent diameter, as well as empirical constants for grinding force and
workpiece materials (Denkena, Tonshoff et al. 2004). A grinding force model developed by
Hahn and Lindsay (Bhateja and Lindsay 1982) is very straightforward and will be used to model
the grinding force. This model assumes there is linear relationship between the material removal
rate and the normal force. In general, the total specific grinding energy is composed of chip
formation, plowing, and the sliding components (Malkin 1989). Correspondingly, there are three
grinding force components. Among them, a threshold force can represent the sum of plowing and
sliding force components. Only when the amount is above the threshold will it contribute to
material removal. For easy to grind material, the energy consumption associated with sliding and
plowing will be insignificant compared to chipping energy, and hence almost all energy is used
for material removal and the threshold force can be ignored. Therefore, the grinding force can be
modeled to be proportional to the material removal rate. For the cylindrical plunge grinding as
illustrated in figure 4, the material removal can be computed as
π bs d w v (8)
where bs is the grinding wheel width, d w is the diameter of workpiece and v is the actual infeed
velocity. However, because of final stiffness associated with work, wheel and contact, the actual
radial infeed velocity will be different from the commanded radial infeed velocity (Malkin 1989).
Neglecting wheel wear for the moment, continuity requires that the difference between the
controlled u (t ) and the actual v(t ) infeed velocities be equal to the time rate change of the radial
elastic deflection of the grinding system:
.
u (t ) − v(t ) = ε (9)

6
 Fn
ε= (10)
ke
where Fn is the normal force component and k e is the effective stiffness. For cylindrical plunge
grinding with a constant radial speed, there is a relationship between the control federate Vr and
the actual velocity Vreff as described by
d F
V −V = ( n) (11)
r reff dt k
As shown in (Denkena, Tonshoff et al. 2004) for cylindrical plunge grinding, the grinding
normal force can be approximated to be proportional to the material removal rate as depicted by
Fn = cπ d w bsVreff (12)
where d w is the workpiece diameter, bs is the grinding width, and c is the proportionality
constant describing the dullness of the grinding wheel. Combining (1) and (2), the relation
between Fn and V can be obtained as:
r
Fn ( s ) cπ d wbs
= (13)
Vr ( s ) 1 + (cπ d wbs / k ) s

2.4 Force Ripple Modeling

As mentioned earlier, cogging is a magnetic disturbance force, which is due to the mutual
attraction between the magnets and iron cores of the translator. It depends merely on the relative
position of the motor coils with respect to the magnets, and is independent of the motor current.
The reluctance force is caused by the variation of the self inductance of the windings with
respect to the related position between the translator and the magnets. Thus, the reluctance force
also has a periodic relationship with the translator-magnet position .The reluctance force exists
only when the motor current is different from zero, and its absolute value depends on the
required force and the relative position of the carriage. Cogging and reluctance force are
commonly called force ripple.

From motion control considerations, force ripple is highly undesirable. They can be minimized
or even eliminated by an alternative design of the motor structure or spatial layout of the
magnetic materials, such as skewing the magnet, optimizing the disposition and width of the
magnets, etc. However, these mechanisms often increase the complexity of the motor structure.
A control algorithm aimed at eliminating the effects of force ripples is therefore highly desirable.
For that purpose, a modeling of the force ripple is much desired. Ripple models suggested by
P.V. Braembussche et al. (1996) will be utilized in this research. In his model, the reluctance
force is described as:
2π 2π
Frelu tan ce = Fdesired [a f sin( x + φ11 ) + b f sin(3 x + φ12 )] (14)
p p
and the cogging force is modeled as:
2π 2π
Fcogging = ac sin( x) + bc sin(3 x + φc ) (15)
p p

7
where Fdesired represent the desired force, p is the average pitch of the magnets for the considered
stroke, x represents the position of the motor carriage, a f , b f , ac and bc are estimated
coefficients, and φ11 , φ12 and φc are estimated phase shift.

3. Experimental Validation

In this section, experimental results from the literature are provided to illustrate the effectiveness
of the proposed modeling strategy. Since experimental results for system which has the exact
same parameters as our simulations in the previous section cannot be found from literature, those
parameters given are utilized in the corresponding experiments to do simulation in order to make
a valid comparison.

Figure 5 depicts the simulated and experimental velocity response for a permanent magnet linear
motor with both friction and force ripple with parameters documented in (Tan, Huang et al.
2002). Since the dynamic friction parameters are not given, they were based on data presented in
(Lampaert, Swevers et al. 2004). From Figure 5, it can be seen that the simulation results match
the experiments reasonably very well, which means our model can predict the response of a real
linear motor system.

Figure 6 and Figure 7 illustrates the system response achieved from both simulation and
experiments. The simulation parameters used and experimental results are documented in
(Hensen, Van De Molengraft et al. 2002). In Figure 6, it is evident that the motion is in stiction
regime since the displacement is very close to bristle deflection. From this figure, it can also be
seen that the simulated presliding displacement matched the measured response very well, which
means the LuGre model can capture the preslidng behavior very accurately. Figure 7 describes
results with another sinusoidal excitation with higher amplitude and period. It can be seen that
the measured response is higher than both macroscopic and presliding displacements obtained
from simulations. The possible reasons are given in (Hensen, Van De Molengraft et al. 2002) .

4. Modeling Simulation Results and Discussion

Numerical simulations were carried out to investigate the effects of friction, force ripple, and
grinding force on the system response. Only open loop simulations were performed since our
focus in this paper is to model the motion behavior of linear motor feed drive under the influence
of friction, force ripple, and grinding force. In this part, the friction effects are studied first. The
dynamic presliding behaviors and the varying breakaway force under inputs with different rates
are examined. Then, the effect of force ripple combined with friction was studied. Finally,
motion behaviors of the linear motor feed drive system with all the three factors combined are
investigated. Friction parameters used for simulation are listed in table 1. Fs represents static
friction which is assumed to be 50% larger than Coulomb friction. The bristle stiffness σ 0 is
chosen be to the same quantitative order as listed in (Lampaert, Swevers et al. 2004). The value
of damping coefficient σ 1 was chosen to give a damping ζ =0.7 for the linearized system
equation in the stiction regime (Canudas de Wit, Olsson et al. 1995).

8
 Table 1 Friction parameters used for simulation

parameter value units

Fs 120 (N)
Fc 80 (N)
υs 0.1 (m/s)
σ0 2 *106 (N/m)
σ1 2 *10 4 (Ns/m)
σ2 0.4 (Ns/m)

4.1 Effects of Friction

To study the friction behavior, two kinds of input were considered. One is step input, the other is
sinusoidal input. For step input, we have considered two cases: one input is below the stiction,
the other is larger than stiction. Figure 8 shows the response of step input with a magnitude of 12,
which is 10% of the stiction. From this figure, it can be seen that the motion is in the stiction
regime where macroscopic displacement is almost equal to the presliding displacement, which is
about 6 µ m . This is in the same order of magnitude as reported in literature (Armstrong-
Helouvry, Dupont et al. 1994). Figure 9 shows the case when input is 10% larger than the
stiction. The upper part demonstrated the motion is away from stiction regime since the
presliding displacement is much smaller than the macro displacement. This conclusion is
confirmed by the lower part because it is very clear that the friction deceased gradually and
eventually stooped at the Coulomb friction.

Figure 10 (a) and (b) show the system response with three sinusoidal inputs with amplitude of 12
but with different frequencies, which are 0.1, 1, and 10, respectively. These first two inputs
didn’t show distinguishable differences regardless of the different input frequency. For the case
of input frequency 10, again the response amplitude is the same as the former two; however,
there is a phase shift.

When the amplitude is 120, the system response varied with input frequency. It can be seen
higher frequency responds to lower macroscopic displacement. Although an explanation can not
be given , it is possible that it is related to friction lag phenomena (Canudas de Wit, Olsson et al.
1995).

Figure 13 (a) demonstrated the breakaway phenomena. To show the breakaway points, the
transition region is zoomed to Figure 13 (b). From this figure, we can tell how breakaway points
vary with different input rate. Higher rate will result in lower breakaway critical point. This
matches the findings in the literature (Canudas de Wit, Olsson et al. 1995).

9
4. 2 Effects of Friction and Force Ripple

Figure 13 shows the system response when both friction and force ripple were taken into
consideration. At first glance, it is hard to tell the difference between figure 9 and figure 13 (a);
close examination of friction as shown in the lower part of figure 13 (a) and the displacements as
illustrated in figure 13 (b) reveal that the corresponding friction force and displacement are
smaller when force ripple effect is considered. This is the force ripple counter balance part of the
input, therefore the resulting displacement is smaller.

Figure 14 shows the system response with amplitude of 130 and a period of 1. From Figure 14
(a), it can be seen that the macroscopic displacement is shifted as time advances which is
reasonable because the force ripple is decreasing with shift as shown in Figure 14 (b).

4.3 Effects of Combined Friction, Force Ripple, and Grinding Force

Figure 15 shows the system response when all of the three factors: friction, force ripple, and
grinding force are taken into account. In Figure 15, (a) shows both the macroscopic and
presliding displacement, (b) illustrates the change of three force components. It can be seen that
the force ripple is negligible among all of the three force components. Although the friction is
increasing during the simulation period, the grinding force is dominant in the whole simulation
period.

5. Controller Design

In this section, the controller design for linear motor drives will be presented after a brief review
of the system model, which is detailed in the modeling paper.

5.1 Modeling of Linear Motor

A current-controlled three phase linear motor with iron core is considered for the controller
design purpose. Figure 16 shows the model of a single axis linear motor feed drive for a grinding
application. In figure 16, u is the control signal; Ka is the current amplifier gain, K f is the motor
force constant, the mass is the equivalent mass of the whole drive. The friction, grinding force
and force ripple are the same as modeled in the modeling paper.

10
 Voltage Current Thrust force
Axis
u + Position
Ka Kf Mass
Control
Signal _
Current Amplifier Motor
_ _ Friction

Grinding Force

_
Force Ripple

Figure 16 Block diagram of linear motor model.

5.2 Model Based Sliding Mode Controller Design

In this section, sliding mode control design, which is based LuGre friction model, force ripple
model and grinding force model as established in the modeling paper, will be discussed. The idea
of sliding mode control was first initiated by utkin (Utkin 1977). The purpose of sliding mode
control is to enforcing the controlled system state trajectory to a chosen sliding mode trajectory,
on which system states will be asymptotically converge to the origin of the phase plane (Hung,
Gao et al. 1993). The sliding mode control approach is very attractive to system with
uncertainties, like the linear motor feed drives, due to its robustness and in many cases even
invariance characteristics, which means the system is complementally insensitive to parameter
uncertainties and external disturbances (Hung, Gao et al. 1993). Assuming all the system states
are available, although the average bristle deflection is immeasurable, the sliding mode control
design can be derived as follows:

First, the sliding surface can be defined as

.
S = e+ λ e (16)
where e is the tracking error evaluated by the difference of xd , the desired trajectory state , and
x , the actual state, as
e = xd − x (17)
Then, the reaching law method (Gao and Hung 1993) will be chosen to establish the reaching
condition due to its merits in reducing undesired chattering phoneme, which occurs in the
implementation of sliding mode control due to discontinuous switch (Utkin 1977). The reaching
law,
.
S = −η sgn( s ) − ks (18)
, consists of a constant rate term, −η sgn( s ) and a proportional rate term, − ks .

The requirement of the above reaching law on the control law, after some mathematical
manipulations, is found to be
m .. . . F f + Fg + Fr
u= ( xd + λ ( xd − x) + η sgn( s ) + ks ) + (19)
Ka K f Ka K f

11
where F f , Fg , and Fr are friction force, grinding force and force ripple, respectively.
To prove the effectives of the used reaching law, the Lyapunov function is defined as
1
V = S2 (20)
2
The time derivative of the above Lyapunov function can be evaluated by
. .
V = S S = −η S − kS 2 (21)
.
V is positive definite and V is negative definite (both η and k are positive ). According to the
second method of Lyapunov, the S will asymptotically tend to zero. Once on the sliding surface,
the tracking error will be guaranteed to converge to zero.

The derived controller law is implemented for the linear motor feed derive as shown in Figure 17,
where LM is the linear motor model as illustrated in figure 16.

+ Output
Sliding Mode
LM
Controller
Reference Input
_

Figure 17 Block diagram of sliding mode control strategy

6. Controller Simulation

In this section, the effectiveness of the proposed controller design will be illustrated by computer
simulations. The parameters used for the linear motor feed drive are documented in the modeling
paper. The design controller parameters are selected as
η = 5 k = 10 λ = 50

6.1 Position Tracking

The positioning performance of the sliding mode controller will be considered first for a
sinusoidal input to demonstrate the effectiveness of the proposed controller, and then three more
reference trajectory will be used in the simulations. For all these simulations,
only friction force is included in the system model.

12
 (a) (b)

(c) (d)

(e) (f)

Figure 18: The simulation results sliding mode control to desired sinusoidal input
(a) Desired positions. (b) Position tracking error. (c) Speed error. (d) Velocity error vs. position
error. (d) Control input. (e) Friction force

The simulation results to the desired sinusoidal input are shown in figure 18. The desired
sinusoidal input has a magnitude of 100 m and a period of 1second.The position error is shown
in figure 18 (b), which indicates a maximum tracking error of around 7000 µ m near the starting
point. The tracking error decreases very sharply towards zero, which means the controlled
system will follow the desired sinusoidal trajectory after a short time. The velocity error as
shown in figure 18 (c) also indicates that the trajectory will be perfectly followed. From figure
18 (d), it can seen that the controlled system states will move toward the sliding line, then once

13
reach it, both the position and velocity error will converge to zero as expected by the sliding
mode theory. Figure 18 (e) shows the high frequency switch of control input with bounded
magnitude. The model based sliding mode controller performed perfectly in the presence of
friction force as shown in figure 18 (f).

The effectiveness of the model based sliding mode control is demonstrated by the prefect
tracking of a desired sinusoidal input. In order to test the effectiveness of the controller with
different inputs. more extensive simulations will be carried out The response to a unit step input,
which jumps from zero at time instant 1 second to 100mm, will be examined. The simulations
results are shown in Figure 19

(a) (b)

(c) (d)

Figure 19: The simulation results sliding mode control to 100 mm step input
(a) Comparison between desired and simulated positions. (b) Speed error. (c) Control input. (d)
Friction force

The simulated position is shown in figure 19 (a). it can be seen, around the time instant of 1
second, simulated response jumps to around 135 mm, and then dives to around -125 mm sharply.
The fast change of position will necessitate an extreme high control effort, as can be seen from
Figure 19 (c). Also, the speed change shown in Figure 19 (b) is significant, and as a result, the
resulted friction force will be big as illustrated in Figure 19 (d). The undesired response to step

14
input is believed to be ascribed to the unsmoothness of the input itself, which has a jump at time
instant of one second.

An acceleration limited trajectory limited trajectory profile, which is smoother than step input, is
selected to evaluate the performance of proposed controller. The simulations results are
described in Figure 20.
(a) (b)

(c) (d)

Figure
20: The simulation results sliding mode control to acceleration limited trajectory
(a) Simulated position. (b) Position tracking error. (c) Control input. (d) Friction force

Figure 20 (a) shows the acceleration, but not jerk, limited reference trajectory. The tracking error
is demonstrated in Figure 20 (b), from which a maximum tracking error of 4.4 µ m is identified.
Compared to the tracking errors for the step input case, an acceleration limited reference
trajectory improved the tracking performance significicantly. In addition, the friction force curve
is much nice than the step input case. However, the control input as shown in Figure 20 (c)
indicates a big control effort occurred at the time instant of 0.5 second, which is a result the
boundless jerk. To get a better tracking performance, a jerk limited reference trajectory as shown
in Figure 21 will be used for simulations.

15
 Figure
21 Jerk limited references trajectory used in simulation

The reference trajectory as shown in Figure 21, is generated as in (Tan, Huang et al. 2001).
A displacement of 100 mm was commanded with a maximum velocity of around 200 mm/s,
acceleration /deceleration values of 115 mm / s 3 , and a jerk limit of 240 mm / s 3 .
The simulated tracking performance for the jerk limited reference trajectory is shown in Figure
22.
(a) (b)

Figure 22: The simulation results sliding mode control to the jerk limited trajectory
(a) Simulated position tracking error (b) Control input.

The superior tracking performance using the jerk limited reference trajectory is indicated by
small tracking error as shown in figure 22 (a), where except a maximum absolute tracking error
of 0.8 µ m , in most time the tracking errors are within a plus and minus 0.2 µ m range. The
magnitude of control input as shown in figure 22 (b) is very well bounded, with an upper limit

16
less than 4 and a lower limit larger than -2.Howver, the controller input switches very fast, which
may be a problem when implemented in real physical system.

6.2 Velocity tracking

When the linear motor feed drives are used for grinding applications, the velocity tracking
performance will play a major role in determining the quality, say, surface finish.
In view of the extreme importance of perfect velocity following, the velocity tracking
performance will be evaluated for the proposed controller design. The desired input for the
simulation is a ramp input with a slope of 1 µ m / s . In order to gets a realistic simulation, friction
force, grinding force and force ripple are included in the system model.

The simulation results for the ramp input are shown in figure 23. The effectiveness of the model
based controller in a low velocity tracking is demonstrated by the small velocity error as shown
in figure 23 (b), where the maximum tracking error is less than 150 nm/s. A zoom view of figure
23 (b) will further discover that the tracking error will be less than 4 nm/s after a period of 0.25
second. The extreme small position error as shown in figure 23 (a) also confirms the
effectiveness of the proposed controller for low velocity application in presences of external
disturbances such as friction, grinding force and force ripple. The friction force is shown in
figure 23 (d), from which it can be seen the friction force is proportional to time. This means
friction changes proportionally with respect to displacement since a constant velocity input is
used. The proportionality of friction with respect to displacement implies the motion of the feed
drive is close to presliding region which is captured by LuGre friction model (Canudas de Wit,
Olsson et al. 1995). The force ripple is shown in figure 23 (e), from which it can be seen the
force ripple changes very small in whole displacement range, which is reasonable because the
whole displacement is much less than the pole pitch of the linear motor. Grinding force is shown
in figure 23 (f), from which it can be seen the grinding force is more significant than friction and
force ripple for the considered case.

(a)
(b)

(c) (d)

17
 (e) (f)

Figure 23: The simulation results sliding mode control to a ramp input
(a) Position error. (b) Velocity error. (c) Control input. (d) Friction force (e) Force ripple.(f)
Grinding force

5. Conclusions

This paper presents a dynamic model of linear motor feed drives for grinding machines. In this
work, a dynamic friction model (LuGre model) is used to capture not only observed static
friction phenomena but also dynamic friction phenomena such as presliding displacement which
is the prevailing friction phenomena for high precision application. Friction behaviors under step
input and sinusoidal input are investigated. Both stiction regime and sliding regime are examined.
It has been found that the presiding displacement is very close to the bristle deflection in stiction
regime. However, macroscopic displacement is much larger than the bristle deflection in the
sliding regime. The varying breakaway phenomena are captured by simulation. It has been found
that larger force rate will result in a less breakaway point. Force ripple was also studied based on
the periodical nature of force ripple with respect to position. It was discovered under some
circumstances the effect of force ripple is insignificant, while in other cases it will induce a shift
of macroscopic displacement. To study the influence of grinding force on linear motor feed drive
positioning performance, an analytical grinding force model proposed by Hahn and Lindsay is
employed. When all of the three force components are taken into account in one simulation, it

18
was found that the grinding force is the most important component, followed by the friction. The
effect of force ripple is not important.

To validate the force ripple model, the simulation result has been compared with the measured
open velocity response of a linear motor motion system. The effectiveness of the model has been
shown by the good agreement between the simulation and experimental results. The friction
model has also been validated by the good agreements between the simulation results and
measured response obtained by frequency domain identification on an electromechanic motion
system.

The report investigates the effectiveness of a model-based sliding mode control for precision
tracking in linear motor feed drives for grinding machines. The model based sliding mode
control makes use of authentic linear motor feed drives modes with the considerations of friction
force, force ripple and grinding force, which are functionally critically for the implementation of
linear motor feed drives in grinding machines. The simulations results show that the system can
achieve submicron positioning performance.For the low velocity tracking, an accuracy of 4nm/s
can be achieved after 0.25 second duration in the presence of friction, force ripple and grinding
force.

References:

Alter, D. M. and T. Tsu-Chin (1996). "Control of linear motors for machine tool feed drives:
design and implementation of H infinity optimal feedback control." Transactions of the ASME.
Journal of Dynamic Systems, Measurement and Control 118(4): 649-56.
Armstrong-Hâelouvry, B. (1991). Control of machines with friction. Boston, Kluwer Academic
Publishers.
Armstrong-Helouvry, B., P. Dupont, et al. (1994). "A survey of models, analysis tools and
compensation methods for the control of machines with friction." Automatica 30(7): 1083-138.
Bhateja, C. and R. Lindsay (1982). Grinding, theory, techniques, and troubleshooting. Dearborn,
Mich., Society of Manufacturing Engineers, Marketing Services Dept.
Boldea, I. and S. A. Nasar (1997). Linear electric actuators and generators. Cambridge ; New
York, Cambridge University Press.
Byrne, G., D. Dornfeld, et al. (2003). "Advancing cutting technology." CIRP Annals -
Manufacturing Technology 52(2): 483-507.
Canudas de Wit, C., H. Olsson, et al. (1995). "A new model for control of systems with friction."
IEEE Transactions on Automatic Control 40(3): 419-25.
Choi, C. and T.-C. Tsao (2005). "Control of linear motor machine tool feed drives for end
milling: Robust MIMO approach." Mechatronics 15(10): 1207-1224.
Dahl, P. R. (1976). "Solid friction damping of mechanical vibrations." AIAA Journal 14(12):
1675-82.
De Wit, C. C. and P. Lischinsky (1997). "Adaptive friction compensation with partially known
dynamic friction model." International Journal of Adaptive Control and Signal Processing 11(1):
65-80.

19
Denkena, B., H. K. Tonshoff, et al. (2004). "Analysis and control/monitoring of the direct linear
drive in end milling." International Journal of Production Research 42(24): 5149-66.
Elfizy, A. T., G. M. Bone, et al. (2004). "Model-based controller design for machine tool direct
feed drives." International Journal of Machine Tools & Manufacture 44(5): 465-77.
Futami, S., A. Furutani, et al. (1990). "Nanometer positioning and its micro-dynamics."
Nanotechnology 1(1): 31-7.
Hensen, R. H. A., M. R. J. G. Van De Molengraft, et al. (2002). "Frequency domain
identification of dynamic friction model parameters." IEEE Transactions on Control Systems
Technology 10(2): 191-196.
Lampaert, V., J. Swevers, et al. (2004). Comparison of model and non-model based friction
compensation techniques in the neighbourhood of pre-sliding friction, Boston, MA, USA, IEEE.
Landis. from http://www.landis-us.com/machines/crankpin/1400.htm.
Malkin, S. (1989). Grinding technology : theory and applications of machining with abrasives.
Chichester,New York, J. Wiley
Otten, G., T. J. A. de Vries, et al. (1997). "Linear motor motion control using a learning
feedforward controller." IEEE/ASME Transactions on Mechatronics 2(3): 179-187.
Pritschow, G. (1998). "Comparison of linear and conventional electromechanical drives." CIRP
Annals - Manufacturing Technology 47(2): 541-548.
Ro, P. I. and P. I. Hubbel (1993). "Model reference adaptive control of dual-mode micro/macro
dynamics of ball screws for nanometer motion." Journal of Dynamic Systems, Measurement and
Control, Transactions of the ASME 115(1): 103-108.
Ro, P. I., W. Shim, et al. (2000). "Robust friction compensation for submicrometer positioning
and tracking for a ball-screw-driven slide system." Precision Engineering 24(2): 160-173.
Slocum, A. H. (1992). Precision machine design. Englewood Cliffs, N.J., Prentice Hall.
Tan, K. K., S. N. Huang, et al. (2002). "Robust adaptive numerical compensation for friction and
force ripple in permanent-magnet linear motors." IEEE Transactions on Magnetics 38(1 II): 221-
228.
Toenshoff, H. K., B. Karpuschewski, et al. (1998). "Grinding process achievements and their
consequences on machine tools challenges and opportunities." CIRP Annals - Manufacturing
Technology 47(2): 651-668.
Toyoda. from http://www.toyodausa.com/gc32m.html.
Tung, E. D., Y. Urushisaki, et al. (1993). Low velocity friction compensation for machine tool
feed drives, San Francisco, CA, USA, American Autom. Control Council.
Van Brussel, H. and P. Van den Braembussche (1998). "Robust control of feed drives with linear
motors." CIRP Annals - Manufacturing Technology 47(1): 325-328.
Van Den Braembussche, P., J. Swevers, et al. (1996). "Accurate tracking control of linear
synchronous motor machine tool axes." Mechatronics 6(5): 507-521.
Yang, S. and M. Tomizuka (1988). "Adaptive pulse width control for precise positioning under
the influence of stiction and Coulomb friction." Journal of Dynamic Systems, Measurement and
Control, Transactions ASME 110(3): 221-227.
Yao, B. and L. Xu (2002). "Adaptive robust motion control of linear motors for precision
manufacturing." Mechatronics 12(4): 595-616.

20
 Figure 1 Schematic illustration of a typical linear motor

Figure 2 Stribeck curve vs. velocity (Armstrong-Helouvry, Dupont et al. 1994)

Figure 3 Bristle deflection (Ro, Shim et al. 2000)

21
 Figure 4 Schematic of cylindrical grinding

Figure 5 Comparison of simulation and measured velocity response of a linear motor

22
Figure 6 Sinusoidal input u=0.01sin (40*pi*t)

Figure 7 Sinusoidal input u=0.035sin (t)

23
Figure 8 Open loop step response (u=12)

Figure 9 Open loop step response (u=130)

24
 (a) (b)

Figure 10 Open loop sinusoidal responses with the same magnitude but different frequencies
(a) presliding displacement; (b) bristle deflection

(a) (b)

Figure 11 Comparison of open loop response excited by sinusoidal inputs having the same
magnitude (stiction) but different frequencies. (a) Macroscopic displacement. (b) Presliding
displacement.

(a) (b)

Figure 12 Breakaway response

25
 (a) (b)

Figure 13 Open loop step response (u=130)

(a) (b)

Figure 14 Open loop step response ( u = 130sin(2* π * t ) )

(a) (b)

Figure 15 Step response with the consideration of friction, force ripple, and grinding force
(u=180)

26
Introduction

The machine tool feed drives control the position and velocities of the machine axes in
accordance with commands generated by the computer numerical control (CNC). The
requirements for a feed driver include: control over a wide range of speeds, precise control of
position under disturbances such as friction and cutting force, rapid response and precise
coordination of multiple axes for contouring operation (Srinivasan and Tsu Chin 1997).

A general machine tool feed drive controller structure as described in Fig. 1 for single axis
application is composed by an inner velocity loop and an outer position loop, both of which with
cascade compensator. Design criteria are based in achieving close position loop bandwidth,
steady state accuracy and rejection of disturbances. Additionally, to reduce the tracking error, a
feedforward compensator, which represents the inverse of the close loop transfer function, can be
added to the close loop configuration. If the close loop inverse transfer function has unstable
poles it can not be implemented directly, to solve this problem a feedforward controlled called
Zero Phase Error Tracking Controller (ZPETC) based on zero pole cancellation was introduced
(Tomizuka 1987). In this way, the overall transfer function will be close to unity, and, accordingly, a
high tracking accuracy can be archived. However, feedforward control will generate high
frequency components in the control signal. To eliminate the high frequency content of the
control signal being generated, the inverse compensation filter (IKF) control strategy was
proposed (Weck and Ye 1990).
+

+
Feedfoward Position Velocity
Compensation
Motor Slide Integrator
Compensation Compensation
_ _

F
igure 16 : General single axis control structure

Linear motor motion systems gain high speed and high accuracy potential by discarding
transmission systems. However, due to the lack of a transmission unit, the tracking behavior of a
direct drive design is very prone to disturbances and model parameter variations (Van Den
Braembussche, Swevers et al. 1996; Srinivasan and Tsu Chin 1997).

One of the major disturbances in linear motor feed drives is force ripple. Among all the available
linear motors, synchronous permanent linear motors (PMLMs) are very good candidate for
machine tool feed drives owing to their benefits such as high force density available, low thermal
loss, etc. However, in addition to thrust force, PMLMs generate undesired force ripple which
causes position dependence of thrust (Van Den Braembussche, Swevers et al. 1996; Otten, de
Vries et al. 1997). Many efforts have been devoted to reduce the effect of force ripple.
Feedforward compensation (Van Den Braembussche, Swevers et al. 1996) was used to cancel
out the effect of force ripple based on a calibrated model. In view of the fact that force ripple is

27
reproducible, a neuro-network-based learning feedfoward controller, consisting of model-based
feedback component and a learning feedforward component, was presented. The effects of this
controller in reducing inaccuracies caused by force ripple was demonstrated by simulation and
experimental results (Otten, de Vries et al. 1997).

Another major disturbance in linear motor feed drives is friction in the guideways. Friction in
servo mechanism is a nonlinear phenomenon which usually causes control problems such as
steady state error, limit cycle, stickslip behavior. Many controller design are based on static
friction model, in which friction force is modeled as a static map of moving velocity (Armstrong-
Helouvry, Dupont et al. 1994). In (Tung, Anwar et al. 1993), a feedfoward friction compensation
strategy was employed to alleviate the effect of static friction. To address parameters
uncertainties and nonlinear disturbances associated with iron-core permanent magnet linear
motor feed drive, an adaptive robust control strategy was proposed (Yao and Xu 2002). The
controller can adaptively estimate the three static friction components, that is, stiction, Coulomb
friction and viscous friction. The proposed controller can achieve guaranteed tracking accuracy
in the presence of nonlinear disturbances. Another approach to compensating friction in a linear
motor motion system was presented (Tan, Huang et al. 2002) , which is an adaptive sliding mode
controller.

Static friction models can not capture presliding displacement which is dominant for high
precison application (Armstrong-Helouvry, Dupont et al. 1994; Dupont, Hayward et al. 2002).A
new nonlinear analytic friction model, i.e., the LuGre model (Canudas de Wit, Olsson et al. 1995;
De Wit and Lischinsky 1997). The LuGre model combines the presliding behavior of the Dahl
model with the steady state friction characteristic in sliding regime, such as the Stribeck curve.
The strength of the dynamic LuGre friction model is the ability to capture a large number of
practically observed friction phenomena as described in (Canudas de Wit, Olsson et al. 1995; De
Wit and Lischinsky 1997). Therefore, the LuGre model serves as a good friction model for
machine tool feed drives, especially for the application where position accuracy requirements
may be down to submicron regime. The application of the LuGre model for friction modeling
and compensation in conventional ball screw feed drive system has been demonstrated in (Ro,
Shim et al. 2000).

The implementation of linear motors in machine tools can simplify the mechanical system in
machine tools. The problem with linear motors is that the cutting forces are directly reflected to
the motors due to directly coupling with severe effects on tracking accuracy. Alter and Tsao
(Alter and Tsao 1994) examined the issue of stable turning using linear DC motor feed drives.
They concluded that stability is not a problematic issue for turning, and that attention should be
focused on issues such as disturbance rejection and trajectory tracking. Alter and Tsao (Alter and
Tsu-Chin 1996; Alter and Tsu-Chin 1998) also examined H ∞ optimal feedback and feedforward
controllers as they apply to linear DC motors. In addition to position feedback, cutting force
feedback was examined. It was found that, in a practical system, both H ∞ optimal position
feedback and H ∞ optimal force feedback could increase dynamic stiffness. H ∞ optimal
feedforward controllers were found to be limited in practical systems, particularly in bandwidth,
due to modeling uncertainty. The application of linear motor feed drive for end milling process
have also been investigated (Choi and Tsao 2005). In their research, a MIMO H ∞ controller was

28
proposed and its effectiveness was shown by the lower tracking error than a conventional PID
controller over a wide range of cutting forces and feed rates.

For linear motor feed drives, besides external disturbances, parameter uncertainties problems
also occur. For example, the mass to be moved by the drive motor not only changes with the
mass of the workpiece, since the load is not known in advance and the robustification of the
motion controller is necessary. Van den Braembussche et al. (Van den Braembussche, Swevers
et al. 2001) have performed controller design and experimental validations for linear motor feed
drive axis. They compared the robustness of the control with respect to load changes up to 300%
in terms of tracking performance. They experimental demonstrated the robustness of the discrete
time sliding model control and enhanced the performance of H_infinity controller using an
improved weighting function.

In this paper, a modeled based sliding mode controller will designed according to the established
model of the linear motor feed drives for grinding machines. The merits of the model-based
sliding mode controller are two folds: (1) By using the authentic model of the linear motor feed
drive, the uncertainty intervals associated with the system model can be reduced as small as
possible. (2) The unaccounted model uncertainties and external disturbances, although already
minimized, can be handled by the sliding mode controller. Because of these merits, the controller
can achieve the robustness in presences of model uncertainties and external disturbances with the
smallest loss of tracking performance.

Linear Motor Feed Drives Controller Design

In this section, the controller design for linear motor drives will be presented after a brief review
of the system model, which is detailed in the modeling paper.

Modeling of Linear Motor

A current-controlled three phase linear motor with iron core is considered for the controller
design purpose. Figure 2 shows the model of a single axis linear motor feed drive for a grinding
application. In figure 2, u is the control signal; Ka is the current amplifier gain, K f is the motor
force constant, the mass is the equivalent mass of the whole drive. The friction, grinding force
and force ripple are the same as modeled in the modeling paper.

29
 Voltage Current Thrust force
Axis
u + Position
Ka Kf Mass
Control
Signal _
Current Amplifier Motor
_ _ Friction

Grinding Force

_
Force Ripple

Figure 17 Block diagram of linear motor model.

Model Based Sliding Mode Controller Design

In this section, sliding mode control design, which is based LuGre friction model, force ripple
model and grinding force model as established in the modeling paper, will be discussed. The idea
of sliding mode control was first initiated by utkin (Utkin 1977). The purpose of sliding mode
control is to enforcing the controlled system state trajectory to a chosen sliding mode trajectory,
on which system states will be asymptotically converge to the origin of the phase plane (Hung,
Gao et al. 1993). The sliding mode control approach is very attractive to system with
uncertainties, like the linear motor feed drives, due to its robustness and in many cases even
invariance characteristics, which means the system is complementally insensitive to parameter
uncertainties and external disturbances (Hung, Gao et al. 1993). Assuming all the system states
are available, although the average bristle deflection is immeasurable, the sliding mode control
design can be derived as follows:

First, the sliding surface can be defined as

.
S = e+ λ e (1)
where e is the tracking error evaluated by the difference of xd , the desired trajectory state , and
x , the actual state, as
e = xd − x (2)
Then, the reaching law method (Gao and Hung 1993) will be chosen to establish the reaching
condition due to its merits in reducing undesired chattering phoneme, which occurs in the
implementation of sliding mode control due to discontinuous switch (Utkin 1977). The reaching
law,
.
S = −η sgn( s ) − ks (4)
, consists of a constant rate term, −η sgn( s ) and a proportional rate term, − ks .

The requirement of the above reaching law on the control law, after some mathematical
manipulations, is found to be
m .. . . F f + Fg + Fr
u= ( xd + λ ( xd − x) + η sgn( s ) + ks ) + (6)
Ka K f Ka K f

30
where F f , Fg , and Fr are friction force, grinding force and force ripple, respectively.
To prove the effectives of the used reaching law, the Lyapunov function is defined as
1
V = S2 (7)
2
The time derivative of the above Lyapunov function can be evaluated by
. .
V = S S = −η S − kS 2 (8)
.
V is positive definite and V is negative definite (both η and k are positive ). According to the
second method of Lyapunov, the S will asymptotically tend to zero. Once on the sliding surface,
the tracking error will be guaranteed to converge to zero.

The derived controller law is implemented for the linear motor feed derive as shown in Figure 3,
where LM is the linear motor model as illustrated in figure 2.

+ Output
Sliding Mode
LM
Controller
Reference Input
_

Figure 18 Block diagram of sliding mode control strategy

Controller Simulation

In this section, the effectiveness of the proposed controller design will be illustrated by computer
simulations. The parameters used for the linear motor feed drive are documented in the modeling
paper. The design controller parameters are selected as
η = 5 k = 10 λ = 50

Position Tracking

The positioning performance of the sliding mode controller will be considered first for a
sinusoidal input to demonstrate the effectiveness of the proposed controller, and then three more
reference trajectory will be used in the simulations. For all these simulations,
only friction force is included in the system model.

31
 (a) (b)

(c) (d)

(e) (f)

Figure 19: The simulation results sliding mode control to desired sinusoidal input
(a) Desired positions. (b) Position tracking error. (c) Speed error. (d) Velocity error vs. position
error. (d) Control input. (e) Friction force

The simulation results to the desired sinusoidal input are shown in figure 4. The desired
sinusoidal input has a magnitude of 100 m and a period of 1second.The position error is shown
in figure 4 (b), which indicates a maximum tracking error of around 7000 µ m near the starting
point. The tracking error decreases very sharply towards zero, which means the controlled
system will follow the desired sinusoidal trajectory after a short time. The velocity error as
shown in figure 4 (c) also indicates that the trajectory will be perfectly followed. From figure 4
(d), it can seen that the controlled system states will move toward the sliding line, then once

32
reach it, both the position and velocity error will converge to zero as expected by the sliding
mode theory. Figure 4 (e) shows the high frequency switch of control input with bounded
magnitude. The model based sliding mode controller performed perfectly in the presence of
friction force as shown in figure 4 (f).

The effectiveness of the model based sliding mode control is demonstrated by the prefect
tracking of a desired sinusoidal input. In order to test the effectiveness of the controller with
different inputs. more extensive simulations will be carried out The response to a unit step input,
which jumps from zero at time instant 1 second to 100mm, will be examined. The simulations
results are shown in Figure 5.

(a) (b)

(c) (d)

Figure 20: The simulation results sliding mode control to 100 mm step input
(a) Comparison between desired and simulated positions. (b) Speed error. (c) Control input. (d)
Friction force

The simulated position is shown in figure 5 (a). it can be seen, around the time instant of 1
second, simulated response jumps to around 135 mm, and then dives to around -125 mm sharply.
The fast change of position will necessitate an extreme high control effort, as can be seen from
Figure 5 (c). Also, the speed change shown in Figure 5 (b) is significant, and as a result, the
resulted friction force will be big as illustrated in Figure 4 (d). The undesired response to step

33
input is believed to be ascribed to the unsmoothness of the input itself, which has a jump at time
instant of one second.

An acceleration limited trajectory limited trajectory profile, which is smoother than step input, is
selected to evaluate the performance of proposed controller. The simulations results are
described in Figure 6.
(a) (b)

(c) (d)

Figure
21: The simulation results sliding mode control to acceleration limited trajectory
(a) Simulated position. (b) Position tracking error. (c) Control input. (d) Friction force

Figure 6 (a) shows the acceleration, but not jerk, limited reference trajectory. The tracking error
is demonstrated in Figure 6 (b), from which a maximum tracking error of 4.4 µ m is identified.
Compared to the tracking errors for the step input case, an acceleration limited reference
trajectory improved the tracking performance significicantly. In addition, the friction force curve
is much nice than the step input case.
However, the control input as shown in Figure 6 (c) indicates a big control effort occurred at the
time instant of 0.5 second, which is a result the boundless jerk. To get a better tracking
performance, a jerk limited reference trajectory as shown in Figure 7 will be used for simulations.

34
 Figure
22 Jerk limited references trajectory used in simulation

The reference trajectory as shown in Figure 7, is generated as in (Tan, Huang et al. 2001).
A displacement of 100 mm was commanded with a maximum velocity of around 200 mm/s,
acceleration /deceleration values of 115 mm / s 3 , and a jerk limit of 240 mm / s 3 .
The simulated tracking performance for the jerk limited reference trajectory is shown in Figure 8.
(a) (b)

Figure 23: The simulation results sliding mode control to the jerk limited trajectory
(a) Simulated position tracking error (b) Control input.

The superior tracking performance using the jerk limited reference trajectory is indicated by
small tracking error as shown in figure 8 (a), where except a maximum absolute tracking error of
0.8 µ m , in most time the tracking errors are within a plus and minus 0.2 µ m range. The
magnitude of control input as shown in figure 8 (b) is very well bounded, with an upper limit less
than 4 and a lower limit larger than -2.Howver, the controller input switches very fast, which
may be a problem when implemented in real physical system.

35
Velocity tracking

When the linear motor feed drives are used for grinding applications, the velocity tracking
performance will play a major role in determining the quality, say, surface finish.
In view of the extreme importance of perfect velocity following, the velocity tracking
performance will be evaluated for the proposed controller design. The desired input for the
simulation is a ramp input with a slope of 1 µ m / s . In order to gets a realistic simulation, friction
force, grinding force and force ripple are included in the system model.

The simulation results for the ramp input are shown in figure 9. The effectiveness of the model
based controller in a low velocity tracking is demonstrated by the small velocity error as shown
in figure 9 (b), where the maximum tracking error is less than 150 nm/s.
A zoom view of figure 9 (b) will further discover that the tracking error will be less than 4 nm/s
after a period of 0.25 second. The extreme small position error as shown in figure 9 (a) also
confirms the effectiveness of the proposed controller for low velocity application in presences of
external disturbances such as friction, grinding force and force ripple. The friction force is shown
in figure 9 (d), from which it can be seen the friction force is proportional to time. This means
friction changes proportionally with respect to displacement since a constant velocity input is
used. The proportionality of friction with respect to displacement implies the motion of the feed
drive is close to presliding region which is captured by LuGre friction model (Canudas de Wit,
Olsson et al. 1995). The force ripple is shown in figure 9 (e), from which it can be seen the force
ripple changes very small in whole displacement range, which is reasonable because the whole
displacement is much less than the pole pitch of the linear motor. Grinding force is shown in
figure 9 (f), from which it can be seen the grinding force is more significant than friction and
force ripple for the considered case.

Conclusions

The paper investigates the effectiveness of a model-based sliding mode control for precision
tracking in linear motor feed drives for grinding machines. The model based sliding mode
control makes use of authentic linear motor feed drives modes with the considerations of friction
force, force ripple and grinding force, which are functionally critically for the implementation of
linear motor feed drives in grinding machines. The simulations results show that the system can
achieve submicron positioning performance.
For the low velocity tracking, an accuracy of 4nm/s can be achieved after 0.25 second duration in
the presence of friction, force ripple and grinding force.

36
 (a)
(b)

(c) (d)

(e) (f)

Figure 24: The simulation results sliding mode control to a ramp input
(a) Position error. (b) Velocity error. (c) Control input. (d) Friction force (e) Force ripple.(f)
Grinding force

References

37
Alter, D. M. and T.-C. Tsao (1994). "Stability of turning processes with actively controlled linear
motor feed drives." Journal of Engineering for Industry, Transactions of the ASME
116(3): 298-307
Alter, D. M. and T. Tsu-Chin (1996). "Control of linear motors for machine tool feed drives:
design and implementation of H infinity optimal feedback control." Transactions of the
ASME. Journal of Dynamic Systems, Measurement and Control 118(4): 649-56
Alter, D. M. and T. Tsu-Chin (1998). "Control of linear motors for machine tool feed drives:
experimental investigation of optimal feedforward tracking control." Transactions of the
ASME. Journal of Dynamic Systems, Measurement and Control 120(1): 137-41
Armstrong-Helouvry, B., P. Dupont, et al. (1994). "A survey of models, analysis tools and
compensation methods for the control of machines with friction." Automatica 30(7):
1083-138
Canudas de Wit, C., H. Olsson, et al. (1995). "A new model for control of systems with friction."
IEEE Transactions on Automatic Control 40(3): 419-25
Choi, C. and T.-C. Tsao (2005). "Control of linear motor machine tool feed drives for end
milling: Robust MIMO approach." Mechatronics 15(10): 1207-1224
De Wit, C. C. and P. Lischinsky (1997). "Adaptive friction compensation with partially known
dynamic friction model." International Journal of Adaptive Control and Signal Processing
11(1): 65-80
Dupont, P., V. Hayward, et al. (2002). "Single state elastoplastic friction models." IEEE
Transactions on Automatic Control 47(5): 787-92
Gao, W. and J. C. Hung (1993). "Variable structure control of nonlinear systems. A new
approach." IEEE Transactions on Industrial Electronics 40(1): 45-55
Hung, J. Y., W. Gao, et al. (1993). "Variable structure control: a survey." IEEE Transactions on
Industrial Electronics 40(1): 2-22
Otten, G., T. J. A. de Vries, et al. (1997). "Linear motor motion control using a learning
feedforward controller." IEEE/ASME Transactions on Mechatronics 2(3): 179-187
Ro, P. I., W. Shim, et al. (2000). "Robust friction compensation for submicrometer positioning
and tracking for a ball-screw-driven slide system." Precision Engineering 24(2): 160-173
Srinivasan, K. and T. Tsu Chin (1997). "Machine tool feed drives and their control-a survey of
the state of the art." Transactions of the ASME. Journal of Manufacturing Science and
Engineering 119(4B): 743-8
Tan, K. K., S. N. Huang, et al. (2001). "Adaptive robust motion control for precise trajectory
tracking applications." ISA Transactions 40(1): 57-71
Tan, K. K., S. N. Huang, et al. (2002). "Robust adaptive numerical compensation for friction and
force ripple in permanent-magnet linear motors." IEEE Transactions on Magnetics 38(1
II): 221-228
Tomizuka, M. (1987). "Zero phase error tracking algorithm for digital control." Transactions of
the ASME. Journal of Dynamic Systems, Measurement and Control 109(1): 65-8
Tung, E. D., G. Anwar, et al. (1993). "Low velocity friction compensation and feedforward
solution based on repetitive control." Journal of Dynamic Systems, Measurement and
Control, Transactions of the ASME 115(2A): 279-284
Utkin, V. I. (1977). "Variable structure systems with sliding modes." IEEE Transactions on
Automatic Control AC-22(2): 212-22

38
Van den Braembussche, P., J. Swevers, et al. (2001). "Design and experimental validation of
robust controllers for machine tool drives with linear motor." Mechatronics 11(5): 545-
562
Van Den Braembussche, P., J. Swevers, et al. (1996). "Accurate tracking control of linear
synchronous motor machine tool axes." Mechatronics 6(5): 507-521
Weck, M. and G. Ye (1990). "Sharp Corner Tracking Using the IKF Control Strategy." CIRP
Annals - Manufacturing Technology 39(1): 437-441
Yao, B. and L. Xu (2002). "Adaptive robust motion control of linear motors for precision
manufacturing." Mechatronics 12(4): 595-616

39