Expert Systems with Applications 38 (2011) 11335–11346

Contents lists available at ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

Fuzzy control of an electrodynamic shaker for automotive

and aerospace vibration testing
K.P.S. Rana ⇑
Instrumentation and Control Engineering, Netaji Subhas Institute of Technology, Sector 3, Dwarka, New Delhi 110 078, India

a r t i c l e i n f o a b s t r a c t

Keywords: A fuzzy logic based digital time domain sinusoidal acceleration waveform amplitude controller for an
Electrodynamic shaker electrodynamic shaker is presented. The purpose of Fuzzy Logic Control (FLC) is to reproduce a pre-
Fuzzy logic controller defined sinusoidal acceleration amplitude profile (in amplitude, frequency and time) at the shaker table.
Active control Sinusoidal vibration profiles (sine and logarithmic sine sweep) are considered for a controlled vibration
Acceleration control
generation in typical automotive and aerospace testing. The difficulty in sine sweep testing is that the
Vibration control
non-rigid load dynamics are unknown and it can severely modify the shaker’s performance during sweep
test. Since a logarithmic frequency sweep is normally used, a controller needs to be robust to un-modeled
dynamics and also fast enough to hold the desired acceleration amplitude within predefined limits
throughout the sweep test. The controller structure is developed based on the usual power amplifier
technology. The control action is implemented on a waveform-by-waveform basis and a FLC is developed
in the LabVIEW environment on a PXI platform for real time control of the shaker. To attenuate the shaker
suspension mode resonance a compensator based on electromechanical model of the shaker is designed
and cascaded to FLC. The shaker model, suspension mode compensator design, FLC synthesis and exper-
imental implementation results are presented in this paper.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction measurement systems. Vibration testing also plays a major role

Vibration is one of the most classical phenomena and it has at-
tracted the minds of great scientists’ right from the days of Sir Isaac
Newton. A mechanical vibration in particular is used to study the
effects of vibration in structures and materials in wide variety of
applications. Generally, the purpose is to simulate the environment
where the Device Under Test (DUT) is used in service and is then
used to investigate if it will survive to the actual usage service con-
ditions. These applications mainly include stimulus response anal-
ysis, reliability testing, vibration sweep testing and resonance
study. Vibration testing refers to subjecting a device to some pre-
defined profile of mechanical vibrations in terms of amplitude
and frequency. Several industries like defense, space, railways, air-
ways and embedded hardware manufacturing, during critical
phase of testing of their products, require vibration tests for pre-
cisely defined profiles. In general there is always a great demand
for increasing the reliability of devices making use of critical com-
ponents. This is accomplished by intensive vibration tests because
this is the main phenomenon responsible for the durability and
serviceability study of various devices. In fact vibration tests have
become one of the essential and critical tests for all the test and
⇑ Tel.: +91 11 25099050; fax: +91 11 25099022.
E-mail address: kpsrana1@gmail.com
in identifying faults in the machineries and in the prevention of
failures leading to the mechanism of fault diagnostics (De Silva,
2000; McConnell, 1995; Tustin, 2005).
Sinusoid vibration testing is very important in the engineering
development to search the product resonances and to determine
the fatigue life. In the most common form, sine testing involves a
logarithmic frequency sweep holding the specified acceleration
magnitude at the base of the DUT. By sweeping the frequency in
either direction in the range of interest according to predefined
profile, adequate test reliability is attained. A PC based modern
vibration equipment normally used to execute close loop sine test-
ing comprises of an electrodynamic shaker, a power amplifier and
an acceleration measurement, controlling and monitoring system
as shown in Fig. 1. The vibration exciter transforms the electrical
energy supplied to it into the physical movements. The interface
to the PC is twofold. First, it is used for measurement of the gener-
ated vibration and second, it generates the base control signal to be
fed to the power amplifier. In the usual configuration, DUT is
mounted to the shakers’ table along with an accelerometer used
to measure the response acceleration. This feedback allows con-
trolling the input excitation signal so as to achieve the desired ref-
erence profile in acceleration magnitude, frequency and time. The
control platform excites the shaker using an appropriate power
amplifier by generating the required drive signal (Tustin, 2005).

0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.02.184
11336 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346
Computer control system for
vibration measurement and
control signal generation
Vibration sensor
Response signal
DUT
Power amplifier Table
Shaker
Drive signal Shaker input
Fig. 1. Vibration testing system components.
1.1. Brief bibliographical review of electrodynamic shaker control
mechanisms
This section gives a brief survey of the recent work done in this
area. The control technique of typical industrial shaker controllers
for sine testing is essentially a frequency domain approach, usually
referred to as transfer function equalisation, before starting the
test; the controller outputs a small broadband random excitation
signal to the shaker. Based on the measured response, the inverse
transfer function of the vibration generation process is estimated.
This information is then used to generate the control signal so as
to compensate the estimated process dynamics. During the test
execution, the controller is required to monitor and constantly up-
date the inverse transfer function as needed (LDS Test & Measure-
ment, 2005). In this scheme no prior very specific structural
parameters of the shaker, or of power amplifier, are required. As re-
ported in (Keller, 2002), this technique relies on the use of very
selective tracking filters to attenuate any frequency rather than
the controlled one, which improves the measurement accuracy
and keeps the focus in magnitude control instead of root mean
square (RMS). Most of the current industrial shaker controllers
have evolved from platforms used to build dynamic signal analys-
ers, so that the high signal-to-noise ratio and the wide dynamic
range of these instruments are also featured for shaker control.
However, despite all the desirable characteristics, the nature of
the transfer function equalisation technique and its implementa-
tion require relatively long processing time to manipulate large
blocks of data samples. Consequently, for sine testing, in particular,
the resulting delay in updating the drive signal in response to
changes in the input acceleration makes it difficult to achieve sta-
bility and good reference tracking under the effects of high Q res-
onances of non-rigid test loads, especially during the frequency
sweep.
Baoliang et al. proposed a Fast Fourier Transform (FFT) based
variable sampling rate method for sine magnitude control. Accord-
ing to simulation results, this approach has improved the ampli-
tude identification accuracy without using tracking filters, which
significantly increases the controller response and precision during
the frequency sweep. However verification and validation of pro-
posed method on a physical system was not presented (Baoliang
& Xia, 2003).
Chen et al. presented an analog acceleration amplitude control-
ler for an inverter-fed shaker using conventional PI control mech-
anism for sinusoid excitation based on time domain feedback.
Reference and disturbance feedforward control were proposed
and further developed for random tests. Robustness to parameter
variations were demonstrated experimentally. This scheme speci-
fies as drive signal not the shaker excitation voltage or current,
but an internal state of an amplifier, which strongly prevents its
application to the usual power amplifier technology in vibration
testing. The settling time of the proposed control logic is reported
as 1 s. Furthermore, the performance of sine sweep test as required
by the industry is not investigated and reported for desired sweep
rate and wider frequency range (Chen & Liaw, 1999; Liaw, Yu, &
Chen, 2002).
Uchiyama et al. achieved the electrodynamic shaker robust con-
trol by means of two controllers. The two control variables are
acceleration and displacement of the shaker’s table. Two control
variables are used in order to control a broad frequency band. Fre-
quencies where the response signals are larger than the noise, the
acceleration control is employed. For lower frequencies the dis-
placement control mechanism is employed. Both of these methods
are coupled in series. Finally the control mechanism was tested on
an electrodynamic shaker and some comparisons were drawn.
They further extended this concept by adding an adaptive filter
to the two degrees of freedom controller. The adaptive filter based
on the H1 filtering mechanism is employed. Performance of the
developed double control loop logic on the resonant load testing
was not reported (Fujita & Uchiyama, 2006; Uchiyama, Mukai, &
Fujita, 2009).
Flora et al. researched the digital acceleration controllers for
sinusoidal vibration tests using a switching-mode AC power source
(ACPS) for an electrodynamic shaker. The proposed scheme made
use of two control loops in interaction. One loop was used for
the acceleration regulation. The other loop was used for the ACPS
voltage output control. Robust model reference adaptive algorithm
was used to reduce the effects caused by the variations on the sys-
tem and to reduce the harmonic and resonant vibrations on the
test piece. The experimental results have shown that the proposed
system is able to attain a good performance in relatively narrow
frequency bands from 20 Hz to 200 Hz and the effect of load reso-
nance was not examined (Flora & Gründling, 2006). In a recent
work, they have also researched the digital acceleration controller
in time domain for sine vibration testing on electrodynamic shaker.
Cascade of compensators were used to attenuate the shakers’
structural resonances. Acceleration control was implemented
using a reference tracking compensator. They implemented sam-
ple-by-sample basis control mechanism for faster drive signal cor-
rection. The controller structure was developed based on the usual
vibration testing power amplifier technology in the automotive
range. Detailed procedure for shaker modeling and control logic
development was presented and supported by the experimental
results. An acceleration tracking error of 10% was reported for ref-
erence tracking in sine sweep test in automotive range of frequen-
cies. The controller performed well for rigid loads and had
limitation to dynamic test loads. No specific method or approach
for reference tracking controller tuning was mentioned (Flora &
Gründling, 2008).
Some other studies reported in the literature are on the model-
ing and controller design of the electrodynamic shaker systems.
Fair et al. carried out the analysis and design of shaker system.
They studied the electrical features of the moving coil vibration
generators (Fair & Bolton, 1993). Darie et al. have theoretically re-
searched computational aspects of electrodynamic system for its
controllability, stability and observability and proposed inclusion
of the device in automatic control systems. No specific controller
was proposed, implemented and experimentally verified (Darie,
Colosi, Vadan, & Balan, 1994). Macdonald et al. presented analysis
and control of a moving coil electrodynamic shaker. They devel-
oped a model and studied the controller design by pole place-
ments. The model was validated to the actual physical system
 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346
with bare table. No results on any type of load testing and sine
sweep profile are reported (Macdonald, Green, & Williams, 1993).
Some other work on analysis and design of electromagnetic mov-
ing coil vibration generators are presented (Lang, 1997; Lang &
Snyder, 2001).
The literature survey, on the electrodynamic shaker control,
presented above clearly indicates that the shaker system has com-
plex dynamics and there are certain un-modeled parameters and
nonlinearities inherent in terms of the damping, heating and sus-
pension system dynamics. There exist deep frequency dependen-
cies on the shaker’s armature electrical parameters owing to
the wide-frequency excitation. Additionally, test load dynamics
changes the mechanical parameters of the shaker, especially the
test mass. Shaker spring-mass suspension resonances and model-
ing errors equally contribute to make almost unusual open loop
operation. Furthermore, very limited types of vibration controllers
have really been studied and actually implemented for vibration
control on the shaker systems. The shaker system development
and so its controllers have been mainly in the industrial domain.
This is one of the reasons that the intelligent controllers based
on fuzzy logic or neural network, studied by the academicians for
several years and by now established to some extent, have not
been tried out on the complex system like vibration exciters. The
most recent and efficient, but relatively complex algorithms, like
fuzzy logic control is yet unexplored on the shakers. In contrast
to the industrial process control technologies, where large varieties
of controllers have been studied and implemented, the work avail-
able in shaker control domain is very limited. Bridging of such
technical gap is one of the main advances of the proposed research.
With the advent of high speed computers, computational com-
plexity is not a limitation anymore because the computing power
has been significantly improved even for high speed industrial pro-
cesses like shakers. This makes FLC an important alternative meth-
od for use in complex un-modeled and nonlinear systems like
shakers. As pointed out earlier for control of these devices most of
the work has been in the typical frequency domain control ap-
proach. In few cases (Chen & Liaw, 1999) classical PI acceleration
amplitude analog controller is implemented. It may be noted that
PID control, which shares around 90% of the process industrial con-
trollers’ space (Åström & Hägglund, 2001), is most suited for many
industrial processes. Furthermore it has set guidelines for tuning
and has witnessed a long journey in its developments over the years
(Chen, 1996; Heong, Chong, & Li, 2005). In case of shaker, some PID
variants like filters for oscillatory systems have also been experi-
mented and reported in the literature (Uchiyama et al., 2009).
The main aim of this work is to develop and implement an intel-
ligent fuzzy logic acceleration amplitude control algorithm on
waveform-by-waveform basis through close loop in typical auto-
motive and aerospace tests frequency range (10 Hz – 2 kHz) on
the single degree of freedom electrodynamic shaker. The main con-
trol variable is the shaker’s table acceleration amplitude. The error
and the change of error in the acceleration waveform amplitude
are minimized using intelligent FLC to achieve the control objec-
tives. The developed intelligent control mechanism makes it possi-
ble to perform tests, such as fatigue life, evaluation of resonant
frequencies, and replication of acquired acceleration’s field signals,
dwell sine test, fall and general modal analysis. A practical ap-
proach is presented and the developed mechanism can be scaled
to other industrial shakers with little effort.
This paper is organized as follows. Section 2 presents the nom-
inal linear dynamic model description and parameter identification
used for spring-mass suspension resonance compensation. Sec-
tion 3 presents the design of suspension mode compensator and
FLC. FLC synthesis, discretisation, normalization of input and out-
put variables and the resulted output control surface are pre-
sented. Tuning procedure used and the performance criteria
11337
adopted for controller evaluation are also presented. Section 4 pre-
sents an experimental setup in block diagram form. Detailed hard-
ware and software components used with their interfacing are
presented in the block diagram. A flow chart is also presented to
demonstrate the procedural steps involved in the experiment. Sec-
tion 5 presents the experimental results. Acceleration amplitude
tracking results for step change of reference, acceleration wave-
forms for some sample frequencies, waveforms for sweep test con-
ducted for rigid load are acquired and the results have been
presented. Finally, the sine sweep tests performance of FLC for res-
onant load is studied and presented. Section 6 presents the conclu-
sion and discussion.
2. Plant description and nominal model
A shaker system is a device which applies the mechanical vibra-
tions to DUT as per the predefined profile i.e. acceleration ampli-
tude, frequency and time. The moving coil principle is almost
universal in electrodynamic shakers for vibration testing. It offers
good linearity (i.e. thrust per current invariance with frequency
and with moving element instantaneous position) achievable over
a wide frequency range. The electrodynamic shaker works on the
basis of the electromagnetic force generated between two interact-
ing magnetic fields. The shaker used in this experiment comprises
of a permanent magnet giving constant magnetic field. The moving
coil suspended in a radial constant magnetic flux and connected to
a table structure in an assembly is normally referred to as the
armature. The moving coil has its own magnetic field proportional
to the current passing through the coil. When a current i is passed
through the moving coil, an axial force f is produced and transmit-
ted to the table where DUT is mounted. Positioning of the coil and
associated fixture table assembly is attended by means of a flexible
suspension arrangement which constraints the movement to axial
direction only and supports the weight of external load. In electro-
dynamic shakers for constant amplitude sinusoidal input signal to
armature, the vibration amplitude changes significantly with fre-
quency, so the power amplifier used to drive the shaker is gov-
erned by a control system. This monitors the mechanical motion,
usually by an accelerometer mounted on the table or the test ob-
jects itself, and corrects for amplitude variation as the drive fre-
quency is swept. The control mechanism in the electrodynamic
Mechanical
Model
K C
Shaker table, mass M x
Sensor DUT
f =αi
R(ω ) L(ω ) e
Electrical
Model
dx
v e=α
i dt
Fig. 2. Electromechanical model of shaker.
11338 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346

shaker is achieved by controlling the voltage applied to the arma-
ture (Lang, 1997; Lang & Snyder, 2001).
An electromechanical dynamic behavior of the shaker table can
be represented by the equivalent circuit as shown in Fig. 2. Under
the assumption of low and medium frequency operating range, as
required in automotive and aerospace testing, of a wide bandwidth
shaker with resonance free load, the test load mass, table and the
moving coil dynamics can be approximated to single body forming
a single degree of freedom spring–mass–damper system. The force
generated by the moving coil is given by Eq. (1):
f ¼ Bil ¼ ai
where f is the generated shaker force, N; B is the magnetic flux den-
sity, T; l is the average length of winding in coil, m; i is the coil cur-
rent, A; a is the ratio of thrust to coil current, N/A.
The mechanical model approximated to Fig. 2 uses a simple
spring–mass lumped parameter approach, to give a representation
as shown in Eq. (2):
2
d x dx
f ¼ ðM þ mÞ 2
þC þ Kx
dt dt
where M is the moving mass of the shaker (armature). It includes
the table and the coil. The DUT has a mass m. K is the shaker suspen-
sion stiffness and C is the damping constant. The relative displace-
ment of armature with respect to the shaker body is denoted by x.
Furthermore, the induced emf in a coil moving in a constant homo-
geneous magnetic field B with velocity v0 is given by Eq. (3):
dx
e ¼ Blv 0 ¼ a
dt
The electrical circuit can be expressed as:
di dx
v ¼ LðxÞ þ RðxÞi þ a
dt dt
The resistance and inductance have deep non-linear frequency
dependencies (Chen & Liaw, 1999; Flora & Gründling, 2008; Lang,
1997; Macdonald et al., 1993).
2.1. Mechanical parameters identification
A shaker of wide bandwidth, working in relatively low and
medium frequency range, can be approximated to single degree
of freedom spring–mass–damper system. There is no relative mo-
tion between the table and the coil. Under the assumption of time
invariant system, a linear representation of the transfer function
G(s) from coil current i to table acceleration a is reduced from
Eqs. (1) and (2).
aðsÞ as2
GMþm ðsÞ ¼ ¼ ð5Þ
iðsÞ ðM þ mÞs2 þ Cs þ K
At a bare table, i.e. m = 0
are from M/S Ling Dynamic Systems. Information about the arma-
ture mass M and the design parameters of the shaker suspension
system K and C may be extracted from a shaker by using it as a
vibration sensor, rather than as an exciter (Lang, 1997). In such an
experiment, the voltage output of the moving coil is monitored
while the shaker is caused to vibrate due to transient mechanical
input. Measuring the dynamic response of the shaker with its table
bare and, again, with a known mass attached to the table permits
the suspension mode characteristics to be determined. The shaker
is forced against its lower stroke-stop and held under pressure.
Now it is abruptly released to allow the shaker to ring. On a shaker,
ð1Þ
such as used in this experiment, the small human hand force was an
ideal tool. The coil voltage is recorded with the help of a high speed
data acquisition (DAQ) card and FFT analysis is carried out. Fig. 3
shows the time histories resulting from a ‘‘pluck’’ with the table bare
and with an attached known mass-load. In the experiment con-
ducted, the differentiating load is 202 g. The resulted response of
the shaker has a classic single degree-of-freedom form – it generated
decreasing sinusoid table movement with an exponential envelope.
Furthermore, attaching a mass reduces the frequency of the ring-
ð2Þ
down and decreases the rapidity of exponential collapse; both
effects are anticipated from a simple spring–mass–damper model.
Fig. 4 shows the FFT results of the unburdened and mass loaded
ring-downs. The frequency at peak response indicates the natural
frequency of the shaker’s suspension system (without bias), as
the measured voltage is proportional to the armature velocity.
Resonant frequency f1 of 39.62 Hz for the unloaded shaker and a
mass-loaded frequency f2 of 29.32 Hz, with the additional mass
m attached, is estimated form the FFT analysis. The half-power
ð3Þ
(3 dB) points bounding f1 were found to be separated frequency
by 3.3 Hz. These observations permit evaluating the mass, stiffness
and damping coefficient of the shaker suspension in accordance
with the following equations. Table 1 lists the identified shaker
ð4Þ
parameters.
f12
M¼m ð9Þ
f12  f22
K ¼ ð2pf1 Þ2 M ð10Þ
C ¼ 2pDf3 dB M ð11Þ
2.2. Electrical parameter identification
Both the resistance R(x) and the inductance L(x) have a non-
linear frequency dependency primarily due to the action of pole
plating, which has the effect of increasing the resistance and
decreasing the inductance with frequency. In addition, these
electrical parameters are subjected to secondary effects such as

aðsÞ as2
GM ðsÞ ¼ ¼ 2
ð6Þ
iðsÞ Ms þ Cs þ K
This leads to the natural resonant frequencies in two different load
conditions as:
rffiffiffiffiffi
K
xNðMÞ ¼ ð7Þ
M
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
K
xNðMþmÞ ¼ ð8Þ
ðM þ mÞ

For the experimental validation of the proposed acceleration con-

troller, an electrodynamic shaker (Model No. V408) driven by the
power amplifier (Model No. PA100E) was used. Both the products Fig. 3. Time-histories of shaker coil voltage generated by ‘‘plucking’’ the shaker.
 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346 11339

compensator designed on the basis of the above identified shaker

nominal model. The compensator makes the shaker frequency re-
sponse nearly flat in the region of interest. Then a FLC is designed
to track the acceleration waveform magnitude and to cater to the
un-modeled component of the shaker dynamics and the effects
of electrical parameter variations. The control mechanism is imple-
mented on waveform-by-waveform basis. In Fig. 5 r(k) represents
the reference amplitude of acceleration of kth waveform at a given
frequency. The response amplitude of the acceleration waveform is
represented by y(k). The output of the controller is the excitation
signal waveform amplitude. Thus an excitation signal in the form
of a complete waveform gets generated and the response wave-
Fig. 4. FFT results. form is acquired with the help of a DAQ card. A low pass filter is
used to attenuate measurement noise.

Table 1 3.1. Suspension mode compensator

Identified mechanical parameters.

S. no. Parameter Value Based on the approximated and identified nominal electrome-
1 Armature mass, M 217 g chanical dynamic model of the shaker and assuming the power
2 Suspension damping coefficient, C 4.42 Ns/m amplifier nominal gain K0 , the linear modeled part of the plant
3 Suspension stiffness, K 13482.96 N/m (generated acceleration to applied voltage) can be reduced from
4 Bare table resonant frequency, xN(M) 39.62 Hz
Eqs. (4) and (5).
5 Force generation constant 10.8 N/A
aðsÞ K 0 as2
GðsÞ ¼ ¼ ð12Þ
thermal and eddy currents within the coil (Chen & Liaw, 1999;
v ðsÞ ½ðM þ mÞs2 þ Cs þ KR þ a2 s
Flora & Gründling, 2008; Lang, 1997; Macdonald et al., 1993). Typ- Based on the inverse of this equation, a suspension mode compen-
ically at the frequencies of interest (2 kHz), the inductance effect sator is defined as:
is much lower to the resistance and hence the electrical parameter
identification is approximated to only resistance. The resistance of ½ðM þ mÞs2 þ Cs þ KR þ a2 s
G0 ðsÞ ¼ ð13Þ
the coil is derived from the measured value of the current and the K 0 as2
voltage applied to it. It is estimated as 2.92 X at operating fre- At low frequencies the implementation of Eq. (13) would drive the
quency of 2 kHz. The variations of these parameters along with excitation voltage to saturation. As a consequence, the pole at s = 0
the shaker and the power amplifier dynamics will be corrected are replaced by two poles at s = 1/l, where l is a positive design
by a FLC at run time as described in next section. Force generation parameter (Chen & Liaw, 1999; Flora & Gründling, 2008).
constant is estimated from the measured acceleration for bare ta-
ble while exciting the shaker by a known voltage. The above esti- ½ðM þ mÞs2 þ Cs þ KR þ a2 s
G0 ðsÞ ¼ l2 ð14Þ
mated value of resistance is used to find the current in the K 0 aðls þ 1Þ2
armature.

3.2. Reference tracking fuzzy controller design

3. Acceleration controller
Emerging intelligent techniques have been developed and
The block diagram of the proposed digital acceleration control extensively used to improve or to replace conventional control
mechanism is shown in Fig. 5. The controller comprises of two techniques, because these techniques do not require a precise
components. The FLC and the suspension mode compensator both model. One of the intelligent techniques, fuzzy logic, developed
coupled in cascade. The compensator accounts for the identified by L.A. Zadeh, is applied for controller design in many applications.
linear model of the shaker and it modulates the FLC output for FLCs have shown some success, but there is a significant need to
an excitation frequency. The controller aims to control the ampli- evaluate their real time performance on specific experiments. Such
tude of the shaker exciting waveform signal for achieving the de- evaluations help to determine the performance of the new intelli-
sired acceleration profile at the shaker table. The dynamics of the gent control method and provide engineers with general guidelines
shaker suspension leading to different performance of the shaker on how to apply them to more complex real-world applications
system at different frequencies, particularly in low frequency (Kuswadi, 2001; Passino, 1993; Passino, 1996, chap. 1; Zumberge
region, is compensated with the help of a suspension mode & Passino, 1998). FLC has emerged as one of the most active and
useful research areas. That is why, the FLCs have been successfully
applied for control of various physical processes. Basically there are
r(k) e(k)
+ Controller Waveform Power Plant two approaches to a fuzzy controller design: an expert approach
Generator Amplifier and a control engineering approach. In the first, the fuzzy control-
- ler structure and parameters choice are assumed to be the respon-
y(k)
sibility of the experts. Consequently, design and performance of a
fuzzy controller depend mainly on the knowledge and experience
of the experts, or intuition and professional feeling of the designer.
This dependence, which is considered far from systematic and reli-
Acceleration Low Pass Vibration Sensor
Amplitude Filter Waveform able, is the flaw of this approach. However, this approach could as-
Estimation Acquisition sist in constructing a fuzzy model or an initial version of a fuzzy
controller. The second approach supposes an application of the
Fig. 5. Proposed acceleration control scheme. knowledge of control engineering and a design of a fuzzy controller
11340 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346

in some aspects similar to the conventional design with the param- waveform amplitude error e(k) and its change of the error De(k)
eter’s choice, depending on the information of their influence on are required and the output of the FLC is the incremental change
the controller performance (Lee, 1990a, 1990b). in the controller output of excitation waveform to be fed to the
The present research work is focused on the design and imple- power amplifier. Increasing the number of inputs to FLC would re-
mentation of a customized fuzzy PI controller on an electrody- quire more processing time for the larger rule base size and hence
namic shaker for acceleration waveform amplitude control on would force a constraint on the implementable update rate in accel-
waveform basis. The error in the acceleration waveform amplitude eration amplitude control. Smaller computational delays also en-
and its change is minimized to achieve the control objectives. Fuz- able the shaker control at high frequencies. In order to make the
zy PI controller is chosen and is designed on the basis of the clas- computations faster a customized fuzzy PI controller was chosen
sical discrete PI controller structure, from which the fuzzy control rather than other combinations such as fuzzy PI + fuzzy PD or fuzzy
law is derived (Lee, 1990a, 1990b). A conventional PI controller fol- PID itself (Carvajal, Chen, & Ogmen, 2000; Pivonka, 2002).
lows the following control low: A physical meaning of the parameters for the fuzzy PI controller
 Z  remains the same like that for the classical PI controller. Fig. 6
1
uPI ðtÞ ¼ K C eðtÞ þ eðtÞdt ð15Þ shows the block diagram representation of the proposed controller
sI implementation on the acceleration amplitude control. The con-
where K C is the proportional constant of PI controller, sI is the inte- troller takes two fuzzy variable e(k) and De(k) and it gives an incre-
gral time constant, e(t) is error in the acceleration waveform ampli- mental change of the control action.
tude, e(t) = r(t)  y(t), r(t) is the desired acceleration amplitude, y(t) Implementation of the FLC requires some preprocessing. Vari-
is the output obtained from the shaker process and uPI ðtÞ is the out- ous tasks for realization of FLC are discussed as follows. Fig. 7
put of the controller. On differentiating Eq. (15) and putting into a shows the schematic of various modules required for FLC
discrete form, implementation.
 
duPI ðtÞ deðtÞ 1
¼ KC þ eðtÞ 3.2.1. Fuzzification
dt dt sI Input and output variables of the FLC are usually quantized into
In a discrete form, for a given sampling period Ts, the above equa- sets of classes defined by linguistic labels. For this experimenta-
tion can be written as: tion, both the inputs, e(k) and De(k), were quantized into five fuzzy
  sets, namely: PB – Positive Big, PM – Positive Medium, ZE – Zero,
½uPI ðkÞ  uPI ðk  1Þ ½eðkÞ  eðk  1Þ 1 NM – Negative Medium and NB – Negative Big. The output fuzzy
¼ KC þ eðkÞ
TS TS sI variable was quantized into seven fuzzy sets, namely: PB – Positive
Big, PM – Positive Medium, PS - Positive Small, ZE – Zero, NS – Neg-
or,
  ative Small, NM – Negative Medium and NB – Negative Big. Seven
TS sets are chosen for output variable in order to have precise correc-
DuPI ðkÞ ¼ K C DeðkÞ þ eðkÞ ;
sI tion to the acceleration amplitude. The quantization of the fuzzy
variables is carried out in the normalized unit range of [1, 1] for
or,
inputs as well as for output. The membership functions for e(k),
DuPI ðkÞ ¼ K C DeðkÞ þ K 0C eðkÞ ð16Þ De(k) and DuPI(k) are all of triangular types, with 50% overlap as
shown in Figs. 8 and 9. Triangular membership function is chosen
where as it is the most economical one. The computational requirements
DuPI ðkÞ ¼ uPI ðkÞ  uPI ðk  1Þ are least in triangular membership functions and these are most
DeðkÞ ¼ eðkÞ  eðk  1Þ suited for real time control applications.

KC is defined as the gain P and K 0C ¼ K C TsSI is the gain I.
The denormalized output of fuzzy controller is given by KUPID
uPI(k), where, gain KUPI denormalizes the normalized incremental
fuzzy output, DuPI(k). The controlled output thus obtained is as:
uPI ðkÞ ¼ uPI ðk  1Þ þ K UPI DuPI ðkÞ
In this work the control action is implemented on waveform basis
and hence the amplitude of the excitation signal to the power
amplifier is controlled to achieve the desired objectives. To imple-
ment the proposed acceleration waveform amplitude fuzzy PI con-
troller Eq. (16), two inputs namely the instantaneous acceleration
3.2.2. Rule base
The rule base for FLC can be imagined to be a two dimensional
matrix as summarized in Table 2. The rows represent the various
linguistic values that e(k) can take and columns indicate the vari-
ous values of De(k). The entries in this matrix would be the control
ð17Þ
action that has to be taken is described in the linguistic terms. The
control action is calculated based upon the experimentally ob-
served process reaction (Lee, 1990a, 1990b). The antecedent pairs
in the rule structure are connected by a logical ‘AND’ operation
which is considered here as algebraic product triangular norm. It
was chosen as it takes into account the effect of all the inputs in

Gain P
∫
e(k) Fuzzy PI Gain Suspension
+ KC Controller KUPI Mode Power Shaker
Compensator Amplifier
r(k)
- Gain I uPI(k)
Δe(k) ΔuPI(k)
K’C
y(k)

Low Pass Filter Vibration Sensor

Fig. 6. FLC implementation block diagram for acceleration amplitude control.

 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346 11341

Fuzzification Module
Defuzzification Module
Scaling Fuzzifier Inference Engine Defuzzifier Scaling

Scaling Factor Membership Functions Membership Functions Scaling Factor

Fuzzy Rule base

Database Knowledgebase Database

Fig. 7. Various modules required for FLC implementation.

Table 2
Rule base for fuzzy PI controller.

Change of error (Δe)

NB NM ZE PM PB

NB NL NM NM NS ZE
NM NM NS NS ZE PS
error (e) ZE NM NS ZE PS PM
PM NS ZE PS PS PM
PB ZE PS PM PM PL
Fig. 8. Triangular membership functions for inputs.

Fig. 9. Triangular membership functions for output.

Fig. 10. Control surface of the FLC.

comparison to the other methods of inference. It may be noticed

that the control rules are non-linear as the output fuzzy sets are
not linearly related to the input fuzzy sets. Since each input vari-
able will have two linguistic terms for each value of the normalized
input, and there are two input variables. Thus four rules get fired in
each case.
3.2.3. Fuzzy inference engine
The basic function of the fuzzy inference engine is to compute
the overall value of the control output variable based on the indi-
vidual contribution of each rule in the rule base. For the present re-
search work, Mamdani minimum inference mechanism has been
used. The differences in using the various implication techniques
are described in (Smith & Shen, 1998).
3.2.4. Defuzzification
This module converts the set of modified controller output va-
lue into a single crisp value. There are many procedures outlined
in the literature for defuzzification which are centre of gravity/
area, centre of mass, centre of largest area, first of maxima, middle
of maxima, and height. Of these, the centre of gravity (COG) is the
most efficient for the fact that it gives a defuzzified output which
conveys the real meaning of the action that has to be taken at that
instant. So, in the present work, the center of gravity defuzzifica-
tion method is used to defuzzify the fuzzy sets into a crisp control
signal (Tang, Man, Chen, & Kwong, 2001). Fig. 10 shows the fuzzy
surface formed by the various values of the two fuzzy inputs
namely, e(k) and De(k) and a fuzzy controller output DuPI(k).
3.2.5. Tuning of the controllers
The tuning of a fuzzy controller is often compared to the tuning
of a PID, stressing the large number of the fuzzy controller param-
eters, compared to the three gains of a PID. In the experiment con-
ducted the tuning was done manually by optimizing the close loop
step response for minimum overshoot at some sample frequencies
(Li, 1997).
11342 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346
3.2.6. Performance criteria
For controlling the acceleration amplitude of an electrodynamic
shaker system the main performance evaluation criterion were the
peak overshoot, rise time and settling time. This is because in the
vibration testing the profile is defined for a given acceleration
amplitude and frequency. Minimizing the settling time would give
faster acquisition of the desired profile. Thus, the main factors
which have been considered for evaluating the performance are
the overshoot and settling time. Both these parameters are con-
trolled within the 12% accuracy for the bare table and with some
small rigid load of 85 g. The main settings which had to be done
in the system were for the desired acceleration amplitude for given
frequency for reference tracking.
3.2.7. Low pass filter
To attenuate the effect of measurement noise, a second order
Butterworth low pass filter having the cut off frequency equal to
the frequency of shaker excitation signal is used.
4. Controller implementation and experimental setup
This section explains the details of the acceleration controller
implementation and experimental setup used to carry out the
experiment. Fig. 11 shows the flow chat for controller implementa-
tion. Fig. 12 shows the experimental setup in block diagram form.
A general purpose DAQ card (NI-PXI-6040E) is used on the PXI con-
troller configured in real time mode. The used DAQ has multiple in-
put output channels, which uses 12 bits resolution for ADC and
DAC. The input and output voltage limits are ±10 V and the sam-
Start
Set various parameters
(Amplifier gain, Accelerometer sensitivity,
DAQ card sampling rate for input/output,
Conditioning module channels, PZT
excitation current, Fuzzy gains, Desired
profile - acceleration amplitude and
frequency for sine or sweep rate etc.)
Excite the shaker for a single waveform of
pre-defined frequency and amplitude at
given sampling rate
Acquire the response acceleration waveform
signal and estimate the amplitude
Find the error e(k) and change of error Δe(k)
of acceleration waveform amplitude
Submit the above to FLC for finding the
change in controller output/excitation
waveform amplitude
Update the shaker excitation waveform
amplitude according to designed
compensator and excite the shaker for a
single waveform
Fig. 11. Flow chart for shaker acceleration control.
PXI controller (NI-8186) with NI-DAQ-6040E
Card and LabVIEW-RT 8.6 software package
Vibration Signal
Power Amplifier
Conditioner
(LDS-PA100E)
(SCXI1000+SCXI1530)
To
Kistler make ICP-8704, Armature
Accelerometer
Shaker
(LDS-V408)
Fig. 12. Experimental setup.
pling frequency varies from DC to 250 kHz for input and from DC
to 1 MHz for output. The DAQ card has a dynamic range of above
85 dB. Shear type accelerometer (Model 8704, ICP based, 4 mA
excitation, acceleration range ±50 g, sensitivity 100 mV/g, 10 kHz
frequency range and 7.5 g mass by M/S Kistler make) was used
to sense the acceleration. For conditioning of the acceleration sig-
nal, NI-SCXI-1000 base chassis along with the NI-SCXI-1530 accel-
erometer input module was used. A sampling rate of 40 kS/s was
set for both the input and the output. The gain of the power ampli-
fier was set at 4 at 1 kHz excitation. A LabVIEW code on PXI plat-
form is developed using the modular function based approach of
FLC development (Mohan & Sinha, 2008). The PXI platform is con-
figured in real time mode for synchronized waveform based exci-
tation, acceleration acquisition and control of the shaker,
implementing all the modules as detailed above. The FLC and the
compensator were cascaded as explained earlier. The design
parameter of suspension mode compensator was set to 0.159. Lab-
VIEW offers good library of digital filters and waveform processing
utilities.
A second order Butterworth low pass filter, as mentioned ear-
lier, was used for attenuating the measurement noise. Amplitude
of the acceleration waveform was estimated using peak detection
method.
The timing aspect of FLC, compensator, filter and the other com-
ponents were also studied. FLC is normally more resource demand-
ing. In this experimentation the developed customized FLC,
compensator and filter produced a delay of 25 ls per iteration. It
may be noted that this computation time is much smaller than
the time period of the highest frequency (2 kHz) to be controlled
for the automotive and aerospace testing. Since the corrective ac-
tion is implemented on waveform basis, the 10 Hz excitation gets
corrected at 100 ms period and on the other hand for the 2 kHz
excitation, it is 500 ls.
5. Experimental results
To characterize the used components in the experimentation
some open loop tests were conducted initially. The open loop fre-
quency response of the power amplifier and the shaker and power
amplifier combination were measured and presented. Effect of
voltage variation on the acceleration amplitude generated by the
shaker is also studied at a given frequency. All these tests were re-
quired to know the capabilities and limitations of the used compo-
nents. Fig. 13 shows the frequency response of the power amplifier.
As seen the gain of the power amplifier is significantly reduced at
lower frequencies typically below 100 Hz. Further, studies on the
shaker and the power amplifier combination frequency response
at a given excitation voltage (2 V) is conducted. Frequency re-
sponse of the shaker and power amplifier combination is shown
in Fig. 14. As expected the two resonances are observed at around
 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346 11343

Fig. 13. Frequency response of the power amplifier,

Fig. 16. Step response of bare table.

Fig. 14. Measured frequency response of the shaker with power amplifier.

Fig. 17. Step response of loaded table.

90 Hz and 11.5 kHz respectively. The lower resonance is attributed

to the resonance of armature mass on its suspension system. The
upper frequency break point corresponds to the axial resonance
of armature which arises from the action of the adhesive joint be-
tween table and coil. Some localized resonances which are distrib-
uted at intervals in the mid-band region are parasitic mass-spring
oscillations which develop in the folded cantilever construction of
the flexible suspension system. These variations in the frequency
response of the shaker are attributed to complex dynamics of the
shaker systems. The effect of voltage variation at a given frequency
(1 kHz) excitation were also studied and presented in Fig. 15. As
seen the response to voltage variation is nearly linear.
The inputs for various gains of the fuzzy controller were pro-
vided as required. Few trials were made and the controller gains
were manually optimized for minimum overshoot. Good accelera- Fig. 18. Step response of bare and loaded table at 10 Hz.
tion tracking results were obtained for Kc = 0.06, K0 c = 0.12 and
KUPI = 1.7. Figs. 16–18 show the reference tracking performance
of the FLC controlled shaker for bare and loaded table at some se- Table 3
lect frequencies. At a given frequency the reference acceleration Step response of bare table.
amplitude was set for 1 g and 2 g alternatively several times for
Frequency Rise time Settling time Overshoot
the frequencies greater than or equal to 20 Hz. For 10 Hz excitation (Hz) (iteration) (iteration) (%)
these reference settings were not permitted by the used shaker, so,
10 9 13 0.0
a smaller reference acceleration range of 0.5–1 g was used to test 20 8 11 0.0
the performance of the shaker at 10 Hz. The rigid test load 500 8 12 0.0
used was of 85 g. The plots show the variations of the filtered 1000 8 12 0.0
1500 8 12 0.0
2000 8 12 0.0

acceleration amplitude (step response) versus acceleration wave-

Fig. 15. Effect of voltage variation.
form iteration number. Tables 3 and 4 summarize the performance
results for rise time, settling time and overshoot for all these inves-
tigated cases. Rise time is defined as the time taken by the shaker
to acquire 90% of the final acceleration value. Similarly, settling
time is defined as the time taken by the shaker to achieve the 2%
error band of the final acceleration value. From these plots it can
be noted that the FLC performs almost similar at the chosen sample
11344 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346

Table 4
Step response of loaded table.

Frequency Rise time Settling time Overshoot

(Hz) (iteration) (iteration) (%)
10 9 13 0.0
20 8 11 0.0
500 8 11 0.0
1000 8 12 0.0
1500 8 12 0.0
2000 8 13 0.0

Fig. 23. Reference and the measured acceleration waveforms at 2 kHz for loaded
table.

frequencies and no overshoot is observed. Fig. 19 shows the FFT re-

sults for 2 kHz excitation of shaker for 1 g reference acceleration.
Figs. 20–23 show the reference and the filtered acceleration
waveforms for 10 Hz and 2 kHz excitation under two different load
conditions. A good acceleration magnitude tracking is observed in
all the studied cases under two different load conditions.
To evaluate the detailed experimental performance of the pro-
Fig. 19. FFT results for 2 kHz excitation.
posed controller in the entire frequency range of interest, swept
sine frequency test were conducted under different load condi-
tions. Considering swept sine frequency tests, the controller re-
sponse was analyzed keeping the magnitude of the reference
acceleration waveform constant and equal to 1 g. Then the fre-
quency was logarithmically swept from 10 Hz to 2 kHz to 10 Hz
at the sweep rate of 2 octave/min. Fig. 24 shows the time variation
of the logarithmically swept frequency. Fig. 25 shows the relative
acceleration magnitude (relative to the reference acceleration of
1 g) for bare table for a sweep cycle from 10 Hz to 2 kHz. As can
be seen, small distortions are observed in the lower frequency re-
gion (below 100 Hz). These distortions are mainly due to the vari-
ations in the power amplifier gain in lower frequency region. As
Fig. 20. Reference and the measured acceleration waveforms at 10 Hz for bare
verified, the relative amplitude tracking error of less than 12% dur-
table. ing the whole sweep cycle. The warning limit and the aborting

Fig. 24. Frequency profile.

Fig. 21. Reference and the measured acceleration waveforms at 10 Hz for loaded
table.

Fig. 22. Reference and the measured acceleration waveforms at 2 kHz for bare
table. Fig. 25. Relative magnitude profile during swept frequency test of bare table.
 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346
boundaries are also shown in the graph to compare the available
margin with the proposed controller. It may be noted that the
industrial shaker control usually require the 1 octave/min sweep
rate with the ±3 dB warning limits and ±6 dB aborting boundaries.
For all the conducted sweep tests the shaker was allowed to settle
for the desired acceleration at 10 Hz before the start of the sweep
cycle. Next the table was loaded by 85 g rigid load and sweep test
was performed. Fig. 26 shows the reference and the measured
acceleration waveforms for loaded table. Fig. 27 shows the corre-
sponding drive voltage signal waveforms. For these two plots data
was collected at 1 Hz frequency interval during the sweep cycle.
Fig. 28 shows the magnitude profile in linear units for loaded table.
Fig. 29 shows the relative magnitude profile for loaded table. It is
noted that the relative acceleration tracking error is less than 8%.
Furthermore, the loaded table tracking is slightly better than the
bare table.
An ultimate test for evaluating the controller’s robustness was
performed for a resonant load. In this test, a resonant load of mass
Fig. 26. Reference and measured acceleration waveforms.
Fig. 27. Drive signal.
Fig. 28. Measured acceleration magnitude profile of loaded table.
11345
Fig. 29. Relative acceleration magnitude profile of loaded table.
Fig. 30. Relative magnitude profile during swept frequency test of resonant load.
equal to 85 g was mounted on the table. The swept sine response
was measured as shown in Fig. 30. As seen the performance
of the sweep test was significantly modified. Specifically at
22.99 Hz, a high Q resonant mode is found and close loop instabil-
ity is disclosed to some extent. Thus, despite a good tracking
performance at bare table and with rigid load the stability robust-
ness to resonant load is somewhat poorly accomplished. However,
here too the introduced error is well within the warning limits
and much smaller as compared to the error reported in the other
literature. Other peaks, as observed, show the higher order modes
of the resonant load at 144.08 Hz, 403.01 Hz and 789.8 Hz
respectively.
6. Conclusion and discussion
This paper has presented a fuzzy logic based intelligent time do-
main digital acceleration amplitude controller for sine vibration
testing using electrodynamic shaker in aerospace and automotive
tests frequency range. The shaker was excited on a complete wave-
form basis and the amplitude of the generated acceleration was
controlled using FLC by minimizing the error and change in error.
Fuzzy logic control was derived from the classical PI technique and
a step-by-step practical procedure was presented to control the
shaker. Experimental results have demonstrated that this solution
is capable of guaranteeing good reference tracking. Performances
on step transition of reference acceleration amplitude and swept
sine frequency were studied and it was found that FLC performs
well for bare table and with rigid load. Further investigation was
conducted for resonant load and it was found that the controller
stability is sensitivity to the resonances to some extent. To over-
come this limitation further research is currently being
undertaken.
11346 K.P.S. Rana / Expert Systems with Applications 38 (2011) 11335–11346

References LDS Test and Measurement, Transfer function equalization, Technical note
TN007REV1, Middleton, 2005..
Lee, C. C. (1990a). Fuzzy logic in control systems: Fuzzy logic controller – Part I. IEEE
Åström, K. J., & Hägglund, T. (2001). The future of PID control. Control Engineering
Transactions on Systems, Man and Cybernetics, 20(2), 404–418.
Practice, 9, 1163–1175.
Lee, C. C. (1990b). Fuzzy logic in control systems: Fuzzy logic controller – Part II.
Baoliang, N., & Xia, Y. (2003). A FFT-based variety-sampling-rate sine-sweep
IEEE Transactions on Systems, Man and Cybernetics, 20(2), 419–435.
vibration controller. In Proceedings of international conference on neural networks
Li, H. X. (1997). A comparative design and tuning for conventional fuzzy control.
and signal processing, Nanjing, China (pp. 1714–1718).
Transaction on Systems, Man and Cybernetics, 27, 884–889.
Carvajal, J., Chen, G., & Ogmen, H. (2000). Fuzzy PID controller: Design, analysis,
Liaw, C. M., Yu, W. C., & Chen, T. H. (2002). Random vibration test control of
performance evaluation, and stability analysis. Information Sciences, 123,
inverter-fed electrodynamic shaker. IEEE Transaction on Industrial Electronics,
249–270.
49(3), 587–594.
Chen, G. (1996). Conventional and fuzzy PID controllers: An overview. International
Macdonald, H. M., Green, T. C., & Williams, B. W. (1993). Analysis and control of a
Journal of Intelligent and Control Systems, 1, 235–246.
moving coil electrodynamic actuator. In Proceedings of the international
Chen, T. H., & Liaw, C. M. (1999). Vibration acceleration control of an inverter-fed
conference on industrial electronics, control and instrumentation, Maui, HI, USA
electrodynamic shaker. IEEE/ASME Transaction on Mechatronics, 4(1), 60–70.
(pp. 2184–2189).
Darie, S. L., Colosi, T., Vadan, I., & Balan, H. (1994). Computational aspects of electro
McConnell, K. G. (1995). Vibration testing: Theory and practice (1st ed.). John Wiley &
dynamic vibrators. In IEEE electrotechnical conference, Antalya, Turkey (pp. 1073–
Sons.
1076).
Mohan, B. M., & Sinha, A. (2008). Analytical structure and stability analysis of a
De Silva, C. W. (2000). Vibration fundamental and practice (2nd ed.). CRC Press.
fuzzy PID controller. Applied Soft Computing, 8, 749–758.
Fair, R., & Bolton, H. R. (1993). Analysis and design of electromagnetic moving coil
Passino, K. (1993). Bridging the gap between conventional and intelligent control,
vibration generators. In IEEE sixth international conference on electric machine
Special Issue on Intelligent Control. IEEE Control Systems Magazine, 13, 12–18.
and drives (pp. 529–534).
Passino, K. (1996). Towards bridging the perceived gap between conventional and
Flora, L. D., & Gründling, H. A. (2006). Acceleration control of an inverter-fed
intelligent control. In M. M. Gupta & N. K. Sinha (Eds.), Intelligent control: Theory
electrodynamic shaker. In Proceedings of the 37th IEEE power electronics
and applications (pp. 1–27). IEEE Press.
specialists conference, Jeju, Korea (pp. 2799–2805).
Pivonka, P. (2002). Comparative analysis of fuzzy PI/PD/PID controller based on
Flora, L. D., & Gründling, H. A. (2008). Time domain sinusoidal acceleration
classical PID approach. In FUZZ-IEEE’02, fuzzy systems proceedings of the 2002
controller for an electrodynamic shaker. Institutions of Engineering and
IEEE international conference, 12–17 May (Vol. 1, pp. 541–546).
Technology, Control Theory and Applications, 2(12), 1044–1053.
Smith, F. S., & Shen, Q. (1998). Selecting inference and defuzzification techniques for
Fujita, M., & Uchiyama, Y. (2006). Robust acceleration and displacement control of
fuzzy logic control. In Proceedings of the UKACC international conference on
eletrodynamic shaker. In Proceedings of the international conference of control
control, 1–4 September (pp. 54–59).
application, Munich, Germany (pp. 746–751).
Tang, K. S., Man, K. F., Chen, G., & Kwong, S. (2001). An optimal fuzzy PID controller.
Heong, A. K., Chong, G., & Li, Y. (2005). PID control system analysis, design and
IEEE Transactions on Industrial Electronics, 48(4), 757–765.
technology. IEEE Transaction on Control System Technology, 13(4), 559–576.
Tustin, W. (2005). Random vibration and shock testing, measurement, analysis &
Keller, T. (2002). on the use of tracking filters during sine vibration testing. Sound
calibration (1st ed.). Equipment Reliability Institute.
and Vibration Magazine, 36(10), 82–85.
Uchiyama, Y., Mukai, M., & Fujita, M. (2009). Robust control of electrodynamic
Kuswadi, S. (2001). Review on intelligent control: Its historical perspective and
shaker by 2DOF control using H1 Filter. Journal of Sound and Vibration, 326(1–2),
future development. IECI Japan Series, 3(2), 38–46.
75–87.
Lang, G. F. (1997). Electrodynamic shaker fundamentals. Sound and Vibration
Zumberge, J., & Passino, K. M. (1998). A case study in intelligent vs. conventional
Magazine, 1–8.
control for a process control experiment. Control Engineering Practice, 6(9),
Lang, G. F., & Snyder, D. (2001). Understanding the physics of electrodynamic shaker
1055–1075.
performance. Sound and Vibration Magazine, 35(10), 24–33.