784 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 2, FEBRUARY 2010
Disturbance and Friction Compensations in Hard
Disk Drives Using Neural Networks
Chow Yin Lai, Student Member, IEEE, Frank L. Lewis, Fellow, IEEE, Venkatakrishnan Venkataramanan,
Xuemei Ren, Shuzhi Sam Ge, Fellow, IEEE, and Thomas Liew, Senior Member, IEEE
Abstract—In this paper, we show that by using two adaptive
neural networks (NNs), each of which is tailored for a specific task,
the tracking performance of the hard-disk-drive (HDD) actuator
can be significantly improved. The first NN utilizes accelerometer
signal to detect external vibrations and compensates for its effect
on HDD position via feedforward action. The second NN is de-
signed to compensate for pivot friction. The appealing advantage
of the NN compensators is that the design does not involve any in-
formation on the plant, sensor, disturbance dynamics, and friction
model. The stability of the proposed scheme is analyzed by the
Lyapunov criterion. Experimental results show that the tracking
performance of the HDDs can be improved significantly with the
use of the NN compensators as compared to the case without
compensation.
Index Terms—Disturbance feedforward, friction compensation,
hard disk drives (HDDs), neural networks (NNs).
I. INTRODUCTION
I
N CONFORMITY with the increase of the data density
on magnetic disk drives in recent years, the data track
width has reduced significantly. This means that the tracking
accuracy of the voice-coil-motor (VCM) actuator must be
improved. However, two major trends of the hard-disk-drive
(HDD) development are posing challenges to the quest for
better tracking performance of the VCM actuator. First, the
HDDs are increasingly used in mobile devices and thus are
subject to more external vibrations and shocks. Second, with
the current trend toward smaller form factors and smaller VCM
torque, the nonlinear pivot friction becomes more pronounced.
Thus, the effect of external disturbances and friction should be
explicitly taken into account during the design of HDDs servo
Manuscript received April 14, 2008; revised July 1, 2009. First published
July 24, 2009; current version published January 13, 2010. This work was
supported in part by the National Science Foundation (NSF) under Grant
ECCS-0801330, in part by the Army Research Office under Grant W91NF-
05-1-0314, in part by the Science and Engineering Research Council, Agency
for Science, Technology and Research (A*STAR) under Grant 052 101 0097,
and in part by the National Natural Science Foundation of China under
Grant 60974046.
C. Y. Lai is with the NUS Graduate School for Integrative Sciences and
Engineering, National University of Singapore, Singapore 117456 (e-mail:
g0601819@nus.edu.sg).
F. L. Lewis is with the Automation and Robotics Research Institute,
The University of Texas at Arlington, Fort Worth, TX 76118 USA (e-mail:
lewis@uta.edu).
V. Venkataramanan and T. Liew are with the Data Storage In-
stitute, A
∗
STAR, Singapore 117608 (e-mail: Venka_V@dsi.a-star.edu.sg;
LIEW_Yun_Fook@dsi.a-star.edu.sg).
X. Ren is with the Department of Automatic Control, Beijing Institute of
Technology, Beijing 100081, China (e-mail: xmren@bit.edu.cn).
S. S. Ge is with the Department of Electrical and Computer Engineering, Na-
tional University of Singapore, Singapore 117576 (e-mail: samge@nus.edu.sg).
Digital Object Identifier 10.1109/TIE.2009.2027257
system, so that the desired positioning accuracy can be achieved
even under the presence of external disturbances and friction.
To reduce the effect of the disturbances on the HDDs, a num-
ber of authors have proposed using accelerometers to measure
external disturbances and injecting the accelerometer signal to a
feedforward controller, which then outputs a feedforward signal
into the system [1]–[8]. Bando et al. [3] identified the transfer
function from external vibration to plant input via recursive
least square algorithm using the acceleration signal and the dis-
turbance signal which is estimated from disturbance observer.
Du et al. [4] applied the H
∞
method in state space to design
the feedforward controller to compensate for external vibration
impact on the positioning accuracy of the VCM actuator. The
identification of both the plant and the disturbance model is
considered by Pannu and Horowitz in [5], where they designed
an adaptive f eedforward controller acting as an add-on com-
pensator to an existing fixed compensator. In [6], Jinzenji et al.
used the dual accelerometers to detect angular acceleration in
HDD and compensate for its effects on the position accuracy. In
[7], Usui et al. established an adaptive feedforward control with
the finite-impulse-response (FIR) filter. However, the gradient
algorithm for updating the coefficients of the FIR controller
relies on the exact internal model. White and Tomizuka [8]
developed an infinite-impulse-response (IIR) filter and the FIR
filter based on the filtered-x least mean-square (LMS) algorithm
to compensate for the effects of the disturbances. To assure con-
vergence of the filtered-x LMS algorithm, however, the product
of the unknown dynamics and its estimated inverse needs to be
strictly positive real. It is evident that in almost all of the earlier
feedforward control schemes, the mathematical models of the
disturbance dynamics must be known or partly known, and the
sensor model and disturbance dynamics are usually restricted
to be linear. The need for the disturbance dynamics is a major
drawback because they can be time varying, nonlinear, and can
rarely be modeled accurately.
Friction can cause tracking errors, large settling time, and
overshoot. The methods for friction compensation can be di-
vided into two categories, namely, the model-based [9]–[14]
and the nonmodel-based compensations [15]–[19]. In the
model-based approach, various friction models are proposed,
their parameters identified either offline or adaptively, and
the output of the friction models injected into the plant to
cancel the true friction force. The drawback of the model-
based friction compensator is the reliance on an accurate model;
thus, a simple model does not provide satisfactory friction
compensation, whereas a highly accurate model is difficult to
parameterize. The nonmodel-based techniques include using
0278-0046/$26.00 © 2010 IEEE
剩余8页未读，继续阅读
资源评论
