con - fernuniversität hagen · 2012-10-09 · proto-col from the w eb. a 3d scenery is created...
TRANSCRIPT
Control Systems Engineering Group
and
Automation Technology Group
Dr.-Ing. Andreas Jochheim (deputy)
Feithstra�e 14058084 Hagen
Phone: 02331-987-1100Fax: 02331-987-354
E-Mail: [email protected]
Academic sta�: Phone:
Control Systems Engineering Group
Dipl.-Ing. Andreas Bischo� -1118Dr.-Ing. Ulrich Borgolte -1106Dipl.-Inform. Jorge Garrido Serrano -1105Dr.-Ing. Michael Gerke -1107Dipl.-Ing. Mohammadthagi Karimian -1105Dipl.-Ing. Dimitrios Lepentsiotis -1111Dipl.-Ing. Christof R�ohrig -1102
Automation Technology Group
Dipl.-Ing. Erwin Haese -4312Dipl.-Ing. Dipl.-Wirt. Dirk Thi�en -4314Dipl.-Ing. Hagen Wenzek -4789
A VRML- and Java-based online robot simulation tool
Andreas Bischo�, Michael Gerke
VRML as a text-based language is a powerful, nevertheless simple language to build'virtual worlds', which include 3D objects, light sources and animations. The descriptionof these virtual environments is a text �le, usually identi�able from its '.wrl'-extension.As VRML supports animations and sensorics, the user can interact with objects in this'virtual environment'.
A VRML browser, ordinary implemented as Web
Figure 1: Web based robot simulator
browser plugin, interpretes any '.wrl'-�les, whicha Web browser has downloaded via HTTP proto-col from the Web. A 3D scenery is created fromthis �le, which is presented in a rectangular areaof the browser window. The 2D viewpoint intothe 3D world is prede�ned during initialization.However, the user can change any viewpoints withthe help of VRML browsers' the control panel.This allows any possible navigation through the3D VRML world. The functionality and degreeof user interaction for a VRML environments canbe enhanced, if the 'External Authoring Interface'(EAI) of VRML is considered for integration of aJAVA applet. User interfaces via comfortable JAVA applets can be build in order to giveaccess to the VRML environment and to allow higher-level modi�cations. AlternativelyJAVA applets can be used to react on events,which occur in the VRML world. The EAIis a browser plug-in, which handles communication with applications embedded into aWeb page.
We have developed a JAVA applet for our robotic
Figure 2: detailed view of the simu-lated PUMA 560 robot
simulation as an EAI, which provides control ofa PUMA 560 robot via buttons. Especially ad-vanced robot programming is feasible with thisinterface via teaching of locations in 3D virtualspace and consequent initiation of a sequence ofmotions between teach-in locations. This controlapplet implements a path interpolation, which �-nally leads to simulation of a synchronized point-to-point motion.
This tool was developed to support the RichODL-EU-Project (Enriching Open and Distance Learn-ing by knowledge sharing for collaborativecomputer-based modelling and simulation) whichis part of the Socrates ODL Transnational Coop-
eration Project.
[1] http://prt.fernuni-hagen.de/pro/richodl
[2] http://www.web3d.org
A mobility aid for older and slightly handicapped persons
Ulrich Borgolte, Mohammadtaghi Karimian
The aim of the MOBIL project is to develop a modular designed assistive device for thesupport of walking, carrying, and rememerance. MOBIL is supported by the EuropeanCommission (Telematics Applications). MOBIL provides a stimulating human machineinterface. The system consists of a low cost but well designed mobile base platform,optionally equipped with a stand-up support, a height adjustment table, a mechanismfor take up of trays and a bedside table function, a module for active walking assistance,and a module for mememberance and stimulation (�g. 1). Particular consideration isgiven to continous and appropriate user involvement within the design phase, adequatefunctionality, aesthetical design, and modularity.
Figure 1: Sketch of MOBIL system used as rollator (left) or co�ee table (right)
The basis for the technical work is an exact re�nement of the known user demands, whichare represented in all system components as well as in the overall system. In more detail,the modules/functions of the system are:
To support the function of walking and transporting, a platform has to be invented,which can cover all necessary system components. In this context, ergonomic and func-tional aspects are in the focus of interest. The basis of the platform is a vehicle thatsupports the function of walking in the same way as a usual walking aid vehicle does.Interviews with potential users during the de�nition phase revealed a demand for walk-ing aids with active stand-up/sit-down support. This will be included as optional partof the basic system. The MOBIL system in its most simple form is pushed by the user.If even this becomes a problem, it is possible to equip the wheels with electric motors.In this case, the user commands the movement by a force sensor. The stimulating
human-machine interface (HMI) is quite an essential point for using the assistant.The aim is to integrate the single dialogue- and interaction-components into a standard-ized concept of human-machine interaction. A component of the HMI will be a daytime
management system. The servant mode of MOBIL o�ers a functionality useful forpeople only needing help in transportation tasks, e.g. arm amputees. Within this mode,the system follows a person with a special marking. While moving, the system keeps acertain distance to the person, avoiding collisions with objects in the environment.
[1] U. Borgolte: A Novel Mobility Aid for Independent Daily Living of ElderlyPeople. 5th European Conference for the Advancement of Assistive Technology(AAATE'99), D�usseldorf 1999, pp. 267-271.
An Assistive Drive Algorithm for Wheelchairs
Jorge Garrido
Within the TMR-MOBINET Project, Mobile Technology for Health Care Services Re-search Network, an assistive drive software is being developed to help users of wheelchairs,to cope with di�erent daily situations.
The aim of this algorithm is to fusion the di�erent advantages provided by previousworks in the �elds of map building, navigation, localisation and robotized wheelchairs,NavChair[1], Vector Field Histogram (VFH), The Dynamic Window Approach to Colli-sion Avoidance[2], and the Active Kinematic Histogram Method developed by Katevas[3]adding kinematic restrictions. It'll be used a Grid-Type Maps, built through two ways,directly from the a priory knowledge of the environment and, dynamically, from thesensor system information, applying the Kalman Filter. All the decisions taken by thesystem rely on this information and the user's input, giving the user assurance that hecontrols the wheelchair all time long and not the other way round.
Knowing the environment and the user's commands, the drive system takes as a directivewhat the user says. Focusing on the direction pointed by the user and setting the speedto one suitable to avoid collisions. Thus as it moves in a free area the focused beamis as wide as possible, seeking on the driving direction. A far as it gets nearer to anobstacle this focus will be narrowed in order to get precise information about the objectthe user is pointing to. By this way, it is able to approximate as much as possible toany obstacle and when the user doesn't want to go in this direction, he'll correct thetrajectory. Despite we work with the focus area heavier, this doesn't mean that the restof the surroundings information will be discarded. Through this behaviour it will be triedto avoid unexpected and undesired rough movements.
In order to test this algorithm, a simulator is being implemented, which is being developedmaking use of the wxWindows library, in order to get a friendly graphical user interface.By this library the code is portable trough di�erent systems like Windows (9x/NT),Unix/Linux machines (GTK, Motif) and Mac. Following a philosophy of modularity andeasy to modify. Providing by this way a useful tool for future works on this area.
The simulator is composed of a GUI that gives the possibility to edit and modify theenvironment and the vehicle itself. Another module is in charge to simulate the sensorsystem. And both sides are working with a module in charge of managing the mapinformation. The other module involved on the simulator is completely independentfrom the rest and in charge of the control and driving of the agent. Using it, diversecontrol algorithms can be tested. The communication between the di�erent modules willuse TCP/IP which allows the distribution of the whole system on di�erent machines andan opened and standardised way of interprocess communication.
[1] R.C. Simpson, S.P. Levine, D.A. Bell, L.A. Jaros, Y. Koren and J. Borenstein: NavChair:An Assistive Wheelchair Navigation System with Automatic Adaptation. Assistive Tech-nology and AI., LNAI 1458, pp 235-255, 1998.
[2] D. Fox, W. Burgard, S. Thurn: The Dynamic Window Approach to Collision Avoidance.
[3] N.I. Katevas and S.G. Tzafestas: The active kinematic method for path planning of nonpoint, non-holonomically constrained mobile robots. Advanced Robotics. Vol.12 No.4,pp 373-395. 1998.
Genetic Path Planning for Autonomous Mobile Robots
Michael Gerke
Long-range route planning is an important precondition to prepare missions for au-tonomous mobile robots. Usually a lack of energy resources has to be considered,when an exploration path for the mobile robot is prede�ned o�-line. This leads to apath optimization problem with respect to known landscape features. The approachdeveloped is intended to solve this problem based on genetic algorithms.
Example:Assuming the satellite imagery of an outdoor operation area, a primitive landscape isderived including all necessary features for path planning. Chromosome paths are cal-culated based on this map. After some hundred generations of evolution (crossover,mutation, reproduction, tightening) and permanent selection of �ttest chromosomes, theoptimal path is found between two given location of the robots mission.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
......................................................................................................................................................................................................................................
............
.............
.............
.............
...................................................
..
.
.................................................................................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
......................................................................................................................................................................................................................................................................................................................................................................................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...................................................................................................................................................................................................................................................................
..
..
..................................................................................................................................................................................................................
.
..
.
..
.
.
..
.
..
.
..
.
.
..
.
..
.
..
.
.
..
.
..
.
..
.
.
..
.
..
.
..
.
.
............................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
..
................................
...............................................................................................................
.......................
.
.
.
.
.
.
.
.
.
.
.
.
..
.......................
.........................................................................
................
.
.
.
.
.
.
.
.
..
.
.........................................................................
.
..
.
.
.
.
.
.
.
�
�
�
�
�
�
�
�
�
�
�
�
��
�
�
�
�
�
� � �
�
�
�
�
�
�
�
�
�
�
�
� �
�
�
�
.
..
..
...............................
......................................................................................
...............................
........................................................................................................................................................................................................................................................................................................................................................................................................................................................................
......................................................
s
s
Figure 1: Modeling of a known landscape and consequently calculated optimal path
[1] M. Gerke: Genetic Path Planning for Mobile Robots.1999 American Control Conference, San Diego, California, June 1999
[2] M. Gemeinder: Genetic Algorithms for Optimal Path Planning.Diploma Thesis, University of Hagen, Germany, October 1999
Fuzzy Collision Avoidance for industrial robots
Andreas Bischo�, Michael Gerke
In advanced robotics, there is still a strong need for a fast reactive collision avoidancecomputable in realtime in order to prevent the robot from hazardous situations duringmotion. Most of these situations occur due to unforeseen events (e.g. collision objectsentering the workspace accidently) or incorrect remote operation (e.g. scenery misinter-pretation during teleoperation).Handling of collision con icts implies a sensor-based collision prediction and a fast strat-egy to resolve the con ict. Advances solutions do not cause simple blockades to preventdamage, but recalculate actual paths on-line and at the same time intend to reach pre-planned goal positions on deviation paths without any time-out.
In this research project, we have chosen a collision prediction based on 3D image pro-cessing and we have designed a fuzzy rulebase applicable to modern industrial robots forconsequent collision avoidance.
Figure 1 shows one of our lab experiments, where we have proven realtime collisionavoidance to be feasible for a stationary robot even in the case of dynamic collisioncon icts.
Figure 1: Realtime collision avoidance in dynamic case
[1] M. Gerke: Fuzzy Collision Avoidance for industrial robots (in German).34th colloquium on control matters, Boppard, Germany, February 2000
[2] A. Bischo�: Realtime Collision Avoidance for industrial robotsby means of 3D sensor supervision (in German).Diploma Thesis, University of Bochum, Germany, January 1999
Multimedia courseware for Control theory
Michael Gerke, Andreas Jochheim
We report here of a multimedia courseware developed by our group �. In order to improvelearner's understanding of our fundamental course on control theory, we decided to createan additional multimedia CD-ROM course.The approach is to build a completely new presentation according to the speci�c demandsand possibilities of multimedia environments. Audio-visual information mostly replaceswritten screen text.In addition to learning environment, an interactive simulation toolbox (incl. BODE andNYQUIST plots, root locus) is provided to make the exploration of control theories morechallenging.Programming the learning environment has been carried out with an authoring tool. Thecourseware consists of about 500 screen pages.
Figure 1: Screen page with sequential display of diverse elements
[1] A. Jochheim, M. Gerke and W. Laaser: Multimedia Courseware forBasic Control Theory. 1999 Int. Symp. on Computer Aided Control System Design,CACSD'99, Hawaii, August 1999
[2] W. Laaser, M. Gerke and H. Hoyer: Teaching Control Theory by Multimedia19th World Conference on Open Learning and Distance Education.ICDE'99, Vienna, June 1999
�The learning environment is a joint development of our group and of personell at the Centre forDistance Study Development (ZFE, a service department of the university).
Optimal Training Parameters in Multilayer Feedforward
Networks
Andreas Wendemuth, Michael Gerke
We present a systematic investigation of the training behavior for multilayer feedforward neu-ral networks. Usually learning is governed by three metaparameters, which are learning rate,momentum and o�set. We apply a (nearly) exhaustive search method to �nd optimal param-eter sets throughout the complete sequence of training cycles. Therefore the training processitself is regarded as a �nite state network in the search space of metaparameter con�gurations.Minimization of training time is achieved by methods of dynamic programming.
A detailed analysis is given forthe proper choice of widths fornetwork 'beams' in search space.Combined 'learning error' and'error descent' puning strategiesas well as thinning strategies areintroduced to make the investi-gation computationally feasible.This method of global search ispresented in the �rst �gure.
It is shown for a representa-tive set of di�erent trainingpattern, that the duration ofany neural network training islargely independent of both,the metaparameter initializationand the random weight initial-ization.An example of metaparameterdependencies is shown in thesecond �gure, where the numberof time steps is given until aminimum error has been reachedfor any combination of parame-ters � and �.
[1] A. Wendemuth and M. Gerke: Optimal Training Parametersin Multilayer Feedforward Networks.1999 Int. Joint Conference on Neural Networks, Washington D.C., July 1999
Advanced Output Coupling of Multivariable Systems
A. Jochheim, R. Dehs
Beside the well-known design methods for linear and nonlinear multivariable systems by output-decoupling, there are situations where speci�c algebraic coupling conditions are part of a desireddynamic behaviour. For dynamic systems described by the nonlinear, time-variant set of equa-tions
_x(t) = a(x(t); t) + B(x(t); t)u(t)y(t) = c(x(t); t) + D(x(t); t)u(t)
and an additional linear output condition
Ty(t) = 0
with the output vector y(t), the input vector u(t) and the state space vector x(t), an analyticalstate space control solution is �rst published in [1]. As a practical example a wheel speedcontroller design is published in [2] and shows the advantages of this solution.
Currently we are working on extended an-alytical examinations in this special �eldof controller design. One aspect is thereplacement of the linear output cou-pling condition with nonlinear equationslike T (y(t); t) = 0. This allows more exibility in the description of the cou-pling equations. The �gure shows an ex-ample of a nonlinear coupling conditiony(t) � f(u(t); v(t)) = 0 between threeoutputs, visualized as a two dimensionalplane in space. The controller design con-cept consists of two parts: a feedbackcontroller and a feedforward �lter. Boththe controller and the �lter part are func-tions of f(u(t); v(t)) and the model de-scription. With the developed algorithms,the resulting dynamic behaviour can bedescribed as a linear multivariable sys-tem with some unrestricted parametersfor pole placement.
References:
[1] A. Jochheim: Reglerentwurf f�ur Mehrgr�o�ensysteme unter der Nebenbedingungvorgegebener Ausgangsgr�o�enverkopplungen. D�usseldorf: VDI-Verlag, 1995.
[2] A. Jochheim: Die Ausgangsgr�o�enverkopplung zur Reduzierung des Radschlupfes beiomnidirektionalen Fahrzeugen. Autonome Mobile Systeme 1996, pp. 140-149, ReiheInformatik aktuell, Springer-Verlag, 1996.
Adaptive Control of Nonlinear Systems
A. Jochheim, I. Ivanov
Subject of our research activities in the �eld of adaptive control are SISO systems, which canbe described by
a0(t)x(n)(t) +
m�1Xi=1
ai(t)fi(x(t)) = u(t)
The system parameters summarized in the vector a(t) = (a0(t); a1(t); : : : ; am�1(t))T are time-
variant and not complete-known. The functions fi(x(t)) are known and nonlinear. The vectorx(t) describes the output variable and its derivatives: x(t) = (x(t); _x(t); : : : ; x(n�1)(t))T .
The controller design is subdivided into three parts
uS(t) = �a0
nXi=1
n
i
!�ix(n�i)(t)
uD(t) = a0
nXi=1
n
i
!�ix(n�i)z (t)
uL(t) = a0x(n)(t) +
m�1Xi=1
ai(t)fi(x(t))
which needs the estimation vector a(t) = (a0(t); a1(t); : : : ; am�1(t))T as input. Vector a(t) is
the output of a parameter identi�cation toolbox. It calculates the estimation vector online withthe help of the measured output variable and its derivatives (x(t)), as well as the referencevariable and its �rst derivative in xz(t) and the calculated system input variable u(t). Thewhole structure of the system is shown in the following �gure:
Browser-based Remote Access to Laboratory Experiments
Christof R�ohrig, Andreas Jochheim
In distance teaching laboratory experimentation is inconvenient because the studentsusually have to be physically present in the universities' labs. A solution to avoid thisdisadvantage is remote access to laboratory experiments. Providing remotely accessibleexperiments, unique or expensive equipment can be shared between di�erent universities.So, a larger number of laboratory resources is available, and students can choose froma variety of lab experiments. Students have access to the experiments via Internet fromanywhere at any time.
In a �rst implementation, we en-
Figure 1: Web Browser with Experiment
hance an existing CAD tool withadditional communication capa-bilities for remote experimenta-tion [2]. In our new approachusers control the experiments ex-clusively with their standard Webbrowser, no additional softwareis needed. A live video streamfor viewing the experiment andan optional audio stream helpsto provide a laboratory feeling.The communication structure isbased on a client/server architec-ture written mainly in the Javaprogramming language. Detailsof the implementation are de-scribed in [3]. Students may workwith any platform that supportsa Web browser including a Java runtime environment. The browser loads the client soft-ware as Java applets from the server and executes them. Java applets on the client's sideallow the permanent improvement of the software since the applets are loaded when theyare needed. Applets are always up-to-date so no user software upgrading is necessarywhen the software is exchanged.The �gure shows a Web browser which controls our \mobile robot" experiment. TwoJava applets are included into the Web page. One applet receives the live video andaudio stream of the laboratory and the other applet controls the experiment. Details ofthe experiment itself are reported in [4].
[1] http://prt.fernuni-hagen.de/virtlab/
[2] C. R�ohrig, A. Jochheim: Remote Control of Laboratory Experiments. Proc. of the 19thICDE World Conference on Open Learning and Distance Education, Vienna, June 1999
[3] C. R�ohrig, A. Jochheim: The Virtual Lab for Controlling Real Experiments via Inter-net. Proc. IEEE International Symposium on Computer-Aided Control System Design,Hawaii, August 1999
[4] A. Jochheim, C. R�ohrig: The Virtual Lab for Teleoperated Control of Real Experiments.Proc. IEEE Conference on Decision and Control, Phoenix, December 1999
Java Tools for Remote Experimentation
Christof R�ohrig, Radovana Ondrejkov�a, Marek Straube
As only one student at a time receives access to an in-
Figure 1: Scheduler Dialog
dividual remote experiment, schedules and exclusive ac-cess procedures to the experiment are necessary. Forthis task an access management system was developed[1]. The user interface of the access management systemis implemented in Java and executed in the Java runtimeenvironment of a Web browser. Users are able to booklaboratory time in advance. They carry out the wholebooking procedure by themselves in order to choose thetime most appropriate to their needs. For the bookingprocedure, the user has to enter a valid user ID andpassword to pop up the scheduler dialog which is shownin �gure 1. A calendar shows a chosen week with freetime slots indicated. The user can reserve the desiredtime directly within this calendar menu. Remaining times from the given time quotaare monitored and available for future appointments. The actually booked appointmentis stored in a database and is used for access control. At the booked time the accessmanagement system gives the user exclusive access to the experiment.
For visualization of measured data, as well
Figure 2: Analyser Window
as support of system identi�cation and con-troller design, a Java program was devel-oped [2]. The program is executed onlineas an applet in a browser or o�ine as anapplication in a Java Runtime Environment.While an experiment is in progress, the in-put and output data of the experimentationplant are measured. The analyser applet cre-ates graphs from measured data and displaysthem in analysis sheets. The analyser pro-vides user selected colors, automatic or man-ual scaling and rastering. Analysis sheetscan be directed to a printer or stored into a�le. This is a useful feature for the students'�nal elaboration of experimental results.Another option of the analyser tool is thesupport of the controller design. Actually
some automatic tuning methods are implemented. Figure 2 shows the system identi-�cation dialog. As result of the approximation the dialog displays parameters of timefunction and of transfer function.
[1] M. Straube: Skeleton of Internet Application out of Deference Safeness. Diploma Thesis,FernUniversit�at Hagen, Slovak Technical University Bratislava, 1998
[2] R. Ondrejkov�a: Graphical User Interface to Evaluate Experiments in Virtual Lab.Diploma Thesis, FernUniversit�at Hagen, Slovak Technical University Bratislava, 1998
Closed Loop Quality Control with Statistical Methods
Erwin Haese
Quality control using statistical methods is well stablished in industry. However, the usedstatistical methods like Shewhart charts are rather simple, which leads to diÆculties ininterpreting the results. Also, a closed loop quality control cannot be implemented to anindustrial process, if the investigated process cant be modeled with explicit mathematicalformulars in respect of quality.
By using high sophisticated statistical methods as the 3SLS-model (three step leastsquares regression) for simultaneous interdependent stochastic multivariate systems andmultivariate cross-correlation, a quality model of nearby any industrial process can bedeveloped, which is further used to implement a closed loop quality control. In the �gurebelow, a typical process is shown, with quality data Q, production data P and a costfunction KP .
Disturbance variables
Production Process
Control value TControl value V
Control value p
controller calculation
Controlled
system
QualityQuality
Qi�1
Q0�Q
i�1
KP
P si P i
Q0 Q
i
Closed loop quality control
By investigating a simpli�ed regression model Qi = �P i+u, derived from a non-closed-loop operation of the production process, a quality controller can be implemented.
To calculate new production control values P si, the equation array above is modi�ed to�Q = Q0 �Qi�1 = ��P , and needs to be constraint by the cost function and othercritical parameters (eg. Pn < Pk, �Pn < �Pnmax
). This leads to a overdetermined setof constrained linear equations, which solutions are to be optimised to a minimum ofthe above mentioned cost function Kp. A solution for this kind of problem can be foundusing the simplex method.
Now having found the needed changes of the production control values �P , the actualnew production control values may be calculated, the control loop is closed.
By implementing this kind of closed loop quality control, a prediction of actual productionquality, as well as a stabilisation of the process in terms of quality can be achieved, leadingto a reduction of cost and a well documentated quality.
[1] E. Haese, B. Scher�: Statistik in der Automatisierungstechnik - ein neuer Ansatz zurQualit�atssicherung. In: Holleczek, P. (Hrsg.): PEARL 99 Multimedia und Automa-tisierung, Fachtagung der GI-Fachgruppe 4.4.2 in Boppard, 25.{26. Nov. 1999, SpringerVerlag, Berlin, 1999
TIMO : A multimedia learning course o�ering adaptability to
di�erent learning strategies and to learning success
Dirk Thi�en
Many of today's computer based learning courses take into consideration the student'sknowledge of a special subject. For example, based on a test in the �rst learning sessionthe student is advised to leave out or to process certain parts of the course. The mul-timedia learning course TIMO follows a di�erent approach. It is adaptable to di�erentstrategies in learning as well as to a student's learning success.
Adaptability to di�erent learning strategies. Based on empirical research two fundamen-tal strategies in learning have been detected. These strategies have to be seen againstthe background of traditional university's electrotechnical curricula, characterised by thedistinction between lectures and exercises.
In a �rst mode, the multimedia learning course presents itself primarily like a specialistbook. Now the module course is in the focus of attention, which becomes obvious as itis the only module that is combined with a table of contents, allowing the students todirectly access each information unit. The use of the other modules like exercise book,laboratory, base knowledge, real life and so on is only possible if using the cross-referenceswithin course.
In a second mode the multimedia learning course is characterised by the exercise bookmodule, that means learning primarily occurs by dealing with exercises. Now the studentcan easily access the single exercises using a table of contents. Analogous of theory basedlearning, in this mode exploring of the other modules' contents depends on the crossreferences within the module exercise book.
Additionally, the whole multimedia learning course can also be used in an independentmode, where all modules are treated in the same way. The choice of one of these modescan be made by the student and can be changed at any time of a learning session.
Adaptability to a student's learning success. For the modules course, exercise book andbase knowledge the students can classify each information unit as easy, medium or diÆcult,which will be recorded. Depending on that classi�cation, a special strategy for using theinformation unit is suggested to the student, which he can either accept or ignore. If astudent didn't work on a certain information unit so far, it would be classi�ed as diÆcult.
References:
[1] D. Thissen, B. Scher�: A New Concept for Designing Distance Education Courses forStudents of Electrical Engineering. In: Proceedings of ED-Media 99, Seattle (WA), USA,June 19-24, 1999, pp. 1306-1307.
[2] D. Thissen, B. Scher�: A New Concept for Designing Distance Education Courses forElectrical Engineers. In: Proceedings of ICTE 99, Edinburgh, Scotland,. March 29.-31,1999.
[3] D. Thissen, B. Scher�: A New Concept for Designing Internet Learning Applicationsfor Students of Electrical Engineering. In: Human-Computer Interaction: Ergonomicsand User Interfaces, Bullinger, H.J. (Ed.), Ziegler, J. (Ed.), Vol. 1, Proceedings of HCIInternational 99, Munich, Germany, August 22.-26, 1999, pp. 590-594.