SOME COMMUNICATION PROBLEMS OF KB CONTROLLED
MANUFACTURING SYSTEMS
George L. Kovács and János Nacsa
CIM Research Laboratory, Computer and Automation Research Institute of HAS,
1518 Budapest POB 63, Hungary
ABSTRACT
According to our recent view knowledge based or
intelligent control of Flexible Manufacturing
Systems (FMS) means a continuous or frequent
observation and evaluation of the status and
condition of the system performance, decision
making based on the evaluation results and on pre-
defined knowledge, and then the operation
according to the decisions. In the case of normal
scheduled operation, there is no need for
intelligence, neither for interactions to modify the
operation. The above given procedure may help
(Smith, et al.,1992) if any kind of disturbance or
irregularity happens, what is rather common in the
case of highly sophisticated, complex systems. Our
paper deals with control and communication
problems of flexible manufacturing systems using
programs based on a real-time expert system
environment taking into consideration the
requirements of openness of up-to-date systems.
Two robot applications will be discussed as
demonstrations.
1. INTRODUCTION
Recent discrete manufacturing and/or assembly
systems (FMS/FMA) are more and more often
using MAP/MMS (Manufacturing Automation
Protocol, Manufacturing Message Specification)
(Brill and Gramm, 1991), because this technology
is widely available from many vendors and gives a
real safe and open solution according to the
demands of OSI (Open System Interconnection).
Many users do not exactly know that they have
such interconnections, they just enjoy the useful
features of MAP. We must also admit that the
MAP or related solutions are still rather expensive.
On the other hand the so called intelligent control
is getting to be a general demand. There is a vivid
discussion in literature and in private
communications among control engineers about
the existence and need of intelligent control
(Franklin et al.,1994). Some experts claim that
there is nothing that really could be called
intelligent control. Most of them are speaking only
about process control, and not about discrete
manufacturing or robot control, however the
control tasks and problems of manufacturing
systems are basically similar to those of batch-like
process control. Now, without going into
discussions in this issue we accept the need and
necessity of intelligent control which we intend to
solve by means of knowledge based (KB) systems,
which we often call expert systems. We admit that
some commercial expert systems provide good
problem description and software development
tools where the programming is closer to the
problem to be solved and to the user and limited
real-time facilities are supported, too (Laffey et al.
1988).
The next chapter shortly introduces our basic
system the developments started from. The third
chapter will discuss the communication problems
of expert systems in FMS applications and give a
problem decomposition. Later on there is a brief
description of the features of the MAP based
Virtual Manufacturing Device model in intelligent
control. Finally there is a presentation of the
detailed basic definitions, and the way we went on
to end up with something what we call "intelligent
robot cell controller with open communication".
2. A KNOWLEDGE BASED HYBRID
SIMULATION-SCHEDULER SYSTEM
We had developed a knowledge based simulation-
scheduler system called SimSched-Q (Kovács et
al., 1993). In the SimSched-Q system the
simulation was implemented in a world-wide
accepted, but 'traditional' simulation/animation
system called SIMAN/Cinema. It was surrounded
by expert system modules (preparation, on-line
quality and scheduler advisor and evaluation)
implemented in the same G2 Ver. 3.0 object-
oriented real-time intelligent environment. The
study in this paper only refers to the SimSched-Q
system as resulting scheduling data are the basic
input for (intelligent) FMS control. Similar efforts
are also being done (e.g. Somlo and El Buzidi,
1996).
There are four expert systems: two of them are
shallow coupled with the simulation (Preparation
and Evaluation), two of them are deep-coupled
(Quality and Scheduler advisors). The structure and
the functionality of the system were designed and
developed in a way that later the simulation might
be changed to a real FMS environment, and the ES
would be the real-time controller.
The prototype application was developed with the
real data (layout, capacities, process plans, machine
parameters, etc.) of some existing FMS, as for
example the Pilot FMS of the Technical University
of Budapest. This system contains four ‘holon-like’
cells (storage with AGV, measurement, assembly
and metal-cutting consisting of a CNC lathe and a
CNC machining center served by robots) on a
network.
To speak about real-time control we need all
scheduling data, as the starting and finishing date
of all operations of all equipment (machine-tools,
robots, transfer units, as AGV, etc.) together with
the possibilities of downloading CNC control
programs to all equipment, and to accept and
evaluate different signals from them, etc. The time
of the loading/unloading should be calculated, too.
3. COMMUNICATION WITH EXPERT
SYSTEMS
The practical problems of the communication of
expert systems in CIM (Computer Integrated
Manufacturing) applications can be divided into
two parts. One is the hardware-software connection
(physical) and the other one is the logical
connection between the controller(s) and controlled
devices. This decomposition was very useful both
in the design and implementation phase during the
last projects of our CIM Laboratory. If this
decomposition is not so sharp many problems may
occur during the development and specially in
maintenance later on.
There are relatively easy programming interfaces
(etc. C/C++) in most available ES shells. These
interfaces provide data transfer and communication
possibilities with external tasks, stations, etc. They
support clear and easy programming to reach
objects, to call procedures, to set and get variables,
etc. The interfaces are dedicated to specific
software tools of the ES and they are general
towards the external world without being able to
take into account the requirements of the given
application. So nearly each CIM implementation
requires special software development to cover this
gap between the external world and the ES.
Clearly the communication functions depend on
the capabilities of the expert system. The way of
learning and knowledge handling determine the
logical levels of the communication. Three
different types of working mode and different
levels of the communication of an intelligent cell-
controller in a CIM environment are defined.
(Nacsa and Kovács, 1994). These levels are
implemented - of course - within the same
protocol.
The lowest level (Data Acquisition and Control
Level) has the basic control and data acquisition
type messages. The other two levels (Knowledge
Acquisition and Knowledge Communication
Levels) have messages if and only if the
‘intelligence of the cell controller is not hidden’.
Hidden intelligence means in this term that the
knowledge based technology is applied only inside
the cell controller and it has no specific actions via
the communication channel. A typical example of
the hidden case if a KB system is built up on the
top of a traditional control system using its original
communication. On the highest logical level the
knowledge acquisition and the so-called knowledge
communications were separated. The first one
contains specific data for modifying or verifying
the knowledge of the given controller. When a KB
system shares its knowledge (new or modified) it
belongs to the knowledge communication level
(Buta and Springer, 1992). The communication
messages of most real and pilot KB applications
for FMS control belong to the lowest, possibly to
the middle logical level.
4. MMS - AN OBJECT ORIENTED
COMMUNICATION TOOL
Going back to the so-called physical connection
there are many possibilities to follow. Most
controller and controlled device vendors offer
good (proprietary) solutions to communicate and
also vendor independent standards are available.
Arcwelding Robot VMD R_ARM
(KUKA-ESS) domain
R_SF
Vendor, Model, Revision
Physical & Logical Status
R_MP49
R_MP50
R_GO
prog. inv. R_MP51
domain
R_WELDST
R_STATE
R_ERRCODE R_OPERID
R_VLOCAL R_AUTO
Fig. 1. Virtual Manufacturing Device model of a
robot (an example)
Also in the CIM area there are more accepted
models or modeling tools to describe the objects of
an FMS. In the communication point of view the
most promising one is the object oriented view of
the so-called MMS (Manufacturing Message
Specification), which is originally an application
layer protocol in the MAP OSI networks. It was
realized that this specification is good on the higher
level of the FMS (Nagy and Haidegger, 1994) to
give a communication oriented view about the
network elements and their resources.
MMS gives a so-called VMD (Virtual
Manufacturing Device) view about each resource
of the FMS. All the objects of the MMS (e.g.
VMD, domains, different types of variables,
program invocations) are object oriented with
special sets of attributes and actions (called
services in MMS) belonging to a given object type.
Fig. 1. shows a simple, but rather general type of
VMD :an arc-welding robot VMD. In a VMD it is
possible to create, read and write different types of
variables, to up- and download domains which can
be programs or machine data, to start and stop
different tasks (with program invocations) and to
handle events. The object-oriented view of MMS
allows to design the VMD model of a certain
device, and it is immediately possible to use this
model as a specification of the communication
where the services (what one can do with a given
object of the VMD) are defined and they are
working in the MMS networks.
5. AN EXPERIMENTAL ROBOT CONTROL
SYSTEM
As the goal of this work is to provide real-time,
intelligent control of FMS systems the next step
was to apply all our development results and to
integrate the independently working and tested
software-hardware modules.
As a first experimental set-up (Fig. 2.) a simple
configuration was chosen. From earlier projects a
Mitsubishi robot was available with serial DNC
connection to a PC that had a MAP interface. The
trivial TIC-TAC-TOE game was served by the
robot according to the following manner: a user
could play against the G2 system. Of course it is
not a problem for that such an intelligent system is
needed. But this well-known game is able to
clearly demonstrate the 'intelligent' features of the
control. On the other hand the full capability of
MMS networking was used.
The structure of the setup shows that a special
gateway was developed between the G2 base
controller and the robot. It couples the G2 system
interface (called GSI) to the MMS standard
communication. In this case it was the 'gap
software' mentioned in the chapter 3. This special
gateway of the G2 allows to develop any discrete
manufacturing applications to develop in G2 using
MMS standard interface without having to modify
the interface. In the demonstration version the
gateway supports the context management and
some program invocation services of MMS.
For this solution a special object hierarchy called
MLGO (MMS like G2 objects) and a procedure set
were needed within the G2 to make the knowledge
base to handle the MMS interface. Three solutions
were examined before developing our MLGO.
 It was examined how the complete MMS
object structure could be built up within the
G2. It took a lot of efforts and gave a relatively
complicated and not too useful structure.
Because of the different services and their
parameters many external procedure calls were
defined with big parameter sets. All
implemented MMS services have their own
GSI procedures and all MMS object types have
their internal G2 object descriptions. The
complexity of the inner structure was the same
as the MMS itself. It did not give a clear and
easy-to-use tool for the G2 user.
 INTELLIGENT GATEWAY
GSI GSI
CONTROLLER
obj. & C
(G2
ENVIRONMENT) proce- inter-
dures face
MMS like G2 objects MMS client

CONTROL CONTROL
RESPONSE REQUEST

MMS server of robot (PNIU)

Fig. 2. Knowledge Based control of a robot with OSI network

 It would have been also possible to define A simple example of the method can be seen in the
independent procedures to all cases when an Fig. 3. When the user wants to take a box to a
MMS services should be used. In this case the given place he or she should push the button of the
whole MMS functionality would be mouse above the place. A G2 'user-defined-action'
implemented within the gateway. It is the belongs to the place which starts the next-move
solution without any MLGO support. procedure. It will pick up the parameters of a start-
 The third solution reduced the number of the program-invocation MMS service from the
necessary objects and procedures to some MLGO and call the GSI mms-serv external
general terms. It uses limited number of procedure that sends the request to the robot across
external GSI procedure calls also, but give a the gateway. This method is rather general and
rather general (few limitations) interface other MMS Application Enabler programs at the
towards MMS. market are using similar solutions (EasyMAP,
1993).
MLGO
PLACE3, a place
Operation "StartPI"
Color gray Param 3
Weight 30
a place Position 3 Last value -1 6. KNOWLEDGE BASED CONTROL OF AN
Remote appl "pc.mms"
table
step Mms object type prog-inv
ARC-WELDING ROBOT
Mms object name "R-GET"
start next-move Domain ""
(user,PLACE3) Named type "" In the last two years a CAD/CAM integrated robot
programming system (PROARC) was developed in
co-operation with German and Hungarian
participants in the frame of an EC East-West
project (Weck et al, 1995). The goal of the system
call mms-serv(operation, is to assist arc-welding with robots in SMEs mostly
remote-appl,object,domain, for small and medium lot sizes with a PC based,
type, param) across mygsi-1 interactive, off-line robot programming system.
Fig. 3. Realization of MMS within the G2 system Most commercially available systems are either too
expensive, or cannot provide enough features for
arc-welding. CAD/CAM integration means that
after adding the necessary technological
information to the AutoCAD based geometric
design the appropriate programs provide automatic
program generation and remote control for the
applied robot to execute the welding tasks.
The PROARC system can be divided into different
modules: Modelling, Welding Planning, Program
Generation and Communication. In this functional
division of the system the different phases of an
offline session are mapped. However, the
underlying structure is not that separable,
PROARC system
AutoCAD R13 ADS function calls
AutoCAD requests
Modelling and welding
planning module
STEP file
RCM3 file
Program generation module
Communication module
MMS network
Fig. 4 Structure of the dual system: PROARC and G2 based KB environment
Fig. 4. shows the connections of the KB based
control and the PROARC system where the
original communication module of PROARC has
been replaced with the KB control module. This
modification will lead us to integrate more parts
into the KB and then the real features and power of
the integrated KB tool will appear.
There are several possibilities to increase the
‘intelligence’ of the recent PROARC system:
 The similarities of the workpiece families
could be managed and analyzed in a more
sophisticated way even if this feature is a
special one in the PROARC. The intelligent
comparison of workpieces in the design phase
especially not for the first two modules. In the
Modelling Module basically the welding-cell and
its components are inputed into the system. The
Welding Planning Module provides the capabilities
to associate seams to the modelled workpieces and
to generate all movements during the welding
session. The Program Generation generates the
robot specific code including some system and
environment specific features as well. The
Communication Module provides the connection
to the robot controller.
G2
environment
Welding
features
management
Workpiece
families
Knowledge based
control system
with MMS interface
can assist of re-using welding settings or even
pieces of robot program parts.
 The present system supports already the
unskilled user with offering default values of
the different welding variables, but a
knowledge based tool could grow the capacity
of this feature.
 Recently the system uses only a simple tactile
sensor to provide safe welding. The intelligent
evaluation of sensor-information could help in
making the system more reliable by means of
analysis and forecasting procedures.
 The area we have already dealt with assists in
performing the control tasks of the system. It
means that the Communication Module of the
 PROARC system is completely replaced a G2
based control with MMS interface.
As the PROARC is completely PC based and the
our recent G2 version is running on SUN some
additional interfacing problems had to be solved,
which are not shown on the Fig. 4. The big arrows
show the data flow through the system in the first
version, the little ones the potential or alternate
connections.
6. CONCLUSIONS
The paper deals with intelligent real-time control of
FMS. As example the simulation of a real FMS
and then a robotics cell (where the robot is defined
as a virtual manufacturing device (VMD)) and
finally an arc-welding robot controlled by a real-
time computer system are presented. The control
software is implemented using a real-time
intelligent environment (G2).
An interfacing problem was solved when the G2
system and the MAP/MMS network were
connected to reach real-time intelligent control.
These results can be very useful and interesting in
the future when they will be applied for more
complex and sophisticated large scale
manufacturing and assembly systems which hardly
can be designed and controlled without using
simulation and expert systems.
7. ACKNOWLEDGEMENT
The PROARC project (Ref. No. 7831) was partly
supported by the EC in the frame of
Copernicus'93. The authors are grateful to the
partners, the WZL of RWTH Aachen and
University of Veszprém for their contributions.
This work was partly supported by the Hungarian
Research Fund (OTKA), projects T 4122 and T
14549.
8. REFERENCES
Buta P, Springer S: Communicating the knowledge in
knowledge-based systems, Expert systems with
applications, 1992, 5, pp. 389-394
Brill M., Gramm U.: MMS: MAP application services
for the manufacturing industry, Computer Networks and
ISDN Systems, 1991, 21, pp. 357-380
Franklin G., Meystel A. et al.: private discussion by e-
mail in Architectures for Intelligent Control Systems
Discussion List, 1994
<AICS-L@UBVM.CC.BUFFALO.EDU>
Kovács G. L. et al.: Integrated application of Real-time
expert systems for FMS evaluation and control,
Applications of Artificial Intelligence in Engineering
VIII , ed. Rzevski, Pastor, Adey, Vol.2, pp. 835-847,
Elsevier, 1993
Laffey T.J. et al: Real-time knowledge based systems, AI
Magazine, 1988, 9, 1, pp.27-45
Nacsa J., Kovács G.L.: Communication problems of
expert systems in manufacturing environment, in
AIRTC'94 (ed. Crespo), pp. 377-381, Preprints of the
Symposium on Artificial Intelligence in Real-time
Control, Valencia, Spain, 1994, IFAC
Nagy G., Haidegger G.: Object Oriented Approach for
Analyzing, Designing and Controlling Manufacturing
Cells, in AUTOFACT’94 (ed. Moody), pp. 12/1-10,
Proceedings of AUTOFACT’94, Detroit, USA, 1994,
SME
Smith P. E. et al: The use of expert systems for decision
support in manufacturing, Expert systems with
applications, 1992, 4, pp. 11-17
EasyMAP User Guide, CRI Ltd., Denmark, 1993
MMS-EASE User Guide, SISCO Inc., MI USA, 1994
Weck M. et al: A PC-based Offline Programming
System to Assist Robotized Welding, ICRAM’95, pp.
533-538. Proceedings of the Int. Conf. On Recent
Advances in Mechatronics, Istanbul, Turkey, 1995
Somlo J., El Buzidi A.: Reverse Scheduling, a New
Approach to the Solution of FMS Scheduling Problems,
RAAD’96, pp 417—425, Proc. of 5th Int. Workshop on
Robotics in Alpe-Adria-Danube Region, Budapest,
Hungary, 1996