Fuzzy Logic and Intelligent Technologies in Nuclear Science
Total Page:16
File Type:pdf, Size:1020Kb
BE9900077 Da RUAN Fuzzy Logic and Intelligent Technologies in Nuclear Science Scientific staff TJ LINS, an acronym for Fuzzy Logic and Intel- and be only a step towards future FL appli- Da RUAN it ligent technologies in Nuclear Science, has cations in nuclear power plants. However, li- Xiaozhong Li been recognized since 1994 as a unique inter- censing this technology as a nuclear technol- national forum on Fuzzy Logic (FL) and intelli- ogy could be more challenging and time con- gent systems for nuclear science and industry. suming. The main task for FUNS for the coming years Programme Started at the beginning of is to solve many intricate problems pertain- 1995, both fuzzy software and hardware spon- ing to the nuclear environment by using mod- sored by OMRON Electronics NV (Belgium) were ern technologies as additional tools and to successfully implemented in BRi. To be al- bridge a gap between novel technologies and lowed by the safety authorities to carry out the industrial nuclear world. Specific prototyp- our on-line fuzzy-control experiment at BRI, we ing of Fuzzy Logic Control (FLC) of SCK«CEN's made several off-line tests and a safety on-line BRi research reactor has been chosen as FLINS' test scheme. first priority. This is an on-going R&D project for controlling BRI'S power level. The project started in 1995 and aims to investigate the In the meantime, we have constructed a demo added value of FLC for nuclear reactors. In this model which is not only suitable for most con- framework, the availability of BRi greatly sim- trol algorithm testing experiments, but also for plifies the effort to validate the model descrip- simulating the power control principle of BRi. tion used. This allows us to concentrate on the optimal implementation of the overall control. Fuzzy logic application to BRi The inter- nal R&D project on FLC of BRI was run according to planning: the hardware interfaces were de- Objectives Based on the progress of the BRi signed and made compatible with the OMRON project (see SCK-CEN's 1995 and 1996 Scientific hardware and with BRI; the experimental re- Reports), we are currently using a permission sults of both the existing control system and from AIB-Vingotte Nuclear to perform the real the FLC proved that the designed rule-base and FLC experiment on-line on BRI. We aim to ben- hardware interfaces provide a solid basis for efit the existing control systems by applying FL the further progress of this ongoing project. as an additional tool for both the safety and economic aspects in nuclear power plants. Based on the present research results, we are now starting the real FLC experiment on-line at The FLINS project reflects a special application BRI. We aim first to realize a real closed-loop domain of FL related to the need for the highest control at BRi, to be sure that both hardware safety standards in nuclear areas. Research in- and software work properly. To ensure the volved in this project will provide a real test maximum safety of the FLC experiment for the (1, fine control):: Figure 1 Electrical circuit connection of Fuzzy Logic Control (FLC) and Normal Control (NC). 100 Exploratory R&D reactor, we designed an automatic switch be- Fuzzy-control demo model The demo tween the FLC experimental operation and the model, based on the working principle of BRI Normal Control (NC) operation. The switch can power-level control, was constructed to show only be connected, automatically, with either the possibilities of using fuzzy-rule-based sys- FLC or NC, to keep at least the NC of BRi (Fig. 1). tems and to demonstrate the added value of practical applications of FLC. We proceeded Following the above step, we change the de- in three steps. The very first approach was a sired value in two directions (increase or de- normal fuzzy-control algorithm based on the crease) when the system becomes stable. We Mamdani model. Our second step was to work then further observe the reaction of the sys- out a PID algorithm and compare the two ap- tem, namely the output power of the reactor proaches. The results have shown that fuzzy and that of FLC. We however only consider a control is more flexible, more robust, and eas- reasonably low power level at BRi during this ier to update than the PID control. As a third very first test, which is acceptable because such step, our demo model can now simulate BRl's an experiment follows exactly the same work- exact control behaviour. That is, the water- ing principle as at the high power level. level control of the demo model is compara- ble with the power-level control of BRI. Our As shown in Figure 2, our safety design of FLC current research in the demo model permits at BRi requires extra support software. After further implementation of both fuzzy hard- turning the power on, the PLC always first reads ware and software at BRI. Moreover, one of the each input signal from the A/D unit. Then it self-learning systems, the automatic rule-base checks each signal against its corresponding built-up for BRi control, can be copied with this alarm boundary. If any of them is outside the demo model. boundary, it stops running and NC automati- cally takes over. Otherwise, it proceeds with An advanced fuzzy-control algorithm, called outputs to control the A-rods and C-rods. Since adaptive fuzzy control, is self-learning. Com- this process is run continuously during the ex- plex conditions, where fuzzy-control rules are perimental period, the FLC experiment at BRi difficult to define, require a dynamic controller will be safe. For extra safety, we set the alarm system of self-regulating fuzzy-control rules. boundary at a smaller value than that under Basically, the system can judge its former per- NC operation. formance and adjust control rules by itself. This algorithm has been successfully applied After having experienced off-line fuzzy-control to the demo model, showing a better perfor- test at BRi, we now foresee an on-line fuzzy- mance than normal fuzzy control when the control experiment at BRi for 1998. All tested fuzzy-control rules are not good. In the exam- algorithms in the demo model will be applied ple of Figure 3, the water level starts at 0 cm in the BRI project as soon as permission is and increases toward a set value of 20 cm, first granted, first in steady-state conditions, then under the analytic controller (near 0 cm), then in dynamic ones. under the fuzzy controller. Actual power i>:; lofA-rd'cis::::: lOesireci power ^Pd IReactorperiod :P;6:sitiorl:of A-rbdsi -;w-'fc Ppsitiisri! Figure 2 Extra yes;/'Any out of\no support software Normal control Fuzzy control i boundary? for safety design :{fuzzy: Cont;ii>l off) of FLC at BRI. Fuzzy Logic and Intelligent Technologies in Nuclear Science 101 Following FLINS'94 and FUNS'96 (the first and Level I second international FUNS workshops), FLINS'98 (September 14-16, 1998) will aim to bring to- Fuzzy adaptive control gether scientists, researchers, and people from industry and to introduce the principles of intelligent systems and soft computing such Fuzzy control _ as FL, neural networks, genetic algorithms, and any combination of the three, knowledge- based expert systems, and complex problem- solving techniques within nuclear industry and related research fields. The FLINS'98 proceed- ings will be published as a book under the ti- tle Fuzzy Logic and Intelligent Technologies for 10: Nuclear Science and Industry (World Scientific, •Tirn:C[rninutes] iM.li Singapore). Figure 3 Given a set of bad rules, the control effect with the learning function is much better than that without it. Scientific partners Universiteit Gent (UG) — Taking advantage of the demo model as a test Physics and Electronics Laboratory (TNO) bed for new algorithms, especially learning ones, we will further research a new algorithm Sponsor European Technical Centre (ETC-OMRON) based on neural networks. Nuclear safety: the special issue of Fuzzy Logic and Intelligent Systems The spe- Books published in 1997 cial issue Nuclear Safety was edited upon in- vitation by FLINS and published in the Korea X. Li, D. RUAN, "Novel Neural Algorithms for Journal of Fuzzy Logic and Intelligent Systems. Solving Fuzzy Relation Equations," in: D. Ruan, ed., Also, FLINS has been invited to edit the special Intelligent Hybrid Systems: Fuzzy Logic, Neural issue Intelligent Systems at FLINS'96 for the In- Networks, and Genetic Algorithms, 59-89 (Kluwer ternational Journal of Intelligent Systems. Nine Academic, Boston, Massachusetts, USA, 1997). papers selected out of a total of 52 in the pro- ceedings of FLINS'96 have been reviewed and D. RUAN, ed., "Intelligent Hybrid Systems: Fuzzy will be published in 1998 in this special issue. Logic, Neural Networks, and Generic Algorithms" (Kluwer Academic, Boston, Massachusetts, USA, 1997). FLINS Volume 6 entitled Intelligent Systems: Fuzzy Logic, Neural Networks, and Genetic Al- D. RUAN, Guest Editor, "Nuclear Safety," Special gorithms was published by Kluwer Academic Issue of Korea J. Fuzzy Logic and Intelligent Systems Publishers. Designed as a natural continuation (Official publication of the Korea Fuzzy Logic and of the previous two FLINS volumes Fuzzy Set Intelligent Systems Society, 1997). Theory and Advanced Mathematical Applica- tions (1995) and Fuzzy Logic Foundations and Publications in 1997 Industrial Applications (1996), it aims to pro- vide researchers and engineers from both aca- X. Li, D. RUAN, "Novel Neural Algorithms Based on demic world and industry with up-to-date cov- Fuzzy <5 Rules for Solving Fuzzy Relation Equations: erage of new results and with an overview of Part 1," Int.