Control in Cement Production

PLENARY SESSIONS

Copyright t?J IFAC Automation in Mining, Mineral and Metal Processing, Helsinki, Finland , 1983

CONTROL IN CEMENT PRODUCTION L. Keviczky Computer and Automation Institute, Hungarian Academy of Sciences, Budapest, Hungary

Abstract. The paper deals with control in the cement manufacturing. A short description of the technology is given first, then the control problems and applied strategies are discussed. After a short summary of the existing control systems technology the possible applications of advanced control algorithms are treated. Keywords. Cement industry; computer control; adaptive control; stochastic control; process identification. large material storages are applied between the main technological units which make the separate, batch operation of them possible. These units consist of several parallel elements (mills, kilns, etc.) for the same purpose (perhaps with different capacities) to increase the overall capacity and reliability of the production. The main parts of the cement manufacturing can be seen on Fig. 1. Here some blocks can be missing in a given factory, but thick lines denote those can be found in most cases.

INTRODUCTION The cement manufacturing is one of the most fundamental industry from several points of view. Its product can be found almost everywhere in the everyday life and the industrial society is unimaginable without cement. This industry is a big power consumer being within the first ten in many countries. The technology has been known for a long time, so it is not accidental that many-many authors have already written about it. The main problems in the automation of the production are also well known from such excellent survey papers as e.g. Bendy (1965), Kaiser (1970), Ross and Bay (1977), Willis and Simmons (1971).

Next the consecutive five main parts of the technology will be surveyed. The basic technology has many variants, differing in smaller or higher extent, all over the world. Here the common properties will be described.

It is very difficult to write considerable new comparing to the above papers, however, we think that the possibilities of the applications of certain methods and algorithms of the modern control theories have been considerably increased by the widespread control engineering applications of the microprocessors in the early eighties. The so-called application threshold has been decreased by the advanced technology.

Quarrying and preparation The raw materials to cement production are selected based on their chemical compositions. The main component is a calcium oxide (CaO) source. Mostly limestone is used but chalk, marl, dolomite and oyster shells can also be applied. The next two components are silica (Si0 2 ) and alumina (A1 203) obtainable from clay or shale.Quartz and bauxite or other minerals can also be used. The fourth important component is ferric oxide (Fe 203) which can be ensured from iron ore, pyrite or blust furnace slag.

Therefore this paper wants to give new contributions to the earlier references by treating the application possibilities of the advanced control theory extensively, because the continuous processes of the cement production always give nice areas to the pioneering, experimental works from the beginning.

These materials are quarried or purchased and are transported on road, railway or conveyor to the cement factory. The raw materials are usually crushed in the quarry, sometimes in the factory, up till the size as machinery can handle or grind.

DESCRIPTION OF THE TECHNOLOGY The recent technology of the cement production has already been known for about lOO years and only the machinery, the specifications (effectiveness, capacity, controllability) have been gradually improved since that time. The technology itself is a continuous process, however,

If the composition of the raw material is strongly varying a pre-homogenization process is usually applied. This can be done in socalled mixing beds, however, the simplest 1

2

L. Keviczky

solution is to built large stockpiles. Their purpose is partly the intermediate storage, but the pre-homogenizing possibility is much more important, if appropriate vertical or horizontal layering and perpendicular random or controlled extraction is applied. The prepared raw materials are stored in silos for the next technological step which is the raw material grinding and blending system. Raw material blending At the next technological unit the prepared raw materials are mixed during a grinding operation by dry process or together with water in the wet process. The materials are proportioned using a control system to ensure the desired chemical compositions in the final ground mix produced by ball mills generally. The raw materials are more easily ground and mixed by adding water to them (grinding is possible separately,too), however, the later evaporation of the water is very disadvantageous from energetically point of view. This is the main drawback of the wet process, therefore we will concentrate to the dry process in the next parts of the paper but the energy consumption is one of the most vital questions. The wet or dry grinding is continued till the particle size which still increases the speed of the next technological step,namely the clinker kilning. The so-called raw meal ground till the prescribed fineness and mixed according to the given chemical compositions is stored in silos using dry process. Many times these silos are homogenizing ones using air to mix the whole silo content in batch operation before feeding the raw meal into the kiln. Using wet process the slurry is stored in basins under continuous blending. Grinding is generally made by ball mills, many times by applying a pre-crusher with hammers. The efficiency of grinding is very law, only a few percent of the invested energy is devoted to the real comminution, therefore the technology is continuously developed to increase the effectiveness. Recently the closed circuit mills are most widely used with a cyclone (hydrocyclone) separator. At the dry process the middle outlet (feeded at the two ends) closed circuit mills are also frequently applied. A typical blending system is shown in Fig.2. Clinker kilning In this step the mix of the raw materials (raw meal or slurry) is burned clinker in a rotary kiln. This is the heart of the entire technology and the most sophisticated chemical and physical processes can be found here. The rotary kiln is a long cylindrical steel tube inclined downward from the feed end and lined with refractory brickwork. It rotates around its axis. The raw meal (or slurry) is fed in at the upper end while heating is applied at the lower one. Natural gas, fuel oil,

or pulverized coal are generally used. The fuel (with the primary air) is burned and combined with the preheated (secondary) air and the hot combustion gases go up the kiln toward the entering raw material forming a special countercurrent heat exchanger. An induced drought fan at the feed end of the kiln speeds up the heat current. Four main sections with different operations can be distinguished along the kiln system (Willis (1968), Kaiser (1970)) in the wet process: (I) Starting from the feed end the first section is the drying zone, where the water is evaporated from the slurry. The heat transfer is helped by hanging chaines, so this part is called chain section,too. (2) The material moves down to the next section and its further heating causes the start of the decomposition of the calcium carbonate. The released carbon dioxide leaves the kiln with the gas stream through the upper end. This sector is called calcining zone. (3) The material enriched in calcium oxide moves closer to the fuel combustion area of highest temperature. This sector is o called burning zone (about 1400 C). The reactions occurring are more complex and they result in the clinker by combining CaO, Si0 2,A1 20 3 and Fe 203 to form dicalciumsilicate (C 2S), tricalciumsilicate (C3S), tricalciumaluminate (C 3A) and tetracalciumaluminoferrite (C 4 AF). (4) The clinker, some centimeters in diameter, leaving the kiln must be cooled, which is done by a planetary cooler installed on the surface of the kiln around or by a travelling grate located under the kiln end. If the cooler is of grate type then the air is blown through the hot clinker bed supplying the preheated air to the kiln. The clinker is cooled to a degree necessary for mechanical handling, then it LS stored in large quantities. In the dry process the raw meal passes through countercurrent cyclone based pre-heaters (in many cases through a precalciner short kiln, too) into the rotary kiln generally. The heat demand of the pre-heater is supplied by the hot exhaust gas of the kiln. There are many variants of these preheating equipments. The gases leaving the kiln system get into a dust-collector and are exhausted to the atmosphere or to other heat recovery equipments. The dust collected is usually returned to the kiln mostly with the raw material. Typical cement kilning systems are shown on Fig.3. Cement grinding The clinker with average diameter of some centimeters is ground to reduce its particle size adding a small amount of gypsum (CaS04) to inhibit fast setting. Closed circuit ball mills are generally used to reach the given size distribution and specific surface area. A very simple technological scheme can be seen on Fig.4.

Control

~n

Cement Production

A closed circuit consists of three ma~n parts: the ball mill itself, the elevator lifting the mill product to the separator and the separator returning the oversize material to the mill as reject. The separator splits the final (fine) product out of the system. CONTROL PROBLEMS AND SYSTEMS The quality of the produced cement depends on the raw materials and the processing operations, too. The control system of the cement production controls these operations to produce maximal quantity of the cement with prescribed quality and minimal cost. The quality depends also on many variables. The appropriate rate of the basic components determining the setting time, strenght, heat of hydration, expansion, etc. is the most important. The free lime content (FLC) also influences the quality similarly to the size distribution and the relativ surface area. A great many open and closed loop controls can be found in th e cement production, however, the proper control of the operations-triplet proportioning-burning-grinding can ensure to reach the overall control aim, the other controls are auxiliary ones. Therefore these three areas will be investigated next. The above operations-triplet can theoretically be controlled together, however, this seems to be a large scale system in the practice, therefore a reasonable decomposition is necessary. The practical decomposition is solved by controlling these units separately maintaining intermediate materials (raw meal, clinker) of constant quality for the consecutive units. By this way disturbances arrising at a unit propagate to the next unit ~n a least extent. In consequence of the drastic increases in the energy prices a much higher attention has being paid to the dry process, therefore - not disregarding the widely used wet process - only the dry one will be discussed and only the most important control philosophies and systems will be treated, because it is impossible to survey all existing strategies. Quarrying and preparations The following objectives should be achieved: - to produce crushed stone of prescribed average composition and proper size as demanded by the next technological processes - to achieve the desired production rate with minimal cost, maximizing the life of the quarry. Usually an off-line computer control is used to reach the above goals. The exploitation of rocks is controlled by probe-drilling analyzing the layers. The main idea is not using the best rocks only, but the worse ones, too. The control can help to receive an optimal allocation and reallocation of the big machines and to minimize the transport and production

3

costs. This control can rather be considered a management guide. The role of a computer can be much higher at the pre-homogenization, where optimal strategies based on the average stockpile (or blending bed) layer compositions can be obtained for the vertical or horizontal extraction. This control ensures the average chemical compositions and decreases their fluctuations by mechanical tools. Raw material blending The main tasks of this subsystem are the following: - to achieve the prescribed average chemical composition of the raw meal with minimal variance by proportioning the raw materials of minimal cost - to ensure the demanded production rate of the ground mix of desired physical characteristics. This latter task can be solved by a conventional mill control system will be shown at the cement grinding later. The chemical composition control of the raw meal is necessary, bec2use the relative amounts of C2S, C3S, C3A and C4AF formed in the kiln strongly depend on the oxide composition of the ground mix. Denote the most important four oxides CaO, Si0 2 , A1 2 03 and Fe,03 by C, S, A and F, respectively. The purpose of the control is to maintain the relative rates of these oxides. The relativ rates can be expressed by the so-called modulus values, the most widely used ones are: Lime standard (or modulus): 100 C

ML=

2 .8S+I. IA+0.8F

AJ.umini'lm modulus:

!-IA=

A

F

Silica modulus: S

MS= A+F and their possible linear combinations. The most difficult problem of the composition control is the measurement of oxide compositions involving a considerable time delay in the closed control loop. Its value is so large (half or one hour), that many experts expect higher improvements in the quality of control by a possible decrease of the delay rather, than by applying the most intelligent algorithm. Earlier the chemical analysis was made by laboratory tools using material sampled by hand. Recently the automatic representative sampling has been solved and usually an X-ray fluorescence analyzer (RFA) gives the oxide compositions.

L. Keviczky

4

The complexity of the control problem of raw material proportioning depends on the number of materials necessary to blend the desired chemical composition. In most cement factories the most important 3-5 basic materials are used, however, there are places where 8-10 different materials are to be mixed. The reachable quality specifications of this control depend on the changes in the basic material compositions. In a very fortunate case there is no need for control. In many cases, however, the fundamentals of the raw materials are so heavily changing that only a computer control can reach the desired goal. It is not accidental that the first computer controls appeared at this part of the technology. The simplified technological scheme of a raw material blending system is shown on Fig.s. This is the case of three most usually applied basic materials. In the feeder silos there are raw materials of different composition. The weigh feeders are controlled by the computer. A conveyor belt feeds the rubble into the raw mill. Afterwards the ground mix of raw materials (raw meal) gets into the silo. Before the silo the meal is sampled and the sample is analysed by an RFA which provides the oxide compositions for the computer. The computer controls this automatic sampling, conveying, preparation and analysing process and computes the new set points (scale factors) for the weigh feeders using the modulus values as references. Hereinafter only the composition control itself will be treated. A control engineering model of the mill-silo system can be seen on Fig.6 (see the details in Keviczky and coworkers (1983», where r(t) and vet) are the input feed and mill outlet (silo inlet), respectively. The vectors ox and M denote the oxides and modulus values, furthermore w stands for the scale factors of the weighs and ~ is the composition matrix. Linear noninteracting (diagonal) dynamics fm(z

-I.

)=d~ag< ...

ci

,

-I , ... > z

-d

I+d.z ~

is generally acceptable for the mill and the first order time lag approximation with delay is proved to be good in most practical cases, because of the relatively large sampling interval (h=20-30 min) necessary to the automatic material sampling, conveying and the RFA. (The faster dynamics of the mill are much more sophisticated, of course, see later.) The input and output variables

r~

and r' -out of this noninteracting multivariable linear part are material flows (lime, clay, pyrite) and the model is finally coupled for the oxides (ox. and ox t) because C is not diagonal. --~np --ou = -~np

The dynamics of a batch silo can be easily derived and is given by a time varying first order lag -I

Qs(z)=

bo

-I

1

I+alz where I

~s

the unit matrix, b

o

and a

l

vary

(Keviczky and coworkers (1983». G is sometimes called "silo integration: :SNote that 1 if the silo is continuous G (z ) is not time varying with a good approx~ation. The blocks denoted by NL mean the nonlinear computation of the modulus values. On the basis of the above model it can be established that a multivariable coupled system has to be controlled. If the problem can be formulated for the oxide variables, then we have a linear system, if it can be done only for the modulus values, then the system is nonlinear. Since the silo parameters vary in time, the system is time varying. The purpose of the control is to proportion a whole silo content with desired average compositions and to reach minimal variances around these reference values. The task of a batch silo is to decrease the composition inhomogeneities, variances by mechanical way homogenizing the silo content before it gets into the kiln. If the changes in the raw material compositions allow and the control operates accurately a continuous silo can also be applied instead of the batch one. In such cases the silo can be abandoned involving a considerable saving in the capital investment. The usually applied control systems of the raw material blending have already reached the level can be expected on the basis of the classical control theory. Two main directions can be traced back essentially. At the first one the modulus values are controlled and the system is also built on them. One of the most complex control strategies is shown on Fig.7. Here Q means a PI and Q does a PI or I dia2 I gonal regulator, respectively, which is understandable considering the first order time delay lag dynamics of the mill. ~r is the desidered modulus reference vector. The most inner loop is a prediction based control using the model of the mill. It works only if the mill behaves in strongly different way as the model. The block of the transformation M/w is also can be found in this loop. This-means the solution of a linear equations system and needs the average composition matrix ~. One of the drawbacks of the system is that ~-is needed and it must be updated (daily~ weekly or monthly) regularly. A further drawback is that the regulators optimally tuned once work properly only in the vicinity of the original working point because of the strongly nonlinear character of the modulus computations. The first problem can be avoided if the prediction loop or the block ~/~ is missing and Q mani2 pulates w directly. In this case, however, the degree of freedom must be taken carefully into consideration because setting modulus values of number k needs k+1 raw materials. Thus Q 2 manipulates k feeders and the last one is presented itself from the constrains. In the other approach the oxide content is controlled, see Fig.8. for a typical system. It is possible to maintain desired modulus values in this case, too, but the oxide setpoints oX have to be computed corresponding r

Control

~n

Cemen t Pr oduction

to the modulus reference values M . This is the t ask of the block M/ox which- ralso contains a matrix inve rsion-essentially . The drawback of th e system i s that it cannot be applied for raw materi als less than four, because the modulus variables consist of four oxid e values. (In such cases ox perhap s can be chosen by expe ri ence .) The--rsystem is very similar to that shown on Fig .7. and the type of fl and f2 can be the same with the same roles. It shou ld be noted tha t on l y the variables and ox ut chara c t eriz ing the mill out put are meas ureg. The values M '1 and oX' ar e --51 0 ---S~ l 0 computed by the a l gorithm " silo integration" al r eady mentioned.

~ut

The above con trol systems relate to s uch unambi guous cases , when the number of raw mate ri a l s is equal t o the number of the contro lled variables ( or excee ds by one using modulus values). The comp utati on of an economical optimum is needed, as we men ti oned ea rli er , if more than nece ssary r aw materials are used. Mostly two methods of minimi zing a cost funct ion are appli ed . At the f irst method a linear programming prob l em is solved t o make the con trol erro r vector e qual t o zero with minimum cos t . The sum of the s qu ares of con trol deviations is minimized by a nonlinear programming technique a t the second one . A s pecia l attent i on mus t be paid to the problem when the cont rol error cannot be made eq ual to ze ro having unr ea li zab le proportioning combination (because of we i ghs or compos iti on or other constrains) at both approaches. The val ue of a pract i ca l system i s considerably incr eased hav ing s tra te gies capable t o handle the previous difficult cases . Clinker k ilning The control ob j ectives a r e the fo llowing : - t o main t ain clinker qual ity and production rate on a high l evel - to minimize the energy cos t (this unit con sumes the highest amount of ene r gy in the manufacturing process) - t o increase the life of the r efrac t ory brickwo rk by decreasing the number of the expensive shut downs. To con trol the r o tary kil n is the most compl ex problem in the cement production. The reason is because the proce sses acc urrin g a r e the most sophisticated ones as it was shown at th e technological description. It is difficult to reach good control for seve ral r easons : th e kiln processes are not complet ely known, it i s dif fic ult t o get accurate measurements, long time de l ays and ve r y short time const ants can a ls o be found, e t c . The uncertainties of the contr ol phil osophies are proved by many appli e d strategies, which are not so un ambigous as at the raw material blending or cement grindin g . Some of the applied con trol schemes wi ll be pre sented ne x t. The highest conformity exis ts at the manipulated v ar iables. The most us ua l ones are the kiln rotational s pee d (KRS), fee d rate (FER), t hey contro l the material residence

5

time a nd the prod ucti on rate,respectivel y . Their comb inati on det e rmi nes the internal l oading. The other t wo manipul a t ed variables are : the fuel r ate (FUR) and the exhau s t gas rate (EGR). Th ese two have the main influence on th e hea t tr ansfe r and the chemical reactions. The EGR is gene r al l y controlled indirectly by the exhaust fan speed or by a prop erl y built damper , e t c . Sometimes the se latter variables are manipulated indirectly through the heat input (HIN) and th e flame temperature (FLT). The control l ed var i ab l es can be diffe r ent depending on th e chosen strategy . These can be the flame temperatur e (FLT) , burnin g zone temperature (BZT) , exhaust gas temperature (EGT) , material temperatures at the end (MT 1) a nd at the front (MT2) of th e kiln , oxigen concentration (02C) in the exhaust gas , ki ln dri ve power (KDP) and the free lime content (FLC) of th e clinke r. The control of a c l inker coo l er is not di s cussed here , however , it should be noted, that the seconda r y air temperature coming from the cooler to th e kiln , should be maxima l . Besides the regulation of this subsys t em the operation of the kiln should also be taken into consi de r a t ion. The increasing ly spreading pl aneta r y coolers do not need so soph i s t icated r egul a tion. Some control con f i guration can be seen on Fig , 9. One of the most widely applied system is (a), whe r e EGT, BZT and KDP a r e controlled by EGR, FUR and KRS , respectively. At th e next system (b) KRS is constan t, fu rth ermore KDP and EGT are con t ro ll ed by EGR and FUR, respectively. Th~ solution (c) uses the combi n a ti on of 02C and EGT (characteriz in g the air flow) an d BZT as output variables , whi ch ar e controlled by EGR and t he combination of FUR and KRS, respectively. The scheme (d) is also us ual,d iff~ring from (a) that 02C is contr olled by EGT. In case of (f) the material t empera tures MTl and MT2 besides FLT are control l ed through the fictio us HIN as a pr ima r y con trol. The FUR and the secondarv a ir rate (SAR) are compu ted from the above va lue s consi derin g the primary air rat e (PAR) by nonlinear functions FUR=f 1 (HIN , PAR , FLT) SAR=f (H I N, PAR , FLT) 2 in a secondary c ontrol loop. (Kaiser ( 1970» The strategy (e) shows that FLC is a l so a controlled variable and KRS ens ur es a cons tant mass transport load to th e kiln in the func tion of FER . (Heste rlund (1981» There are essentia ll y two reasons of contro l systems of many types. One of th em is the multipl e input/multiple output (MIMO) charac t e r of the process. The best comb inations of variab l es whe n the interaction is the le as t and the contro l is most effec tive a r e continu ously investigated. The other r eason i s that i t is dif fic ult to find the effec ti ve, stable (perhaps op timal ) set points on the s tati c cha racteristics of the process, which r ea lly corr es pond t o each o ther. This fact involves the very high importance of th e l ong experience at th e rotary kiln control. This experi enc e

6

L. Keviczky

can ensure the corresponding set point values and their acceptable domains. The usual manipulated variables are directly or indirectly constrained. The EGR is regularly manipulated indirectly by the induced fan speed for example. When we want to maximize the production rate then the higher production needs more heat, higher EGR and finally we reach a quantitative limit. The setpoint of 02C should be minimized approaching a lower limit necessary to prevent a possible explosion. The specific energy used can be minimized through FLC. Only such a high level of FLC can he maintained which is still under the upper quality standard avoiding the overburning. The determination of the appropriate limits for the above three setpoints and the high quality control to approach them as close as possible can be the direct aim of a good rotary kiln control system. Cement grinding The control has the following goals: - to produce cement with desired fineness (specific surface and size distribution) - te maximize the production rate minimizing the ene rgy cos t . The grinding consumes a lot of electric energy, this is the highest percentage of the production cost. In many cases the use of lower cost off-peak energy is applied if possible. Most of the invested energy c.onverts to hent and or.ly 8 few percent is used for the real size reduction. The energy consumption of most mills hardly depends on increasing the material transport load, therefore the minimal specific consumption CAn be reached by maximizing the production rate. The closed circuit grinding is a sophisticated nonlinear system with considerable time ct~lays and varying time constants, where there are few variables to be manipulated. They can be the fresh feed rate and the speed of the separator spreading plate (shortly separator speed) which give a very limited chance comparing to the complexity of the process. (Sometimes the rotating speed of the mill can also be used.) In a good case the main three material flows: the fresh input feed, reject (or their sum, i.e. the mill input) and the final (fine) product are measured, sometimes the mill output, too. The elevator drive power, which is a very noisy signal, however, proportional to the mill output, is usually measured. The electric ears (microphones) located along the mill give also very noisy signals. Their signals are proportional to the quantity of the material in the appropriate chamber of the mill. This variable is called filling degree. The on-line measurements of the physical characteristics of the finAl product, as particle size distribution or specific surface area, are solved only in a very few places, mostly with experimental character. (The situation is a little bit better at the raw mills in the wet process, where considerable progress has been achieved by laser technique.) In most places

the specific surface area or the proportional Blaine value is measured only by laboratory tools in each half or one hour. The measurements of the other variables as temperatures, ventillation, etc. serve only for safety monitoring the process. A control engineering model of the closed circuit grinding is shown on Fig.IO., where r(t), get), x(t), met), vet), f(t), c(t) and set) are the fresh input feed rate, reject, mill input, mill output,final (fine) product, separator speed and relative surface area, respectively. Here r denotes the size distribution of the input feed which is also input variable but not controllable. The other variables are inner state variables. The roles and explanations of blocks can be found in Keviczky and co-workers (1982). The main noise sources are the changes in the particle size distribution and the grindibility of the input fresh feed. These fluctuations should be eliminated from the production rate and from the cement fineness. The cement specific surface area (s(t)) is well controllable by the separator speed (c(t)). Only a set point control based on static relations can be applied because of the very infrequent measurements. The separator speed influences the separation characteristics: the oversize rate is increased by increasing the speed and vice versa. The system is forced for further grinding increasing the reject, which results in higher residence time and finer final product. Increasing the fresh input feed rate the final product is also increased due to the material conservation law. There exists an optimum of the filling degree where the efficiency of grinding is the best. This point is a stability margin for the input feed rate which should be approached as close as possible to maximize the production rate. It is possible to develop such an algorithm which controls the system to the vicinity of this optimum (see Cs~ki and co-workers (1978)) theoretically. In the practice, however, this point is beyond one of the machinary constrains, therefore it cannot be reached. Regularly the gear of the elevator cannot take the higher load. In each case when the purpose of the control ~s to achieve a given limit as close as possible (see e.g. the control of FLC earlier), the most effective strategy is to decrease the variance of the variable in the first step, then to approach the setpoint close to the limit taking the resulting control error variance into consideration. According to this the best mill regulation is obtained when it ensures the stationary operation of the mill without oscillations and incregpes the input fresh feed till the capacity limit gradually. (The automatic control of this setpoint will be shown within the advanced methods later.) The simplest, most widely used control system - beside the simple setting of the fresh input feed r(t) - is essentially an open loop control

Control in Cement Production with the block-scheme shown in Fig.ll. This so-called input regulator partly ensures the constant mill input x(t) because it computes r(t) from the desired X taking the reject g(t) into consideration: It is possible to apply limits both for x(t) and r(t). The control is open loop, all variables can be found at the input of the mill which results in the name of the regulator. This regulator prevents the fill -up of the first chamber of the mill when large material wave occurs in g(t). The more developed control systems form single or mUltiple closed cascade loops. The most difficult problem in controlling the material flows is that only the fresh input feed is the manipulated variable. Since the other variables of the system as the filling degree e.g. can be used only to speed the control up and to improve its quality in an inner cascade loop. The different control systems - can be found in the practice - are summarized on Fig.12. In the outer control loop using the regulator G 2 the controlled variable Y2(t) can be cfiosen as: - the mill output m(t) (or the proportional elevator power) - the reject g( t) - the circulation ratio g(t) /v(t) (or equivalent variables). In the inner control loop with regulator G I the controlled variable YI(t) can be chosefr as - the filling degree f(t) (signal(s) from the electric ear(s)) - the mill output m(t) (if Y2(t)=g( t)/v(t)). The reference value y depends on the selection of y (t) and it is gefrerally set by an experiencea operator. The advantage of the selection Y2(t)=m(t) is that y is proportional to the production and the c~pacity limit can be approached step by step if the disturbance rejection of the control system is good.enough. The inner control loop is, of course, optional. (The above mentioned things are essentially also valid for the raw mills with dry process.) Since the specific surface area (Blaine value) is mostly measured by slow laboratory methods, these values are generally applied for checking purposes and they are not included in the control . Therefore it is general that the ce ment is overground due to the not very frequent quality control. CONTROL SYSTEMS TECHNOLOGY The cement industry is one of the basic, traditional, continuously developing industry with a long history. The continuous clinker production is more than 100 years old. The life of cement plants is relatively large several decadesand factories built at the beginning of this centur y still work in many countries. This means that the pressure of the capital investment is not so very high as in other, more modern industries. The mechanical handling processes, involving solids flows and not liquids or gases, are essential in the cement manufacturing. Therefore the develop-

7

ment of the instrumentation and automation is relatively slow. The above factors result in a very wide spectrum of the automation in the existing cement factories. The cement manufacturing went through the different phasis of automation (as instrumentation, centralized control, automatic control, comput er control), thus we can still find technological units where the long time human experience (burner, grinder,etc.) is the most important and -at the same time - almost completely computerized control systems with an operator supervision can also be found. The current actual cont r ol system technology is generally applied at the newly designe d and installed factories. In the last decade the application of the programmable logic controllers (PLC-s) has been introduced, then the colour TV displays ensuring a very high level man/machine communication have appeared in the central control rooms. Recently the application of microprocessors and single board computers (SBC-s) for control problems comes into general use and the application of the process control local area networks (PCLAN-s) involved by the large area of a cement plant, is on the eve of success . This development corresponds to the general directions of the control engineering technology and it would probably be much faster if the ge neral crisis of th e eighties would not afflict the building industry. The high quality control of the technolo gical. units in the cement production arrised mea suring, sensoring problems. Besides these the problems of the control look small. At the proportioning control the automatic representative sampling from large solid material flows and the fast analysis of the samples had to be solved . This problem has al ready been solved since most of the computer controls can be found here. At the rotary kiln control the measurements of very high temperatures, the 0 2 content and th e free lime content had to be solved . Th e first two have been solved, however, the FLC can be measured in laboratory until now. The experts still beware of the complete computer control of a kiln. At the cement grinding the automatic measurements of the specific surface area and the particle size distribution are still missing steps, however, promising experiments are reported continuously. APPLICATION OF ADVANCED CONTROL THEORY Theoretically many algorithms of the advanced control theory, methods published on the latest conferences can be applied in t he cement production. The approaches shown below are probably sUbjective ones and other authors with other experience would suggest other strate gies. Herebelow we try to show what kind of advanced control algorithms can be applied

L. Keviczky

8

with a high chance of success at the main operations-triplet. Raw material blending This process is multivariable and coupled with considerable time delay where the disturbances occurred in two frequency domain. The fast fluctuations show a stochastic character, the slow changes can be explained by parameter variations. Therefore it is reasonable to apply a stochastic multivariable regulator with certain adaptivity. Because of the large delay a predictive control is suggested. If the process consists of a batch silo, the control should include a special final value regulation, too. These goals are met by the multiple input/ multiple output (MIMO) minimum variance (MV) self-tuning (ST) regulators based on d-step ahead adaptive prediction. The main goal of the blending control is to decrease the variance of the chemical compositions of the raw meal rejecting the disturbances caused by the changes in the raw material compositions. Furthermore - in case of batch operating silos - every full silo content is homogenized to reduce the composition variations around the average values. These double control objectives can be formulated using a general loss function 2

2

V=E{ I ly(t+d)-y -r II +y l ly -a (N)-y - r II I d=min where ~(t+d), ~r and ~a(N) are the measured composition vector, the reference vector and the average composition vector at the end of a silo filling (t=N), respectively. Here N is the length of the silo filling and y is a positive penalizing coefficient. (Note that instead of ~a(N) any ~a(t+k) can be applied in V.) Here E{ ... } stands for the expectation. The minimization of V results in the so-called MIMO-ST-MV-RAFT strategy introduced by Keviczky and co-workers (1978), ensuring a required average for the finite time of silo filling. This optimal strategy gives a special cascaded control loop shown in Fig. 13. The inner loop is a MIMO-MV-ST regulator and a time-varying K(y,F,t) gain for each channel is applied in the outer loop. Keviczky and co-workers (1978) applied this control strategy with the selections y(t)= M (t) and y (t)=M '1 (t). The operation of -out -a -Sl 0 the adaptive regulator can be seen on Fig. 14., where the controlled variables are ML(t) and MA(t), furthermore the average values MLa(t) and MAa(t) are also plotted. The multivariable regulators contain many parameters to be adapted. A high effort is now devoted to decrease their number drasticly. In some special cases promising possibilities have been found. Hetthessy and co-workers (1983) has shown for a special component matrix ~ that the MIMO system can be decoupled to SISO subsystems introducing new input/output variables. Such a control system is shown on Fig. IS. ,where

a silo pole-cancelling PI regulator is applied in the outer loop instead of the proportional K(y,F,t) which is also possible. Fig.16. shows the operation of this control system where MLa(t) and MAa(t) are again plotted. A SISO-MV-ST regulator was successfully applied

by Westerlund and co-workers (1979) to control the CaO content. They used the identified parameters of a continuous silo to compute the d-step ahead prediction. It should be noted that in case of many raw materials there is no way to apply the MIMOMV-ST-RAFT strategy because of the high dimensions. If a natural decoupling based on the actual ~ cannot be reached then such adaptive strategIes can be suggested where only the static relationships are learned (adapted) and the dynamic control is solved by regulator with constant parameters. This system is shown in Fig.17. where multiple linear regression (in a recursive form to get adaptivity)Ais applied for the adaptive estimation of ~. The working point computation can easily be- combined with the optimization of a cost-function. A Eoninteracting PI matrix regulator is used to eliminate the remained control errors caused by the unaccurate estimates. Now we are at the problem what must be mentioned here and at the following sections, too. There is a serious contradiction between that a good adaption of the working (or set) points needs considerable changes in the variables and that a good control results in minimally varying variables. Because the good adaption needs to disturb the stationary operation of the system a designed artificial intervention may generally be necessary. Cement kilning It has already been mentioned that if the purpose of the control is to achieve technological or standard limits then this task is suggested to be solved by a two step control philosophy. First to reach minimal variance, then to move the setpoint close to the limits. Such cases can be seen on Fig.18. for upper and lower limits. This philosophy can be applied for the rotary kiln control in the form of a two level control system shown in Fig.19. Here an optimal MIMO regulator is in the lower level and an adaptive algorithm can be found in the upper level which identify the static relationships between the variables. Using the identified static process model the optimal setpoints are computed by the next block. It is also possible to apply adaptive regulator in the lower level to learn the dynamics of the process and to improve the transient quality of the control. The static adaptation shown in the figure can be performed between the pairs of variables or by a multivariable regression including interactions. FLC(KDP) shows a decreasing curve. Identifying its parameters the minimal setpoint for KDP can be computed taking the standards and FLC variance into consideration. This subsystem was introduced by Westerlund (1981),(1983) obtaining very nice results.

Control in Cement Production The system can theoretically be extended by identifying the other two functions, too. The situation is very similar for 02C, too, only a minimal, still acceptable setpoint has to be computed here. In connection with EGT, EGR the situation slightly differs because the constraint is generally known for the manipulated variable EGR. Therefore the appropriate setpoint for EGT should be computed from the identified parameters and from the EGR constraint. The two level control has the same problem mentioned earlier. Cement grinding This process is a multivariable, nonlinear one with an inner feedback and large delays. It has distributed parameters where a lumped parameter approximation is load dependent. The particle size distribution and the grindability of the clinker change stochastically causing continuous disturbances to the mill. A two level adaptive control system can offer the best possibilities for the application of advanced control algorithms,here too. A possible control loop can be seen on Fig.20. A cascade loop is suggested here to control the material flow if a signal proportional to the filling degree (f(t» is available. This solution always improves the transient properties. A ST regulator can also be applied in the inner loop using feedforward on the separator speed (c(t» forming a way to decrease the material waves when sudden change occurs in c(t). It is sufficient to use a slower PI regulation in the outer loop to keep the system on the desired working points. The controlled variable is the mill output met) here (or the elevator power proportional to it). The other control loop regulates the specific surface area (s(t» of the cement where set) is usually measured in every hour. The manipulated variable is c(t). Because of the very infrequent samplings this regulator may be an lone, however, a ST regulator to improve the dynamic properties can also be applied independently of the previous loop. The setpoint S is apriori known from the customer's dem~nds. The operation of a ST regulator is shown on Fig.21. with sampling interval h=15 minutes. The figure shows the learning period of the regulator and its stationary operation for another reference value. An adaptive algorithm can be applied on a higher level to control the setpoint M optimally. In most cases it is probably en6ugh to set M properly close to the limit formulated to th~ elevator power apriori. It is also possible to develop such algorithm which approaches this upper limit gradually observing the operation of the closed loop control (computing dynamic variances). This process would maximize the production rate. More complicated control strategy is worth to be applied if the optimal working point needs less elevator power than its constraint. In this case the optimum can be obtained by the adaptive identification of the closed circuit

9

process. Different strategies can be applied depending on the available measured variables. The optimal working point can be computed by simple correlation technique, linear regression or more complicated nonlinear dynamic models. A self-tuning extremum regulator (ST-ER; see Csaki and co-workers (1978» can also be applied. These more sophisticated systems have reached only the experimental simulation level. It should be noted that besides met) the variables get) or g(t)/v(t) can also be selected in the setpoint adaptation loop, identifying another static characteristics in this case. SUMMARY The technology of the cement production was shortly described from control engineering point of view first. The different control problems and the applied general philosophies of the specific areas were then discussed. Some words were devoted to present the recent states and the future trends in the control engineering technology. A separate section dealt with the application possibilities of the advanced control algorithms. It was stated that mostly MV type regulators should be applied combining with adaptive setpoint computations in an upper supervision level of a two level hierarchical control system. REFERENCES Xstrom, K.J. (1970). Introduction to stochastic control theory. New York,Academic Press. Xstrom, K.J., B. Wittenmark (1973). On selftuning regulators. IFAC Journ. Automatica, Vol.9, pp.185-199. Bang-Pedersen, H. (1975). Complete chemical control of the cement plant processes. IFAC Congress. Bang-Pedersen, H., F.L. Smidth (1976). New possibilities in automatic control of cement plants? IEEE Cement Industry Technical Conf.,Tucson,USA. Barton, A.D., D.G. Jenkins, F.P. Lees, R.W. Murtagh (1970). Commissioning and operation of a direct digital control comput e r on a cement plant. Measurement and Control Vol.3. pp.T77-T88. Bay, T., C.W. Ross, J.C. Andrews, J.L.Gilliland (1968). Dynamic control of the cement process with a digital computer system. IEEE Trans. Vol.IGA-4,5. Bendy, W.R (1965). Why automate a cement plant? IEEE Cement Ind. Techn. Conf.,Allentown Cascio, J., J.C. Querido (1976§. Instrumentation and computer control strategies of Arizona portland cement. Csaki, F., L.Keviczky, J.Hetthessy, M.Hilger, J.Kolostori (1978). Simultaneous adaptive control of chemical composition and maxi7 mum quantity of ground materials at a closed circuit ball mill. 7th IFAC Congress, Helsinki. Dozortsev, V.M., E.L.Itskovich, I.V.Nikiforov, 1.1. Perelman (1983). IFAC/IFIP Syrnp.on

10

L. Keviczky

Real Time Digital Control Applications, Guadalajara. ---Gautier, E.H., M.R. Hurlbut, E.A.E.Rich (1971) Recent developments in automation of cement plants. IEEE Trans. Vol.IGA-7,pp. 458-469. Hammer, H. (1972). Computer controlled rawmeal production in cement industry. Regelungstechnik und Processdatenverarbeitung, Vol.5, pp.190 198. Hetthessy, J., I.Vajk, R.Haber, M.Hilger, L. Keviczky (1983). A microprocessor based adaptive composition control system. IFAC Symp. on Digital Control Applications-,-Mexico~ Guadalajara. Higham, J.D. (1971). Dynamic computer regulation of a dry process cement kiln. Proc. of lEE, Vol.1 18,3/4, pp.609-617. ---Hoeni~(1972). Component control in a cement plant using a process computer. Zement-Kalk-Gyps, Vol 1.,pp.31-36. Kaiser, V.A. (1970). Computer con trol in the cement industry. Proc. of the lEE, Vol 58,1 ,pp. 70-77. Keviczky, L., J.Hetthessy, M.Hilger, J. Kolostori (1978). Self-tuning adaptive control of cement raw material blending. IFAC J. Automatica,Vol.14.,pp.525-532. ----Keviczky,L., M. Hilge r, J.Kolostori (1982). On control engineering models of cement grinding mills. XIV Int. Mineral Processing Congress, Toronto. Kunze, E.G., M.Salaba (1979). Praktische Erprobung eines adaptiven Regelunsverfahrens an einer Zementmahlanlage . PDV-Bericht Kfk-PDV 158, Kernforschungs-Zentrum Karlsruhe Gmbh. Lundan, A. O. Mattila (1974). A system for the control of the homogenization of the cement raw meal. 4th IFAC Symp. on Digital Computer Appli ca tions Process Control, Zurich. Ohta, T., K.lshida (1983). Real-time di gital control systems for the cement industry . IFAC/IFIP Symp. on Rea l-Time Digital Control Appli cations, Guadalajara. Otomo, T., T.Nakagawa, H.Akaike (1972). Statistical approach to computer control of cement rotary kilns. IFAC J. Automatica, Vo1.8,pp.35-48. Phillips, R.A. ( ).Automation of a portland cement plant using a digital control computer. Riegel, R.W. ( ). Improved process control of raw material blending and kiln/cooler operation result in major energy savings. Ross, C .W., T.Bay (1977). Energy conservation through automatic controls in the cement industry. 4th Ann.Adv.Contr. Conf. on Conservation of Process Energy Through the Application of Advanced Control,Purdue University . Schulz, R. (1973). Adaptive Regelung einer Rohmehlmahlanlage.Sprechsaal, Vol.106, pp.223-225. Schulze, H. (1974). Adaptive Verfahren zur digitalen Regelung von Kugelmlihlen in Zementwerken. Rege lungs technik,Vol.22. pp.174-177. Schulz, R. (1983). Adaptive control of a ball mill with self-tuning reference model. IFAC/IFIP Symp. on Real-Time Digital

Control Applications, Guadalajara. Steelman, D.M., M. Cochelin (1975). A cement plant control centre on a desk. IEEE Cement Industry Technical Confer~ Montreal. Teutenberg, J. (1971). Prozessautomation einer zementfabrik in argentinien. ZementKalk-Gyps, Vol 4. pp.141-151.--Teutenberg, J. (1981).Decentrale Prozessautomatisierung in der Zementindustrie. Zement-Kalk-Gyps, Vol.34.,pp.113-122. Westerlund, T., H.Toivonen, K.E.Nyman (1979). Stochastic modelling and self-tuning control of a continuous cement raw material mixing system. Control and Decision Conf. Florida. Westerlund, T. (1981). A digital quality control system for an industrial dry process rotary cement kiln. IEEE Trans. on Aut.C. Vol. AC-26,4,pp.885 890. Westerlund, T. (1983). Experiences from a digital quality control system for cement kilns. IFAC/IFIP Symp. on Real-Time Digital Control Applications, Guadalajara. Wieland, W. (1967). Automatic fineness control. Minerals Processing, pp.30-31. Willis, V. (1968). The Symposium on cement automation. Measurement and Control.Vol. I. ,pp. T85-T86. Willis, V., R.T.Simmons (1971). The effect of computer on the desi gn, control and management of cement processes. Ying, P. (1980). Method and parame ters for automatically controlling rotary cement kiln. Zement-Kalk-Gyps, 4,pp.171-176. Young, S.C.K. (1971). Raw material blending - a multi variable control problem. 4th UKAC Control Convention on MultivarIable Control, Manchester.

QUARRYING

CRUSHING

I

I

!----I'---s_To_R_AG_E--'~--~ P URCHASING

--I-RA:-I~~LL

I

\CRlJS.H ING

~

11

! PRE-

HOMO~EUI Z AT ION I

- - --- - - - - 0 -

L-1r--;;;;;;;;-l------, I , STORAGE

,-------, I

HOMOGENIZA TION

I

:~ ~] ~

D

='---- ___________"'_

___ p-,==S,-TO=R=A=G-,E

I p"EHE"TER I IROTAR~ KILN I CO~L ER

.~o

I

y

~

STORAGE

-- -- -------- - -l-----'=-i--

-

-t- -

-

,--____, I STORAGE

ICE "";EN T '''I LL I

r-------'

~

'EO £~

~ :3,

-- -- -- -- -- -- -- -- -- --

! SCi-UPPING

Fig. I Typical basic technological operations

Control in Cement Production

IWET

PROCESS

I WATER

PREHEATER

~W'l '- - - - ' I ORY PROCESS

11

ROTARY

KILN

COOLER

raw meal calc ining

burn ing

I (planetary or grate) ;' typE'

I

RAW MATERIALS

, I

J

I'

\;

V

_~"A'.'IMEAQ

\/

O~ _I_._~~ _ ~ ------

I air

"

! clinker

§ Fig. 2. Typical blending systems Fig. 3. Typical clinker kilning system

I

CEMENT MILL

SEPARATOR

FINAL PROO UCT

REJECT

_FE_E_D_L-_~,--_

__ _MILL __ BALL

~LI

~ w

Fig. 4. Typical cement grinding mill

Fig. 5. Simplified technological scheme of raw material blending control

Fig. 6. Model of the raw mill-silo system

Mr

Q!,

~5ilo

W

2~~~~Q~-=" 11

r

I: I.

rL.-:

L

__ ___

.1

y iI

~I ~=utJ 1'

11

i

\i

I ~ .. -- - - - -'------------------ -

-- -~--

Fig. 8. Typical proportioning system using oxide values

Fig. 7. Typical proportioning control system using module values

12

L. Keviczky

-''0 ','0,'.fU~

~~

'0'

EG T_

co"'!

,<,

Cb '

,'",0"'-'""_[

fl.)R

61 T _

.EGR

!RS.

1IE!'_

fER

,020 _ EGT_

KOP

b

-0'.'£

I

'RS

'0'

,.,

(I '

Fig. 9. Typical rotary kiln control configurations

Fig. 10. Control engineering model of closed circuit grinding

r-------,

g(l

I

~"',

CLOSE D

~ -'~

CIRCU IT

m et)

MILL

/ /

~f(t l

) m (t )

\

I \

~~~----~'

I

1

!

L-~--r_~

I

Fig. 11. Input regulator Fig. 12. Closed circuit mill control systems

'j,IO ~

~!

OF M!fr040-S T·"" " ·R.AFT

-===:::':i,,~~t . ~~----,-~:::::,:!,,~u:;i
"lr o..4 l1 11

"

I

-1-y.,*+-'1hi~"+t-"""'+\~>f¥-I-rnf'oll--!~ i

78~1==;::==;:====:;::===~

Ml<:t~- -="- ~f ~~ ---'l-r-l

7.~_==_ :=====;==~==::=~

Fig. 13. Scheme of an optimal MIMO control system

~:-~ ! : ~" '- - - '-'-" "'tr-I

" ':+-1

·"':Jvf~M+I~~1 ."O"::~t ~~ ~ F~ :"

t

·"'·.':1 I

I

H~I ~I

:

Fig. 15. A decoupled control scheme Fig. 14. Operation of the MIMO-MV-ST-RAFT regulator

13

Control in Cement Production

c_____ ~OU I(t )

I

I "

(or

P! :

--

i\

--:-- .--==-'.:-::: :~-==---~

--=~-.-.- ~ 9!sI (o( l)

Fig. 16. Adaptive controlled average modulus values

Fig. 17. Adaptive control with linear re gression

FUR O FER --

OPTIMA L _ _E.,2 R _ MIWO

REGULATQ

[]

02C K

-

~

KQ~

N

L

11

D-0 _

02CIFU')

-

E~T -

_K ~S

meosured

mJ

-, ~I 1',

- ----' 1 1

~~TIEG!ll

!

FLCIKC~L

sta tic a daptation

outputs

Fig. 18. Quality improvements by optimal control Fig. 19. A two level adaptive kiln control system

~ - --- - - -- - -- - - --- -l

L _---r-- - ' Sr__

e(l)

--L ~ __:-.. . - -

CL OSED CIR CUI T

«I)

JLABl I ~ l_~.~.~~_rS(\} ",( 1)

MI L L ,~ fll )

Ca)

j

5.~2900 r\tv'VA~V if'.f\6y'P,yrv:v
V(1rl a~l es

Fig. 20. A two level adaptive cement mill control

) ',00 )000

p r :'C€'SS

I Iv

2700 1 ~ 1

Cb)

Fig. 21. Adaptive control of the specific surface area