Вы находитесь на странице: 1из 4

OPTIMISATION OF MATHEMATICAL MODEL OF CLINKER PRODUCTION IN CEMENT FACTORIES Emir M. Omerdi1, Naser M. Prljaa2, Jakub M.

Osmi3
1

Cement Plant Lukavac, Lukavakih brigada bb, 75300 Lukavac, BiH Tel:+387 35 553 182, E-mail: emiromerdic@yahoo.com

2and 3 University of Tuzla- Faculty of Electrical Engineering; Franjevaka 2 75000 Tuzla E-mail: 2 prljacanyahoo.com , 3 jakub.osmicuntz.ba

Abstract: A lot of industrial processes are not optimally adjusted in relation with costs of production. Controllers in factories are adjusted mostly with Ziegler-Nichols method with criteria of error minimisation of controlling loops. In this paper, optimisation is based on mathematical models (quadratic regression function) which are developed for prediction of clinker production in cement industry. These models are optimised with criteria of maximum profit but with technology and quality constrains. Mathematical method used to find optimal profit is nonlinear programming with constraints. This work intends to demonstrate a cheap and efficient procedure for approaching to optimum production. Copyright 2007 IFAC Keywords: Nonlinear programming, Multiple Regression, Profit, Modelling.

1.

the objective BACKGROUND When function f(x) is strictly In industry , it is 1. PROBLEM common practice to FORMULATION A linearly constrainedconvex for all feasible document/store optimization problem withpoints the problem has a a quadratic objectiveunique local minimum important processThe problem of finding variables to paper,technological parameters function is called awhich is also the global quadratic program (QP).minimum. A sufficient disk or other mediawhich will produce the maximum profit is: Because of its manycondition to guarantee during production applications, quadraticstrictly convexity is for Q process. These tasks programming is oftento be positive definite. are performed c1 m( x1 , x 2 , x3 viewed) as a discipline in ,.... x n periodically by shift Max of c 2 Maz + c3 Ugljandc 4 Ost itself. More3.2 Karush-Kuhn-Tucker + workers (writing importantly, though, itConditions report data usually forms the basis of several (1) every hour) or SCADA general nonlinearWe now specialize the or systems (storing data first-order programming algorithms.general to hard disk). We begin this section bynecessary conditions given The process of finding examining the Karush-in Section 11.3 to the x c1 m( x1 , x2 , x3 ,....n ) a best-fitting data Kuhn-Tucker conditions forquadratic program. These M a x model for quantity of c2 M a z c3 C o a lc4 O s tthe QP and see that theyconditions are sufficient for clinker production in turn out to be a set of lineara global minimum when Q positive definite; function of process(2) equalities andis otherwise, the most we can variables includes theWhere are: complementarity following: c1 price per ton of constraints ( Marti 1973).say is that they are 1. Talking with clinker Much like in separablenecessary. the c2 price per ton of programming, a modifiedExcluding persons who oil (mazut) version of the simplexnonnegativity conditions, work on kiln c3 price per ton of algorithm can be used tothe Lagrangian function for and finding out the quadratic program is coal find solutions. which process c4 total rest price data are per unit (workers + controllable 1 electrical 2.1 Problem L( x, ) = cx + x T Qx + ( Ax b ) , and which of 2 power+spare Statement them are parts+raws) important. (6) Maz - oil The general quadraticWhere is an m2. Making table of Coal - Coal program can be written as dimensional row vector. shift reports. Ost - rest MinimizeThe Karush-Kuhn-Tucker 3. Starting stepconditions for a local wise With next constraints: 1 T minimum are given as regressions, f ( x ) = cx + x Qx follows. which find 2 a1 x1 b1 some of (5) (7) models a 2 x 2 b2 subject to Ax b and x L 0, j = 1, ... .., c + x T Q + A 0 n automatically. 0 where c is an n- x j a 3 x3 b3 4. Manually dimensional row vector assessment of (8) .......... .......... . describing the coefficients L these models 0, i = 1,. ... .., A x b 0 m the and finding the .......... .......... . of the linear terms in Q is i (9-a) objective function, and best between a n x n bn an (n n) symmetric x L = 0, j = 1,.. ... , (9-b) them. n matrix describing the j xj 5. Analysis of the (3) coefficients of the quadratic model, ANOVA (10) terms. If a constant term x T c T + x T Q + A = 0 table and exists it is dropped from the (11) pictures of theConstrains (3) are definedmodel. As in linear g ( x ) = 0, i = 1, . .. . .. , ( A x b ) = 0 m i i best model. with physical laws andprogramming, the decision (12) n On the base of reportstechnology of production. variables are denoted by the x j 0, j = 1, .. ... , x 0 with importantAnd final solution will notn-dimensional column 0, i = 1,. .. ... , 0 m process variables thebe aceptable if nextvector x, and the constraints i best equation whichinequality for quality ofare defined by an (m n)To put (7) (12) into a A matrix and an m-more manageable form we represent quantity ofclinker is not satisfied: dimensional column vectorintroduce clinker production for nonnegative b of right-hand-sidesurplus variables y n months March 2004 and June 2004 in 28_days , x ,..x , z ,coefficients. 45 assumeto the inequalities in (7) (x1 2 k 1 that,...z m feasible solution z 2 a ) We Cement Factory and nonnegative slack and that thevariables v m to the Lukavac were (4) exists

INTRODUCTION quadratic function.

regression

constraint

region

is

2. THEORETICAL bounded.

inequalities in (8) to obtainThe goal is to find thewill be insignificant if theT1b = Temperature of the equations solution to the linearalterations are kept small. cyclone 1b, program that minimizes the T3a = Temperature of sum of the artificial4 APPLICATION OF cyclone 3a, variables with the QP FOR T3b = Temperature of c T + Qx + AT T y = 0 additional requirement that MAXIMISATION OF cyclone 3b, complementarity CLINKER PRICE CaCO3 = (13)the slackness conditions be Concentration of and satisfied at each iteration. IfConcrete goal function for Ax b + v = 0 the sum is zero, the solutionmaximisation, which itCaCO3,Temperature of Tsz = will satisfy (15) (17-b). (14)To accommodate (17-b),should be optimised are (1)sinter zone, or (2) with constrains (3). the rule for selecting theResult (predicted clinkerMaz = Quantity of The KKT conditions canentering variable must be mazut (oil) spent in now be written with themodified with the followingquantity) which satisfied24 hours, (1), (2) and (3) must also constants moved to therelationships in mind. t = Working time of have a quality which satisfy right-hand side. inequality (4) and then itkiln (min/day), SB = Quantity of raw xj and yj are complementaryshould be accepted. (15) for j = 1, . . . , n Function m is differentmeal (t/d), Qx + AT T y = c T i and vi (16)arefrom month to month duem = Quantity of clinker complementary for i =to changes in object ofproduced Ax + v = b 1, . . . , m (t/d). (17-a) modelling, but variations in x 0, 0, y 0, v 0 goal function could also be (17-b) The entering variable willresult of changes in priceConjecture is that prices yTx = 0, v = 0 be the one whose reducedof clinker - c1 or energyare: c1=90, c2=500 and cost is most negativeprices c2,c3. Constrains (3)c3=150. Price c4 is not that itsand (4) are result ofimportant because it is co The first two expressionsprovided are linear equalities, thecomplementary variable isphysicality and they are notnotated that optimisation in does not third restricts all thenot in the basis or wouldsubject of change fromtechnology influence on costs of rest variables to be nonnegative,leave the basis on the samemonth to month. (workers + electrical and the fourth prescribesiteration. At the conclusion power+spare parts+raws). complementary slackness. of the algorithm, the vectorThe software tool x defines the optimalEssential Regression Using MATLAB software solution and the vector (Steppan, et al., package for looking of defines the optimal dual1998) is used to find maximum profit we get 3.3 Solving for the variables. next parameters which Optimum the solution of This approach has been produce the maximum multivariable The simplex algorithm canshown to work well whenregression problem. profit for March of 2004: be used to solve (15) the objectivefunction is definite, andMultiple regression SB = 1480 t (17-b) by treating thepositive computationalfunctions m, ( Myers and T1b = 320.3 oC complementary slacknessrequires Montgomery 2002) which Tsz = 1399.8 conditions (17-b) implicitlyeffort comparable to a programmingare subject of CaCO3 = 77.8 with a restricted basis entrylinear optimalisation in this paper Mazut = 67.8 rule (Luenberger 2003). Theproblem with m + n are: t = 1440 min/day procedure for setting up theconstraints, where m is the linear programming modelnumber of constraints and n is the number of variablesFor March of 2004: For these parameters follows. in the QP. Positive semim = 3,7 8 +10,5 8 6S B 1,8 8 3 0 308 predicted production of 1 clinker is m = 868,44.(18) Let the structuraldefinite forms of the 2 1 0 constraints be Eqs. (15) andobjective function, though, T T + 2,3 1 6 3 8 C a C O a tz 10 1b s z 3 M NOTE: c2 = 0 for this (16) defined by the KKTcan present computational difficulties. Van De PanneAnd for June of 2004: month, production on kiln conditions. (1975) presents an was only with oil and same 1 7 If any of the right-hand-extensive discussion of the m = 1,3 6 10,6 1 7S B+ 1,4 1 8 0 T1b saving is for the (19) situation: with no coal, side values are negative,conditions that will yield a 2 1 9 multiply the correspondingglobal optimum even when T3b C a C3 O 7,5 6 7 0 T1bT3a + only oil for heating of kiln f(x) is not positive definite. but managing of process equation by 1. 2 1 1 4 The simplest practical + 1,6 6 4 0 T1b M a z+ 1,4 8 7 with better parameters. If each day in month the Add an artificial variableapproach to overcome any 4 l difficulties caused by semi- 1 0 T3a C o a M a z parameters on kiln are to each equation. definiteness is to add a managed same as calculated for maximal Let the objective functionsmall constant to each ofThe following be the sum of the artificialthe diagonal elements of Qnomenclature is usedprofit, then diagram of in such a way that the costs would be: variables. modified Q matrix becomesin equation (18) and (19): Put the resultant problem positive definite. Although Fig 1. Earning without fix the resultant solution will (a) into simplex form. costs , costs of raws not be exact, the difference and electrical

(b)

power for March of Those controllers do not 2004 a) EarningImplementing the sameSaving for 20 days exist, but PLC with all without fix costs,procedure on equation (19) sensors and actuators exist S13 * and application of obtained costs of raws andfor June of 2004 it could be electrical powerobtained optimal parameters should not be a Saving for 300 days without parameters for maximum hard job. S13 * Financial optimalisation b)profit in clinker production. saving Saving or profit with implementing results of parameters obtainedFor these parameters optimalisation could be (in by optimalisation predicted production of ideal situation, with no clinker in June of 2004 is Clinker produced withfailures of equipment and parameters obtained bywith ideal ) 19-30% and it As final results about m = 1049,4. optimisation of model (19)is very impressive. savings which will be achieved if the personal NOTE: In this monthfor June 2004 would have from control room manage heating of kiln was donenext hardness (indicator of the process with optimal with mix of coal and oil andquality) given in next table. parameters, are given in saving is related to the continuation: same situation but with Table 2 Hardness after 3 REFERENCES and 28 days of clinker as . better parameters. indicator of quality of Luenberger D.G. (2003), clinker for June 2004 Linear and Nonlinear If the saving in i-th * Programming, Kluwer day is Si and Sk Hardness after 3 days (a) Academic Publishers is saving till i-th Hardnes after 28 days day then Marti Lj. (1973), Nelinearno programiranjeOdabrana poglavlja, 5 CONCLUSION k 20 Informator - Zagreb, (b) S k * = Si S 2 0* = Si The process of (b R.H.and D.C. i=1 i =1 optimalisation of profitMyers Montgomery (2002). function in clinker (20) Surface production in cement Response factory was successful and Methodology - Process Product i f the earning in iobtained results are and Optimization Using th day is Zi and presented in this paper. Designed Experiments, earning from1-st In process of looking for till k-th day is Zk* multiple regression no. chapters 14, pp. cumulative then which best 793, John Wiley & Fig 2. Earning without fix functions technological Sons, New York costs , costs of raws and predict electrical power for June of process there were chosenSteppan D.D., J Werner., 2004 a) Earning without fix controllable parameters and and. R.P Yeater (1998). optimal Essential Regression, costs, costs of raws and obtained k 20 electrical power without parameters should be set up Software. Z k * = Zi Z 20 * optimalisation b) Saving values for controllers. = Zi i =1 i =1 or profit with parameters (21) Relative earning obtained by optimalisation

Curve a) on figure 2 is related to earning without * fix costs, costs of raws and S P % = 13* 100 electrical power without optimalisation and curve b) Z13 is related to saving or profit (22)with parameters obtained by optimalisation. Day 10 was the day with Table 1 Relative cumulative savings with malfunction in working of impleme-ntation of optimal kiln (more then 14 hours parameters for March 2004 the kiln was in stand by waiting for reparation) and for this day it could not be Saving for 20 days implemented optimal S 20 * parameters. Saving for 300 days

P% is then

S 300

Table 2 Relative cumulative savings with * impleme-ntation of optimal parameters for June 2004

Вам также может понравиться