You are on page 1of 9

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3661629

Fuzzy logic control of automated screw fastening

Conference Paper  in  Robotics and Computer-Integrated Manufacturing · November 1995


DOI: 10.1109/IAS.1995.530506 · Source: IEEE Xplore

CITATIONS READS

34 262

3 authors, including:

Zhiqiang Gao
Cleveland State University
207 PUBLICATIONS   10,973 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ADRC: a new paradigm of the science of automatic control View project

Engineering Cybernetics: a renewal View project

All content following this page was uploaded by Zhiqiang Gao on 12 July 2019.

The user has requested enhancement of the downloaded file.


Robotics & Computer-Integrated Mterufacturing, Vol. 12, No. 3, pp. 235-242, 1996
Pergamon Copyright O 1996 Elsevier Science Ltd
Primed in Great Britain. All rights reserved
0736-5845]96 $15.00 + 0.00

PII:SO736--f,845(96)O0009-9

• Paper

FUZZY LOGIC CONTROL OF AUTOMATED SCREW FASTENING

NITIN DHAYAGUDE,* ZHIQIANG GAO* and FOUAD MRADt


*Electrical Engineering Department, Cleveland State University, Cleveland, OH 44115, U.S.A.
and tElectrical Engineering Department, American University of Beirut, PO Box 11-236,
Beirut, Lebanon

In this paper, we present an intelligent control strategy for automated screw fastening. In automated assembly
processes, there are often found dectlcated stations for various types/sizes of screw fastening. Problems found in
current processes include cross-threading, screw ]ammlng, slippage and the need to apply precise torque. The
intelligent controller developed here supervises the integrated process of an electric driver mounted on a robotic
positioning system to fasten screws. The new scheme controls confinuonsly the motion and driving stages to avoid
process-cansed failures, to achieve a desired precise torque and to detect bad parts at early stages of the assembly.
Initial simulation results are encouraging. Copyright © 1996 Elsevier Science Ltd.

1. INTRODUCTION 1.1. Screw-fastening process overview


Research and development in the area of robotics In today's flexible and efficient assembly processes,
and automation has been mostly concentrated in the we find many workstations dedicated for screw
past on the advancements of individual technologies, fastening. This present situation does not seem to
such as manipulator control, or sensing technolo- be changing in the near future, since the disadvan-
gies. However, for complex control problems, we tages of alternative fasteners in joining remain
often need to address the integration of technolo- obstacles facing design and production engineers.
gies, specifically torque feedback, position feedback The disadvantages of joining-alternatives (such as
from the tool encoder and the robotic positioning adhesives) include out-gassing, difficult reworking,
system. This is especially important in automated and difficult application when a precise clamp load is
screw-fastening problems owing to their unique needed.
requirements. Unlike standard closed-loop control In general, the screw-fastening process is required
problems, where the output of the system is to be to detect faulty joining conditions resulting from
driven to a desired setpoint, the screw-fastening cross-threading, jamming, slippage, partially taped
control is designed to maintain a particular torque holes, or stripped threads. More sensitive applica-
vs number-of-turns relationship in a process full of tions, such as clamping a stack of magnetic disks on
disturbances and uncertainties. Such problems have a motor hub in a hard disk drive (illustrated in Fig.
not been studied from the control-system-design 1), is'21 or fastening a transmission gasket 26, need a
point of view, and no systematic design approach precise clamp load produced by a set of fasteners. In
exists at this juncture. In this paper, a fuzzy-logic- this paper, we are addressing the concerns of the
based control system is proposed. Section 1.1 will sensitive assembly operations. In the process of
review the screw driving process, related concerns fastening a screw, there are three phases, illustrated
and available technologies. In Section 1.2, we in Fig. 2. The first phase is the engagement, where the
present a brief review of various intelligent control screw is contacting the part for at least one
schemes and discuss the selection of a fuzzy-logic- revolution. The second phase is the screw-down
control approach to the fastening problem. Section period, where initial torquing is applied and the
2 will present an overall description of our proposed screw head comes into contact with the part. The
control scheme with a block diagram of the process third phase is the torquing period, in which the clamp
steps. The fuzzy-logic controller (FLC) is designed load increases linearly and data are analyzed by the
and analyzed. Section 3 discusses the potential fastening system for monitoring and control strate-
implementation aspects of the proposed methodol- gies. ~8'26 These strategies are customized for the
ogy with a complete simulation model. The results specific application using 16-46 parameters. 26 Some
presented in Section 3.1 illustrate the technical of the factors that have an impact on the final
feasibility of our approach and highlight the claimed torquing process include: size of the screw, workpiece
advantages. Section 4 presents concluding remarks material, screwhead geometry, rotating speed of the
and possible future steps. driver, linear speed of the positioning system
235
236 Robotics & Computer-IntegratedManufacturing• Volume 12, Number3, 1996
Screw -~ screw-fastening technique that provides monitoring
and control during the entire duration of the process.
Our proposed scheme will achieve precise clamp load
in a joint and detect, as well as prevent, potential
faulty process cycles. The major components used in
this scheme are popular in today's applications.
Specifically, we use an electric driver mounted on a
servo slide or a robotic manipulator that is
Disk ~ l ~ " - ~ ~ Spacer computer-controlled. Electric drivers offer the ability
of closed-loop control, precise and fast control of the
stop phase, as well as efficient and simple main-
tenance. 26 Our intelligent control scheme does not
count on a mathematical model of the fastening
process with all the various factors that affect the
outcome of the assembly. Instead, it acts to adjust the
driver input torque (or current) in order to maintain
an acceptable torque vs angle behavior during the
torquing process. This scheme will prevent damaging
parts by process steps and tools throughout all three
phases of assembly. It also offers early detection and
Fig. 1. A disk pack assemblyin a hard disk drive. correction of bad joints or parts. Our proposed
scheme is most valuable when a precise clamp load is
required in a joint, especially when simultaneous
Torque screws are present in the same joint. 26'21

1.2. Intelligent control techniques


The successful application of intelligent control
techniques to a wide range of problems has been
demonstrated in research publications and commer-
cial products (see Refs 1-4, 7-9, 12, 14-17, 22-24,
Engag~ n e n t f . / " 27). The main avenues of intelligent control include
expert systems, fuzzy logic, neural networks, self-
Draw-In
learning controllers and hierarchical intelligent
Time/angle controllers. The intelligent control mechanism re-
places an analytic controller in a conventional
Fig. Z. Screw-fasteningprocessstages. control system with a high-level decision-making
scheme that generates the control signal based on a
manipulating the driver, and the parts-coating qualitative or heuristic understanding of the process.
lubricant characteristics. 1s'25 Typically, the intelligent controller output is based
Ideal automated fastening systems are capable of on known trends or rules-of-thumb of how the
monitoring and controlling the fastening process to process reacts in a dynamic environment. It is
produce a consistent, precise clamp load. Existing capable of capturing the expert knowledge and
technologies can be classified into two families. The heuristics in various forms and incorporates this as
first group is characterized by the monitoring of a part of the control mechanism. To find an
torque vs time or angle at the end of the fastening intelligent control approach for an automated
process. 6A°'26 In this scheme, angular encoders and screw-fastening process, the following main meth-
torque transducers are required to verify the success odologies are evaluated.
of the joining operation. The second group consists
of torque vs angle control strategy. In this control 1.3. Expert systems
strategy, a reference torque value indicates the An expert system uses a knowledge base of
starting of the torquing phase. After a predetermined information consisting of rules, and an inference
number of driver revolutions, the expected torque engine that usually searches the rule base for the best
should be achieved. Otherwise, a faulty joint is solution. Expert systems for control are available in
detected. some commercial products as well as in a large
In the sensitive assembly applications mentioned volume of research literature (see, for example, Refs
earlier, 1s-21 these techniques are preprogrammed, 16, 17, 22, 24). In our control problem, an expert
rigid and problem-dependent. In addition, they can system approach would grow significantly in com-
detect a faulty joint only at the end of the process, plexity with various types of screws, processes and
but they cannot prevent or correct a potential faulty specifications. It is quite possible to develop an expert
joint. This motivated us to develop an intelligent system to handle a particular screw-fastening
Fuzzy logic control of automated screw fastening • N. D H A Y A G U D E et al. 237
application. But this expert system must be modified execute this qualitative idea in a real-time situation, is
manually by a human operator to accommodate each an attractive feature in the control of screw-fastening
change in application. Such an expert system processes. In summary, the fuzzy-logic design
approach would be too rigid and problem-dependent approach is selected for the screw-fastening problem
as a general methodology for industrial applications. for its ability to characterize relations and depen-
dences that are not exactly known. The simple
1.4. Artificial neural networks structure and low computational complexity make
Artificial neural network (ANN) models have been this approach clearly preferable to the two intelli-
studied for many years in hopes of duplicating the gent-control methodologies above.
complicated behavior and capacity for learning
demonstrated in the human brain. They consist of
many non-linear computational elements operating 2. INTEGRATED CLOSED-LOOP-FASTENING
in parallel, and typically are modeled by sigmoid PROCESS
functions. It has been demonstrated that ANNs have The overall fastening process can be illustrated as in
good potential in areas such as non-linear systems Fig. 3.
and self-learning control, where conventional control The need to monitor precisely the clamp load
methods become overly complex to represent (torque or force) achieved by a given joint
adequately the characteristics of the system due to necessitates torque signal feedback. In d.c. electric
inherent non-linearity, time delays, saturation, time- drivers, the applied torque is proportional to the
varying parameters, and model uncertainty. Some input current of the driver: torque =constant × (/),
investigations along these lines have been done in where (/) is the armature current in an armature-
Refs 7, 8, 23, 27, primarily using the back- controlled motor, or the field current in a field-
propagation training technique. controlled motor. Therefore, torque can be computed
The major effort of an A N N solution involves the via a current measurement. However, in small torque
selection of a control system configuration, a applications the reactive torque applied by the
network configuration, A N N training algorithm, workpiece against the driver shaft cannot be ignored.
and training data set. An A N N is a viable approach Thus the introduction of a torque (or force)
to screw fastening, but it provides little under- transducer to measure the actual clamp forces,
standing of the nuances in the control problem. typically a network of strain gages connected by a
The main disadvantage in using an A N N for the wheatstone bridge. 6'26 The torque sensor is installed
fastening problem is the lack of flexibility of the at the end of the driver shaft contacting the bit. These
controller. The A N N controller can be trained for a transducers are available in different sizes, depending
specific fastening application. However, when the on the torque range of interest. 6 The driver is
screw sizes and processes change and the adjustments equipped with an encoder, which supplies the angular
must be made on-line, the requirement for timely position of the driver shaft and the corresponding
changes in the controller may well exceed its angular speed. The positioning system manipulating
limitations. the driver can be any computer-controlled position-
ing system that can move the end-effector tool in a
1.5. Fuzzy logic straight line trajectory. The various levels of
Fuzzy logic uses the concept of degrees of member- interactions among the positioning system speed,
ship rather than strictly true or false to characterize driver rotational velocity and weight, the screw
variables of interest. It is based on fuzzy-set theory geometry/diameter and finishing, the workpiece
that describes a set using a smooth transition rather material and other factors make the model of the
than a sharp boundary between membership and integrated process very uncertain. A FLC is designed
non-membership. 1.:4 The degree of membership of an to handle this high degree of uncertainty and
object in a set is subjective in nature, a matter of
definition and not necessarily measurement. * Applied torque
* Driver (rpm) angle
Fuzzy control is an appealing alternative to * Robot position
conventional control methods when systems follow
some general operating characteristics and a detailed
process understanding is unknown, or traditional Control /~
system models become overly complex. The cap-
ability to capture qualitatively the attributes of a
control system based on observable phenomena is a CONTROLLER
main feature of fuzzy control. These aspects of fuzzy @ Driver Scheme
control have been demonstrated in the research
@ X-Y-Z Table
literature (see Refs 1, 2, 4, 5, 9, 12, 14), and in
commercial products from control vendors like Conveyor Control Strategy
Reliance Electric and Omron. The ability of fuzzy
logic to capture system dynamics qualitatively, and Fig. 3. Fastening workccll layout.
238 Robotics & Computer-Integrated Manufacturing • Volume 12, Number3, 1996
Positioning
System
Desired Input, Fz
Torque
& Angle
Fuzzy ~ / Toroue
e(k)
Logic
Controller
-s l

Torque& angle
Fig. 4. Control system block diagram.

Measured A typical FLC is thus composed of three basic


joint parts: input signal fuzzifieation, a fuzzy engine and
High '~ ¢~:
Specifications ~ ~. ':i: defuzzification that generates the input to the
Low system as a control action. Figure 6 shows a typical
FLC. The fuzzifieation block transforms the con-
tinuous input signal into linguistic fuzzy variables
such as small, medium and large. The fuzzy engine
_ _ ~ Slippage carries out rule inference where the expert knowledge
Acceptable Driver of the process to be controlled can be injected
torque Low High angle through linguistic rules. The defuzzification block
measurement
Start Specifications converts the inferred control action to a continuous
encoder
linear region signal.
Fig. 5. Fuzz-logic control specifications.

complexity in the coupled system, I1 keeping in mind


that the system input (driver input torque or current, 2.2. Structure of the fuzzy-logic controller
the manipulator input torque or force) and the The FLC developed here is a two-input single-output
output (the achieved joint torque or force) are not controller. The two inputs are the deviation from
fuzzy signals. Figure 4 is a block diagram of the setpoint, error e(k) and error rate Ae(k). The FLC is
proposed control system showing the major utilized implemented in a discrete form using a zero-order-
signals. Note that the. purpose of the integrated hold as shown in Fig. 4, with membership functions
approach is to maintain an acceptable (between the defined in Fig. 7. The membership functions are
maximum and minimum technical specifications of divided into eight regions, linguistically labeled as:
the product) level of joint torque at the correspond- positive large, medium, small and zero; negative
ing angular position (or depth) of the screw. Figure 5 large, medium, small and zero, i.e. {PL, PM, PS, P0,
illustrates the purpose of the FLC strategy. NL, NM, NS, NO}.
The FLC is developed based on the under-
standing of how the process works. A fuzzy-logic
2.1. Development of fuzzy-logic controller
control law was set up in discrete-time form, which
In the process of developing a FLC to control a
generates a control signal similar to the conven-
dynamic system, a number of key items need to be
tional PI-type controller. The controller output u(k)
specified. A universe of discourse must be defined for
is given by
each variable and associated with linguistic labels.
Membership functions must be defined for each u(k) = u(k - 1) + F(e, Ae)
linguistic label. Fuzzification and defuzzification
mapping procedures must be specified to relate each with u(0) = 1 and F(e, Ae) as the output of the FLC
linguistic label to its universe of discourse. Rules must for the kth sampling interval. The output member-
be developed to capture the control strategy qualita- ship for F(e, Ae) shown in Fig. 7 is developed using
tively relating the input to FLC output. an intuitive generalization of the PI controller. 13

Controller Control
Action
e Fuzzification Fuzzy Rules Defuzzification

Ae
Fig. 6. A schematic diagram for a typical fuzzy-logic controller.
Fuzzy logic control of automated screw fastening • N. D H A Y A G U D E et aL 239

Input Membership of straight lines for each linguistic label, starting at


NL NM NS / I PS PM PL the origin and rising to the maximum grade of
membership, i.e. a value of 1 at the endpoint. The
only exceptions are the rules with output P0 and NO,
which have a value of 1 at the origin and have the
Input Value e,~e minimum membership grade of 0 at the endpoint.
Output Membership
2.2.2. Fuzzy-logic controller tuning. Since there is no
NL NM NS I1 as PM PL established theory for developing rules and
membership functions for FLCs at this time, tuning
I -r-.~',,~ I /~./j--r J i is a heuristic process. The rules described above were
developed from expert knowledge of such processes.
Output Value u The membership partitions for e and Ae are
defined symmetrically over the universe of discourse,
Fig. 7. Fuzzy membership functions.
a region from - 1 to 1 (to permit scaling), using
positive and negative linguistic labels such as zero,
Input e Input Ae Output u small, medium and large. In general, the number of
Membership Membership Membership
Positive Small Positive Zero Positive Small membership functions can vary to provide the
resolution needed, but resolution is directly related
to the control rule generation effort, as can be seen
from Scheme 1. The eight membership functions for e
and Ae generate 64 possible partitions. The tuning
(If) e is PS e (And) he is PO &e (Then) u is PS u effort resulted in 12 regions that correspond to the
rules previously described and effectively control the
Fig. 8. Fuzzy-control rules. system.
The triangular membership functions for e and Ae
2.2.1. Rule and membership development. F(e, Ae) had their initial membership tuning center values at
includes 12 rules, shown graphically in Fig. 8 in the 0, 0.2, 0.35 and 0.6, corresponding to zero, small,
form: medium and large, respectively. The left and right
membership functions for each linguistic label were
IF " e / s label 1" A N D "Ae is label 2" arbitrarily chosen to provide membership overlap
THEN "'output is label 3". with adjacent membership functions. The member-
ship overlap provides a transition region in the FLC
The output from each rule can be treated as a
as the plant moves between operating states which
fuzzy singleton. The FLC control action is the
have different rules to control the plant. The straight
combination of the output of each rule using the
line output membership functions, stated as a
weighted average defuzzification method and can be
percentage of the maximum incremental step size,
viewed as the center of gravity of the fuzzy set of
had initial end points specified as 10, 30, 70 and
output singletons. The rule set is listed in Scheme 1;
100%, corresponding to the labels zero, small,
shaded areas indicate zero control action.
medium and large, respectively (see Fig. 7). Both
The rules and membership functions are developed
the input and output membership functions are
using an intuitive understanding of the screw-
symmetric with respect to the origin.
fastening process. A PI control strategy is captured
Tuning of the FLC was accomplished by adjusting
linguistically and implemented in the FLC as the rule
the membership functions based on the fuzzy output
set shown in Scheme 1. The rule output membership
set used to calculate the control output. Using this
functions, as shown in Fig. 6, are defined by a series
tuning approach, the existence of a dominant control
rule is indicated by a single membership function
,de
with a high membership grade. Conflicting rules are
I us I 01,01 ,s I PMI ,L indicated by two or more membership functions with
NL NL a high membership grade.
NM NM PS By evaluating the fuzzy output set for each control
action, weak and conflicting control rules can be
:::::::::::::::::::::::::::::::::::::: I :::::::::::::::::::::::::::::::::::::::::::
modified by adjusting the e and/or Ae membership
NO NO
functions of the linguistic labels used in the rule
e P0 P0 premise. Modifying the output membership function
~?~.~.~t~'~V:'~.~'.;.'%~
PS ~ NS adjusts the rules contribution relative to the output
PM I NS PM universe of discourse. Once the dominant rule tuning
PL PL
is completed, fine tuning of the membership
functions is performed to achieve the desired control
Scheme 1. FLC control rules performance.
240 Robotics & Computer-Integrated Manufacturing • Volume 12, Number 3, 1996

This rule optimization scheme is accomplished 0 ~


through trial and error. The tuning and rule
development relies on an intuitive understanding of -2
how the closed-loop system operates. The method
provides tuning based on a dominant rule. If the rule
base does not have sufficient knowledge or variable z -4
resolution to provide adequate performance, more N
It.
knowledge or rules may be required along with a -6
returning effort.
-8
3. IMPLEMENTATION FEASIBILITY
The proposed control scheme of this paper would
have to be verified and tested in an experimental -10
0 0.1 0.2
setup. The practical application in an assembly line time in sec
has to be customized for the particular product and
parts as well as for the technical specifications of the Fig. 9. Manipulator input force, with pulse disturbance.
production line. The sophistication and cost of the
proposed technique have to be justified. We believe in Fig. 2. This control scheme adjusts the driver input
that the applications requiring precise clamp load torque in order to maintain the desired torque/angle
from a set of screws need the advanced scheme of this relation during the entire fastening process. This
paper. At this stage, we have some positive feasibility FLC was developed using MATLAB and implemen-
results based on a simulation model of the process, ted on a VAX-750 system. The simulations were
keeping in mind that the uncertainty and incomplete- carried out with noisy transducer measurements and
ness of the integrated model required the fuzzy pulse disturbances in the manipulator input force.
control in the first place. Therefore, we acknowledge The disturbances are injected to mimic the commonly
that our simulation model is incomplete and seen motion jerks and jamming in such processes.
imprecise. However, it does verify some of the The sensor noise was added to simulate closely the
coupling effects of which we are certain, and it actual screw-fastening process. The noise was
proves some of the advantages of our proposed assumed to be normally distributed.
scheme. We do not attempt to model the fastening From the simulation results, we note that the FLC
process complexity, neither are we capable of doing was able to handle a rather large and impulsive
that owing to the large number of variables and disturbance in F= as illustrated in Figs 9, 10 and 11.
unmodelled couplings in the integrated process. We In particular, Fig. 9 shows the manipulator input
present next the approximated model of the fastening force with a significant pulse disturbance at 0.1 see.
process and the simulation results of applying the Without proper adjustment to the input torque "Cin,
proposed FLC scheme on the computer model. such a disturbance would directly affect the joint
torque Zjoint. It is shown in Figs 10 and 11 that the
3.1. Simulation FLC successfully modified the input torque and kept
The model adopted for the simulation is derived the joint torque in the proper range. In Fig. 12, the
using the Euler-Lagrange equation of motion of the viscous friction constant changes gradually by 100%
system presented in Fig. 3 earlier. The following from t=0.075 sec to t=0.125 sec, simulating the
dynamic equations describe approximately the variations in different parts and operating conditions.
motion of the screw-fastening process. The fuzzy controller accommodated such a change

m~ =Fz + mg - [kz~ + ko 0] 2.5


J0 =~in - Boa - Bz~, = ~joint
where Zjoint is the achieved actual joint torque (load)
(Nm), zi, is the input torque (Nm), m is the mass of
the driver mounted in the end effector (kg), F= is the
input force of the linear servo slide (N), g is the
gravitational acceleration (m/see2), k= is the constant
zE 1.51
.=_

o
1,

0.5'
Ij
~5
r,i
of friction, ~ is the linear velocity of servo slide (In/ "5
see), ko is the constant of friction, 0 is the rotational 0 0
velocity of the driver (rad/see), J is the total inertia
-0.5
(kg m2), Bo is the viscous friction constant, B~ is the
viscous friction constant. -1
The FLC system presented in Section 2.1 is applied 0 10 2o
"fheta in tad
to the process described above. The goal is to achieve
a desired torque and angle relationship as described Fig. 10. Joint torque vs driver angular position.
Fuzzy logic control of automated screw fastening • N. D H A Y A G U D E et aL 241

the achieved joint torque at the specific driver angle is


3.5 novel and, to the best of our knowledge, has not been
performed. We believe that the complexity and c o s t
3 of implementing the proposed scheme are justified
E
z 2.5 only in certain fastening steps and products, for
.m
instance, when a precise clamp load is needed from
2 one or more screw joints. The proposed scheme is
.9,o portable to other fastening applications, such as bolt
"5 1.5
0... tightening. The feasibility study was carried out
1 based on an approximate model simulation. The
promising results wiU be experimentally tested, where
0.5
the actual integrated fastening process does not need
to be modeled or approximated. Our study verified
0 011 0.2
time in sec the usefulness of the F L C in adjusting the driver
motion to guarantee the acceptable joint torque.
Fig. 11. FLC output, driver input torque• Furthermore, the robustness of the proposed scheme
was demonstrated in the presence of noisy measure-
2.5 4.5 ments and jerky manipulator motion. The same
4 control scheme is thought to be capable of detection
of early error, such as jamming or slippage
3.5
E 1.5 situations.
z
.=_ c

"~ 2.5 Acknlowedgements---This work was partially supported by


E
"q 0.5 g2 an American University of Beirut Research grant. The need
to solve this particular problem in the fastening process was
~1.5 identified in the IBM hard disk drive assembly plant of
o 0
1 Rochester, MN. The feasibility study was done in the
-0.5 0.5 Electrical Engineering Department of Cleveland State
University, OH.
-1 0
0 10 20 0 0.1 0.2
Theta in rad Ume in sec
REFERENCES
1. Abdelnour, G., Chang, C., Huang, F., Cheung, J.:
Fig. 12. Output and input torque. Design of a fuzzy controller using input and output
mapping factors. 1EEE Trans. Systems Man Cyber-
successfully by increasing the input torque. In Fig. netics 21: 952-960, 1991.
13, the system experiences a drastic jump in viscous 2. Bernard, J.: Use of a rule-based system for process
friction constant at t=0.1 sec, and appears to be control. IEEE Control Systems Magazine 8(5): 3-13,
1988.
headed to a jam. The fuzzy controller was able to 3. Chen, F.: Back-propagation neural networks for
detect this problem, and it increased the torque nonlinear self-tuning adaptive control. IEEE Control
accordingly to maintain a desired torque vs number Systems Magazine 10(3): 44-48, 1990.
of turns relationship. 4. Chiu, S., Chand, S., Moore, D., Chaudhary, A.: Fuzzy
logic for control of roll and moment for a flexible wing
aircraft• 1EEE Control System Magazine 11(4): 42-48,
4. C O N C L U S I O N S 1991.
5. Dawson, J., Gao, Z.: Fuzzy logic control of linear
In this paper we presented a feasibility study of systems with variable time delay. In Proceedings of the
implementing a F L C in the screw-fastening process. 9th 1EEE International Symposium on Intelligent
Such continuous control of the driver speed based on Control, August 1994.
6. GSE data-stat: Torque-angle monitor; force displace-
4 ment, Model 560. Farmington Hills, MI, GSE Inc.
1993.
3.5 7. Gu, Y.: On nonlinear system invertability and learning
3 approaches by neural networks. In Proceedings of the
American Control Conference, San Diego. 1990, pp.
z=E2.5 3013-3018.

2/
"~ 1 8. Guez, A., Selinski, J.: A neuromorphic controller with
a human teacher. In Proceedings of the International
osh p Joint Conference on Neural Networks, San Diego. 1990,
~ 1.5 pp. 11595-11602•
1 9. Huang, L., Tomizuka, M.: A self-paced fuzzy
-0.5 0.5 tracking controller for two-dimensional motion
control. IEEE Trans. Systems Man Cybernetics 20:
-1 0 1115-1124, 1990.
0 10 20 ) 0'.1 0.2
10. Jergens Portable Tools: Controlled Torque Screwdrivers
Theta in rad time in see
for Assembly. Cleveland, OH, Jergens Inc. 1988.
Fig. 13. Output and input torque. 11. Kim, M., Kim, K.: Feasibility study of fuzzy control
242 Robotics & Computer-IntegratedManufacturing• Volume 12, Number 3, 1996
systems in machining processes. Robotics Computer- 20. Mrad, F., Ahmad, S.: Adaptive control of flexible joint
Integr. Mfg 9: 211-217, 1992. robots using position and velocity feedback. Int. J.
12. King, P., Mamdani, E.: The application of fuzzy Control 55(5): 1255-1278, 1992.
control systems to industrial processes. Automatica 11: 21. Mrad, F., Reid, E., Malek, S.: Magnetic heads loading
235-242, 1977. in disk drive assembly using rectangular robot with
13. Kuo, B.: Automatic Control Systems, 6th edition. computer visual feedback. In Proceedings of the IEEE
Englewood Cliffs, NJ, Prentice-Hall. 1991. Industrial Electronics Society 19th Annual Conference,
14. Lee, C.: Fuzzy logic in control systems: fuzzy logic Maui, Hawaii. November 1993.
controller, Parts 1 and 2. IEEE Trans. Systems Man 22. Neat, G., Kaufman, H., Roy, R.: Expert adaptive
Cybernetics 20: 404--435, 1990. control for drug delivery systems. IEEE Control
15. Lee, W., 81otine, J.: Neural network control of unknown Systems Magazine 9(4): 20-24, 1989.
nonlinear systems. In Proceedings of the American 23. Sannar, R., Akin, D.: Neuromorphic pitch attitude
Control Conference, Pittsburgh. 1989, pp. 1136--1141. regulation of an underwater telerobot. IEEE Control
16. Lingarkar, R., Lui, Li, Elbestawi, M., Sinha, N.: Systems Magazine 10(3): 62-67, 1990.
Knowledge-based adaptive computer control in man- 24. Shin, K., Cui, X.: Design of a knowledge-based
ufactudng systems: a case study. IEEE Trans. Systems controller for intelligent control systems. IEEE Trans.
Man Cybernetics 20: 606--616, 1990. Systems Man Cybernetics 21: 368-375, 1991.
17. Litt, J.: An expert system to perform on-line controller 25. Shoham, M., Srivatsan, R.: Automation of surface
tuning. IEEE Control Systems Magazine 11(3): 18-23, finishing processes. Robotics Computer-Integrated Mfg
1991. 9: 219-226, 1992.
18. Malek, S., Chiang, Y., Mason, J.: Multivariable effects 26. Smith, B.: Automated systems for accurate fastening.
on an automatic screw-torquing process. J. Mfg Machine Design 61(17): 44-50, 1989.
Systems 12(6): 457--462, 1992. 27. Wu, Q., Hogg, B., Irwin, G.: A neural network
t9. Mrad, F., Ahmad, S.: Control of flexible joint robots. regulator for turbogenerators. IEEE Trans. Neural
Robotics Computer-Integr. Mfg 9(2): 137-144, 1992. Networks 3: 95-100, 1992.

View publication stats

You might also like