Fuzzy-Guided Mould Level Control in Continuous Steel Casting

Fuzzy-Guided Mould Level Control in Continuous Steel Casting

Copyright © IFAC Automation in Mining, Mineral and Metal Processing, Sun City, South Africa, 1995 FUZZY-GUIDED MOULD LEVEL CONTROL IN CONTINUOUS STEE...

2MB Sizes 14 Downloads 182 Views

Copyright © IFAC Automation in Mining, Mineral and Metal Processing, Sun City, South Africa, 1995

FUZZY-GUIDED MOULD LEVEL CONTROL IN CONTINUOUS STEEL CASTING

Markku Inkinen*, Pentti Lautala* and Erkki Saarelainen**

*Tampere University of Technology Control Engineering Laboratory P.O. Box 692, FIN-33J01 Tampere, Finland Phone: +358-31-3162661, Fax: +358-31-3162340 Email: [email protected]@ae.tut.fi **Finx Oy Revontulentie 6, FlN-02100, Espoo Phone: +358-0-4351602, Fax: +358-0-4351603

Abstract: In steel production the casting machine and especially the liquid steel level behaviour in the mould have major impact on steel quality. In normal conditions conventional level control with two PI-controllers forming cascade control system works fine. However in certain repeatedly occurring situations conventional level control cannot operate in the best possible way. The control system should have some kind of intelligence for observing these situations and deciding how to react. Fuzzy logic is one way to implement the required intelligence. In this paper the structure, operating strategy and control results with the fuzzy-guided mould level control are presented. Keywords: Fuzzy control, Fuzzy supervision, Level control, Setpoint control, Steel industry, Steel manufacture

1. INTRODUCTION

1.2. Automation of the casting machine

1.1 Definition of the casting machine from mould level control pOint of view

In steel production the casting machine (Fig. 1), which consists of ladle, tundish, mould, withdrawal unit, submerged nozzles and stopper mechanism between them and the automation system involved with measuring instruments, controllers and actuators, has extremely significant impact on the steel quality (Kudrin, 1985; Louhenkilpi ed., 1990). The casting process is fast and inherently unstable what means that good automation has very important role in the production. So it is obvious that better operation of the casting machine can be directly seen in better steel quality.

157

When casting machine is defined as above, the automation system can be divided in to three independent parts. The first part is tundish level control by using measurement of the tundish weight. The second part is casting rate control by using measurement of the liquid steel temperature in ladle and knowledge of the steel grade. The third part and the part that has been examined and improved in this research project is mould level control by using measurements of the mould level and stopper rod position. The mould level control is usually implemented by using two PI-controllers which form cascade control system (Fig. 2). Input to the primary controller is the

and control can be best applied to tasks that rely heavily on human experience and intuition, though it is not proved that fuzzy control gives better results than conventional methods. But the power of fuzzy logic is in the use of multiple-valued logic and easy to understand cause-efIect relationship based on linguistic rules, which makes it easier to associate with human reasoning than conventional methods.

Ladle Wcigth: - 60 In

Tempcmure: - I SOO C

Ar-feed 1.S Vmin

Hydraulic

1.4. Project history

Powderfccd -IOkglh

actuator

Mould - oociIIation 3 - eIec:IramopeQc

Stopper rod

I#!!:~ ~--.--Ih.-~

Two years ' industrial research program was conducted in Imatra Steel Oy Ab, Imatra, Finland to study and improve the behaviour of the mould level control system (Saarelainen et al., 1995). Tampere University of Technology's Control Engineering Laboratory participated to this research during the second year from December 1993 to November 1994. During this time the fuzzy-guided mould level control (FGMLC) was developed, tested and implemented.

otiniaa

--cooIiq

Figure 1. Schematic picture of the casting machine

Level ca1rd.ler

-hnish -~ra:\

-nnid

In this paper the structure, operating strategy and control results with the fuzzy-guided mould level control are presented.

Figure 2. Block diagram of the old mould level control system 2. CASTING PROCESS deviation between the mould level setpoint and the measured mould level, and output of the primary controller is remote setpoint of the hydraulic cylinder which is connected to the stopper rod. Input to the secondary controller is the deviation between output of the primary controller and the measured cylinder position. This kind of mould level control system works fine in normal circumstances. In case of disturbance, however, control system does not operate best possible way. The control system should have some kind of extra intelligence for observing these situations and deciding how to react. Fuzzy logic is one way to implement the required intelligence.

2.1. Casting process from mould level control point o/view

The environment and the circumstances that affect to the mould level control and which can be affected in the casting process are restricted to the casting machine as shown in Fig. 1. Liquid steel flows from the ladle to the tundish and from the tundish to two independent moulds (two-strand caster), where it is cooled by water and stirred electromagnetically. When steel, moving downwards vertically, leaves the mould it has been solidified. The liquid steel flow in to the mould is controlled by the stopper rod. Argon is feeded to the steel through the stopper rod. The mould oscillates with amplitude -3 mm and frequency -1.7 Hz, frequency depending on the casting rate. This oscillation prevents steel from adhering in to the mould walls. Casting powder is feeded manually to the mould. Time required for casting one ladle of steel (-60 tn) depends on the casting rate and is approximately 45-60 min. Casting rate (speed) depends on the liquid steel temperature in the ladle and steel grade being usually between 0.6 and 0.9 mlmin.

1.3 Fuzzy logic and control

Fuzzy logic and fuzzy control have been studied and applied to several fields in engineering and other sciences in recent years. For general ideas of fuzzy control, see for example (Driankov, et al., 1993 ; Lee, 1990). Tuning of fuzzy controller can be done based on the values of the PID-controller (Viljamaa, 1992), Makkonen has studied optimization of fuzzy controller (1994) and applied it to fuzzy control of a nonlinear servomotor model (with Koivo, 1994), Frank has applied fuzzy logic to fault diagnosis and process supervision (1994). Bowerie has compared fuzzy logic control with other automation control approaches (1991). It is often said that fuzzy logic

2.2 The level measurement system

The level measurement system (Fig. 3) and the stopper rod mechanism in tundish moved by hydrau-

158

...

quantitIes are not useful in mould level control because they have minor role compared to the steel flow behaviour from tundish to mould itself which is very nonlinear, unstable and difficult to model. Also the possible effects caused by electromagnetic stirring and water cooling are not taken in to consideration in mould level control.

.... . ... \Vater cooling

According to the process measurements the stopper rod position (i.e. measurement of the hydraulic cylinder position) and the steel flow correlates well only during a short time period, i.e. the rod position does not tell much about the actual flow. In normal process conditions these unidealities are not severe but the PI-controllers (Fig. 2) in the cascade system have to be tuned so lazy that in case of disturbance the control result is not good enough.

-1200 mm

Figure 3. Layout of the level measurement

2.4 Disturbances in casting process from mould level control pOint of view

150

200

The most common disturbance is caused by casting powder. Casting powder is added manually to the mould every 5-10 minute leaving approximately 10 mm thick powder bed on the top of the liquid steel. This powder bed has three important functions in casting. The powder bed protects liquid steel from oxidation, absorbs impurities from liquid and travels between liquid steel and mould walls protecting the walls and acting as a lubricant. Therefore casting powder is not a disturbance from casting point of view but it causes fast change to level measurement which control system tries to compensate causing harmful fluctuation in liquid steel level (Fig. 9 a). With FGMLC harmful effects of casting powder can be decreased.

250

Figure 4. Level measurement signal recorded in offprocess situation lic actuator are the most important parts in the mould level control. The level measuring instrument, Berthold LB 300-2, is based on C060 radiation, measuring range is 100 mm and accuracy is ±3 mm. The measurement signal is filtered, first order analog filter with time constant 1 s, before using in control. The analog level measurement signal, filtered (analog, l-ord., T=l), recorded with sampling frequency 20 Hz (same sampling frequency was used by the old level control system and is used in the FGMLC) in off-process situation when mould was half full of solid immobile steel and mould was not oscillating, is presented in Fig. 4. The level measurement error seems to be composed of two components: an approximately white noise and colouraI noise near the frequency 0.2 Hz.

Formation of impurity blocks, clogging, in subentry nozzle between tundish and mould is very undesirable, but quite usual event in casting. Clogging depends mostly on the steel grade. With some alloys it happens always and with some other alloys never. When blocks are formatted, the liquid steel flow in the subentry nozzle decreases, mould level lowers and control system opens the stopper valve. Typical behaviour of stopper rod and mould level when clogging occurs is shown in Fig. 5 (scale in Fig. 5 is cm for mould level and for stopper rod position one interval equals 20 mm). In the final end stopper valve is completely open, but the mould level is still decreasing because impurity blocks are reducing the liquid steel flow too much. Decreasing of the subentry nozzle flow caused by clogging does not usually propagate straightforwardly. Impurity blocks in the subentry nozzle are also disintegrating continuously because of the flow pressure and at the same time new blocks are formed (Fig. 5 at time instant -28 min quite big block disintegrates and mould level rises rapidly).

2.3. Factors that affect to the mould level behaviour Steel flow from the ladle to the tundish, the tundish level and Argon-feed through the stopper rod are all affecting to steel flow from the tundish to the mould and that way to the mould level behaviour but these

159

installing to the real process. For more detail, see (Lautala and Inkinen, 1995). Final tuning of the FGMLC parameters was done in real process environment.

3.2. Structure and operating strategy of the FGMLC

'0

FGMLC is composed of two portions (Fig. 6). The mould level control is implemented by ~ing two controllers forming cascade system same way as before and described in chapter 1. These two controllers are guided by the intelligent portion of the control system. This portion that consists of programmable controller and fuzzy logic unit gives the parameters and operation strategy of the cascade control system by using measurements of the mould level setpoint, mould level and level controller output. In normal process conditions the system acts like ordinary cascade control system, but in the case of disturbance cascade system is guided by fuzzy logic.

10

Figure 5. Example of clogging. Stopper rod drifts upwards during the casting At this kind of situation the flow dependence of the stopper rod position is very unstable and mould level is difficult to control, as can be seen from Fig. 5. (Mould level behaviour was so unstable that this time the casted steel was scrapped.) With FGMLC clogging can be observed and the control strategy can be changed so that clogging can be decelerated and the undesirable effects prevented.

Fuzzy logic unit has two outputs - modified mould level setpoint and modified level controller output and seven inputs. Three of these inputs are mould level setpoint in normal conditions, set by the operator, mould level measurement and level controller output. The other four inputs are constructed from these three inputs in programmable controller, so that they carry the necessary information of powder throws and clogging.

3. FUZZY-GUIDED MOULD LEVEL CONlROL 3.1 Steps in the development of the FGMLC

The first step in improving mould level control was to examine the nature of the casting process and mould level control by collecting data, i.e. doing process measurements. Following measurement signals were recorded: mould level, hydraulic cylinder position, casting rate, level-controller output, servo-controller output. Recording was done mainly during the actual casting process (i.e. from the closed loop), but also off-process tests and recording were done with level measurement system (Fig. 4) and hydraulic actuator for having better knowledge of their real behaviour.

The actual powder throw detection and clogging detection are done in programmable controller, and the actual decision making in all situations is done by fuzzy logic unit. When powder throw is detected the mould level setpoint is modified so that the real steel (i.e. the liquid steel level without powder floating on the top of it) fluctuates as little as possible. Mould level sctpoint

Next step was to construct mathematical model of the process from process data for simulation purpose. Model parameters for steel flow from tundish to mould were estimated from normal operating data. Closed loop identification is not convenient in this kind of noisy process, but it is necessary because of the open loop unstability (Lautala and Inkinen, 1995).

Modified mould level sctpoint

CyJ.inder co: cxroth position hydraulics controller

After the model was validated with recorded data several control strategies were tested with the simulator. The final solution was also implemented to the actual controller and the controller's program was tested with the simulator and debugged before

Fig. 6. Block diagram of the FGMLC

160

-tundish -stopper rod

-mould

..

The modified setpoint is then brought back to its original value in certain time. There are upper and lower limit for the modified setpoint.

Level Control (FGMLC) and control results in actual process environment are presented. a) Old control system

When clogging is detected the PI-controllers parameters are changed so that stopper rod movement becomes stronger and bigger. Then the steel flow from the tundish to the mould suffers from increased flow and pressure impulses that tend to disintegrate formatted blocks from subentry nozzle. With this kind of action the clogging is decelerated. When clogging is detected the setpoint can not be modified without risk, because it can not always be detected whether the rise in the mould level is caused by powder throw or by disintegrating of blocks. But this is usually unimportant, because drawbacks in steel quality caused by powder throws are minor when compared to the effects caused by clogging, as can be seen from Fig. 5.

b) New control system

6

J

.....

I.oJ

cdl

....,j.. T

400

200

600

1000

800

1200

1400

Figure 7. Control results in normal casting conditions with the old mould level control and FGMLC

4. RESULTS

a) Old control system

An interesting feature of steel flow behaviour is

presented in Fig. 7. It shows that stronger stopper rod movement in fact stabilizes the steel flow from the tundish to the mould and that way the mould level. However stronger movement of the stopper rod, i.e. tight tuning of the controller parameters, gives good results in normal conditions but may cause trouble with disturbances and stronger movement is more stressing to the actuator. It should also be pointed out that in Fig. 7 the steel grade behaves very well, i.e. no clogging. The scales in Fig. 7, and in Fig. 8 and 9 too, are same as in Fig. 5. In Fig. 7 b) the setpoint modification is not used.

b) New control system

100

200

300

400

500

An example of control results with old control

system and with FGMLC in case of clogging is presented in Fig. 8. The benefits of FGMLC can clearly be seen. In Fig. 8 a) the stopper rod drifts upwards continuously but in Fig. 8 b) drifting is not so fast and disintegrating of blocks occurs at time instants -200 and -300 seconds.

Figure 8. Control results with the old mould level control and FGMLC in clogging situation a) Old control system Stopper rod position

Behaviour of the old control system and the FGMLC system is compared in Fig. 9. With FGMLC system the real steel level does not fluctuate as much as with the old control system because stopper rod does not move so much when the mould level setpoint is modifi.t when powder throw is detected.

10

20

30 b) New control system

6.5

Stopper rod position

5. CONCLUSION In this paper the mould level control in continuous steel casting was discussed. The conventional mould level control is compared to control strategy that uses fuzzy logic in guiding and supervising role. The structure and strategy of the. Fuzzy-Guided Mould

Figure 9. Control results with the old mould level control and FGMLC when casting powder is thrown to the mould

..'

161

lASTED International Conference in Modelling, Identification and Control. 20-22 February 1995, Insbruck, Austria. Lee, C.C. (1990). Fuzzy Logic in Control Systems: Fuzzy Logic Controller, parts I-II. IEEE Transactions on Systems, Man and Cybernetics. Vo120, no. 2, pp. 404-418. Louhenkilpi, S. ed. (1990). Continuous Casting of Steel 1984-1989. TEKES, Helsinki, Finland. Makkonen, A. and H.N. Koivo (1994). Fuzzy Control of a Nonlinear Servomotw" Model. Proceedings of the 3rd International Workshop on Advanced Motion Control. 20-23 March 1994, Berkeley, CA. Makkonen, A. (1994). Optimization of Fuzzy Controller. MSc. Thesis (in Finnish). August 1994, Tampere University of Technology, Tampere, Finland. Saarelainen, E. , M. Inkinen, P. Lautala and J. Johansson (1995). Steel Caster Mold Level Control by Using Fuzzy Logic. ISS 78th Steelmaking, 54th Ironmaking & 13th Process Technology Conference. 2-5 April 1995, Nashville, TN. Viljamaa, P. (1992). Tuning of Fuzzy Controller. M.Sc. Thesis (in Finnish). November 1992, Tampere University of Technology, Tampere, Finland.

There are two kinds of disturbances in the casting process that affect to the mould level control. The other ones, like high nonlinearity and unstability of the liquid steel flow from the tundish to the mould, the quite big error in the C060-based level measurement and bad clogging feature of some steel grades, can not be removed by FGMLC or by any other control strategy. Of course these characteristics depend on the type of the casting machine and actuators and measuring instruments used. On the other hand disturbances like mould level swinging caused by casting powder throwing and clogging as an event can be handled suitable way by FGMLC and the control results in these situations are clearly better than with the conventional mould level control system. The FGMLC system was commissioned at the beginning of October 1994 to Imatra Steel Oy Ab 's steel plant in Imatra, Finland. Mould level variations have been suppressed clearly. Unfortunately there is not yet a statistical basis to say anything about long tenn effects on steel cleanliness, but the first results are positive. In long tenn the plant expects to benefit from more consistent product quality and decreased rejection rate of blooms.

6. ACKNOWLEDGEMENTS The authors would like to thank following persons for their work and co-operation during the project: Pekka J. Aunola from A&D automation Oy; Jukka Salo and Jukka Saastamoinen from Omron Electronics Oy; Tenho Hatonen, Janno Johansson and Mauri Kuisma from Imatra Steel Oy Ab.

7. REFERENCES Boverie, S. et al. (1991). Fuzzy Logic Control Compared with Other Automatic Control Approaches. Proceedings of the 30th Conference on Decision and Control. December 1991 , Brighton, England. Driankov, D ., H. Hellendoorn, M Reinfrank (1993). An Introduction to Fuzzy Control. SpringerVerlag, Berlin, Heidelberg. Frank, P.M (1994). Application of Fuzzy Logic to Process Supervision and Fault Diagnosis. Preprints of the SAFEPROCESS '94, IFAC

Symposium on Fault Detection, Supervision and Safety for Technical Processes. Vol 2, pp. 531538. 13-16 June 1994, Espoo, Finland. Kudrin, V. (1985). Steelmaking, Chapter Eleven. Mir Publishers, Moscow. Lautala, P. and M Inkinen (1995). Tuning of Fuzzy Level Control in Continuous Steel Casting. 14th

162

PAPER PRESENTED BY MINKINEN

DISCUSSION FOLLOWING TIIE PRESENTATION: J Zietsman: Why is the mould powder added in a batch fashion and not continuously?

M lnkinen: The automatic (continuous) powder addition was used in Imatra steel plant but it did not work very well. So they went back to using the manual addition which is still used. A van Cauwenberghe: It would be interesting to have some information on the Fuzzy Logic Control unit. particularly to see why a second output (modified level controller output) is useful. Allnkinen: Modified level control output is useful especially in clogging situation. The input membership functions are mostly triangular type but there are also some non-convex membership functions. Output membership functions are singleton type. The inference method is max-min and the defuzzication method is centre of gravity. K Szajinicki: I. Not convinced by the results Fig. 8. 2. How is the actual steel level estimated? M lnkinen: 1. It is true that mould level does not behave well with FGMLC, but that is because of the compromise we have to do in a clogging situation. If we have to choose between preventing the total stoppage of the flow from tundish to mould caused by clogging and having as stable mould level as possible, we choose the first goal, because we save more money that way in the long term. 2. The actual steel level is not very important. As long as it is between 3-7 cm. We are more interested in the actual steel level fluctuation and the stopper rod movement (hydraulic cylinder position measurement) which gives us an estimation of this.

163