ETRE EE) Fuzzy Logic Based Cement Kiln Control AK Swain, Non-member B Subudhi, Member This paper describes the techniques of computer control of rotary cement kiln through fuzzy log ‘automatic kiln contol strategy is based on the priority management of different kiln control objectives and sub-objectives. These objectives for the rotary kiln have been developed 10 control the kin effectively by tackling the unknown interdependence among the different process variables. The design methodology described in the present paper has been implemented and tested on a 2500 tons per day (pd) capacity rotary 4iln cement plant. eyrors: Puliment fae, Weight fer, Clsker, Kiln cont, Pony managemeet INTRODUCTION In 1965 fuzzy logic was developed by Lotfi Zadeh! ofthe University of California. tsa very powerful straight forward ‘method for controlling multivariable nonlinear industrial processes. The cement kiln process is highly nonlinear, ime ‘varying, and few process parameters available for its conto Nonlinearity an time varying behaviour of the process can te tackled by predictive control? and Self tuning conto,” respectively. For inffequent measurements of lw frequency self tuning controllers may converge very slowly or even may not converge. This isthe exact cae in a cement kiln process ‘where some measurements like live weight and fee lime samples are not avilable at regular intervals due to thee dependence on laboratory evaluation." Also ils with identi cal measurements behave dllerenlly. Hence, itis very dif cuttoestablish an accurate mathematical made forthe rotary ln! Human operators contol the cement kiln fom their experience based on houses. These rules are vaguely de- fined linguistic quantities like, High Small, OK etc. For ‘example, i the cocontent in the ext gas from the kiln at the preheater inletis High then reduce the coal fed tothe precal- siner. By Fuzzy Logic Technique, this type of impeci ‘vague linguistic expressions are controlled. The fuzzy logic Aan conto syster* was commercially availabe in 1980, by FLL Smidth (FLS) of Denmark. FLS has developed its own fuzzy programming language: the fuzzy control language? ECL), Til now FLS is working constantly for the further evelopment ofa high level fuzzy logic kiln contol satay. “Thispaper deals with the design methodology adopted forthe evelopment of a very high level fuzzy logic kiln contol system by Ramco Electronics Division, Madras, India FUZZY LOGIC PRINCIPLES ‘The basic structure ofthe rule based on fuzy loge system is shown in Fig 1. It is basically’ consisting of a fuziier, inference engine and defuzzfier.Herit would be appropriate (o define fuzzy set_as a set of ordered pats, ie, S (wpe) | w 6 Uj withthe generic element and rts ade of membership function, where O'S jy S_1.IET;G) is ‘5 K Sain nd B Sabuthl ae ith Department of Hla Engeering, Indra ‘Gandhi Tasos af Tecmo, Sart, Denna Ori 72946 ‘Tis peeved wat ced in Ob 6 195, Ween dctton ca tis per il beer ll abe 3, 1998. 58 FIGURE T BLOCK SCHEMATIC OF FUZZY LOGIC PRINCIPLE, ; ee alpen >| MRE | loename fr ocoed aaroes mS) [eazy mee | OBE weir | EE| | eur aes neat 2 hazy murber and Mi) nis membership function fo all _rina universe of discourse U, then xis called as linguistic variable, wz temperature, pressure ef; ng (i the er set for xz low, medium, high. The fuzzy logic sytem ia ‘mapping from UCR* to R with the assumption U = UL Us Xeessves X Um Where UIC Ri= 1,2... «and x means intersection. Now the function ofeach of the blocks in Fig | hasbeen described ‘The fonzifir is a mapping from a crisp input space web aty cesseael @ U to a fuzzy set S inthe input universe of discourse U. The furry rulebas is aset Rf fuzzy IE-THEN rues for a mult-inpat mol-outpt (MIMO) 5- ‘em, R= (Ramos Ramos where the ith fuzzy rule is Flow} Reo + TE (xis Tx, and and nis Tx) THEN (is Ty, +n is Typ) w where X= (614 9 scssee0tm) © U and Y= Qu. y2s-- soso a) © Rare the input and output ofthe fuzzy logic system. The precondition of Rlgnso form a fuzzy set Tay X Tag X «++. Tig and the control action of Rkao isthe tunion of m independent outputs. Here, these rules can be represented by a fuzzy implication’ 1E() Journal-EL,ha: (Pa Ta X oe Tin) + (By ts oA Tyg) Which sa fuzzy set defined inthe product space UR, where the sign + represents union. “The fuzzy inference gagine maps fuzzy sets in Uto fuzzy sets i R, based upon thd fuzzy IF-THEN rules in the fu22 rule base and compositional rule of inference® i, it matches the precondition of from a fuzzy set Te, X.-..--- Try and perform implication. One example of compositional rule of inference is given in Appendix I. ‘The defuzzitir maps fuzzy setsin Rw crisp point in R. There are three wide spread methods for defuzzification, amongst these the center of area (COA) method" isthe superior one. ¥uzzy logic contro stategy has beon selected fr cement kiln control, asi utilises all he process information from sensors and human experts, which are imprecise and vague. This inetd is easy to understand and inexpensive to implement in cal time. I is non-parametric and ean approximate any nonlinear system in ime domain. FUZZY LOGIC KILN CONTROLLER ‘The block schematic ofthe controller implementation ofthe combined kiln-cooler section of the plant is shown in Fig 2 Kiln-cooler section includes preheater, precaleiner, rotary Kiln, and cooler. Te total control mechanism is divided into two distinct zones one course contol and the other fine Za] FeDREZ **) BLOCK SCHEMATIC OF THE KILN CONTROLLER IMPLEMENTATION. PLANT (uN cooLER SECTION) CONTROLER || CoM SECTION || SECTION ——— MODE SPs SELECTION oureur|| input Ca] PID SETPONTS Ps) PROCESS | CONTROL VALUES MODE ourpur Fuzzy ‘CONTROLER Vol 77. August 1996 T] FOuRES {| INPUT ourpur DESCRIPTION oF THE FUZZY CONTOLLER KILN TORQUE UTRE WEIGHT FREE UME nmogen |) ‘CARBON MONOXIDE SETPONT ‘aL PRODUC. "TION LEVEL, control. Fuzzy controller alone gives the course control and the PID controllers provide fine control ofthe process variable ‘under normal operation periods. But during severe distur- ‘ances, the control action for the set point change is drastic Which makes the PID controller entirely ineffective. Under these conditions fuzzy control bypasses PID, and gives the control action directly tothe kiln,” which can be achieved by ‘bumpless transfer," usually followed in practice for a smooth transfer from manual to automatic control without disturbing the plant. ‘The basic input output details of the controller is shown in Fig 3. This controller is basically designed for «2500 ipd rotary kiln employing dry process. This controller, of eourse ‘can be used for any cement kiln control with modification of the rules in the rule base, and the input ouiput membership functions. The input parameters are available from the respec- tive transducers in tems of voltage. All the inputs were nor- ‘malized to a unit scale, The final normalised output is then scaled tothe required range. * KILN CONTROL STRATEGY ‘The entire kiln operation is very complex, but itis possible divide this into smaller and simple sub-parts soas to facile beiter control. Now each small partis called an objective. The different objectives to be fulfilled during control mechani ® ai Sticky kiln Loss of coating 0FIGURE FIGURES KILN CONTROL oBsecTIvE MODULE “4] PRIORITY MANAGEMENT FOR KILN CONTROL oovecrve FOREN FACTOR 1OS50F | [Sree PROBE eT Seer hom waorr Tana Pacton cALMUATON y MATERIAL LOW GROUP UKE JK CONTROL ACTION ‘I stdce a CARBON | [GEN iy Stability ofthe kiln “Ywoxoxbe || tow iv) Quality ofthe clinker (Production ofthe plant ONGEN \ Low crouP (i) Carbon monoxide (CO) control HIGH os (vil) Oxygen (0, contol Out of the above seven objectives first five are material flow ‘objectives and rest two are gas flow objectives. Some of the hove objectives are contradictory to each other, like quality ‘or production of clinker, and calorific consumption improve- ‘ment. A priority to each objective has been assigned. Now a priority management system is required. The priority manage- ‘ment system used in this paper checks the highest priority jective which should be fulfilled before next objective is tired. For knowing the fate of objective a fulfillment factor is «ssigned (0 each of these objectives. This is called, Grade of Fulfillment (GFF). This priority manageiment system ensures ‘hat the particular objective is to be fired. by introducing a factor called Weight Factor (WE). One objective module is described in Pig 4. The basic elements of an objective module (i) Balfitiment factor (i) Rule block for fulfilment of objective (Git), Weight factor (iv) Timing calculation Fulfillment factor (GFE) indicates the degree of closeness of the particular objective to the actual process value. It is culculated from a set of fuzzy rules giving a value between 0 vid 1. For this fuzzy logic controller each objective has its ‘own rule block and the priority management system takes care ‘of the interaction among the rule blocks, The weight factor (WF) close to unity means that all other higher priority objec- lives are fulfilled. fuzzy rule base is ereated forthe caleu- lation of weight factor. ‘Timing calculation is one of the nportant factor in entire control strategy. It determines the ¢ forthe next objective rules to be fired. This depends on the fulfillment factor. For this case the timing is chosen as. follows «= (GFF x a + b) x minimum control action (3) Where a and b are (wo constants. This is essentially atime delay in the process objective, oo PRIORITY MANAGEMENT SYSTEM FOR OBJECTIVE MODULE , ‘The priority management of different objects are shows Fig 5. From this figure it is clear that loss of coating (LOC) and carbon monoxide (CO)has got the same priority. Whereas the sticky kiln and stability of the kiln in the material flow ‘group has got the same priority. Os low and O; high has got the same priority inthe priority diagram inthe gas Now grup, ‘These equal priority to different objectives has been allowed as they will not occur simultaneously. The priority diagram ‘indicates that if LOC is not fulfilled then the conte state will give less importance to stability of sticky kiln conditions LOC has been given the highest priority, as whem this will ‘occur the kill has to be stopped, which will in turn hamper rest of the operations. Quality is less important than stability of the kiln, as only a stable kiln ean provide good quality clinker. So, quality or production depends on a kiln and iis stable running conditions, ‘Thus, priority scheme is so important for an automat kiln control, and is the basis for the priority diagram to be set Priority management system splits a complex problem inte small and relatively simple subproblems. FEATURES OF THE SOFTWARE ‘The software developed is an integrated package with the ‘major modules, ie, — Communication module to communicate with plant — Controller module = User interface ‘Amongst the other features the operator can access the history of the plant, on-line trends of different process variables, alarms for any process deviation, ‘The data structures used in this software package for imple ‘menting the fuzzy control strategy are: — Name {}, Unit J. Week {}, Day , Hour (, Actual Value, Nos ‘mal Value, Low Limit, Higlh Limit, Previous Val (0 Input Parameters FEC) Journali) Controt Parameters — Time Last Action, Time Next Action, Along with Input Pa- raneters. — Name {), GFF, WE, previous GFF, Time Last Action, Time Next Action, Connected Ob- jective, Preceding Main Objec- tive (ii) Main Objective Gy) Sub-objective _ — Till Time Next Action, Input Parameters, Output/control Pa- rameters, Parent Objective and all the Data suuctures of Main, Objective. ‘Objectives and Rule Base — control ation, WF, fulfillment ming caleulation, RESULTS AND OBSERVATIONS “The computer program of this contol strategy has been de- veloped in C programming language in an Intel $0486 based system. This has tested for a 2500 tons per day (ipd) plant situated at Jayanthipuram, Andhra Pradesh. It has been ob- served that te plant performance improves dramatically with this contol scheme, The clear benefits out of this are: (Higher run fact6r: up to an extent of 8 to 20 percent increase, ’ (Gi) Improved control performance. Gil) Longer equipment lifetime: due to theless interven- tion of human operator. Improved product quality: due to the constant vigil of the fuzzy controller on the process. (7) Greater production: in respect of the constant supervi- sion ofthe fuzzy controller. Greater operational economy: due to the less human ‘operator interaction, Easy maintenance through a much better understand ing ofthe contol strategy ‘All the benefits cited have been derived from the experience ‘of the plant personnel. A clearcut picture can not be drawn, without comparing the plant performance between a manual, fe, by operators anda fuzzy controller one. This is not possible as two different contol strategies can neither be used simul- taneously on arunning plant nor on two identical kilns as they behave differently. One example ofa rule block implementa- tion s piven in Appendix 2 thas been observed that with his improved automatic rotary kiln control system, the plant performances such as clinker quality, cement production, ete ‘increase significantly. This control strategy iseasily mainain- able, adaptive to new kiln and operating conditions. CONCLUSION ‘This paper deals with the fuzzy logic design technique used fora rotary cement kiln control. This control scheme ean be used in many other process industries like refineries, paper mils, polymer plants ere. The entire program requires a very ‘ew changes for adaptation in other processes, ACKNOWLEDGEMENT “The authors acknowledge with thanks Ramco Industries Lim- ited. Madras. and the reviewers for their helpful comments ww ww wip Vol 77, August 1996 Jayanthipuram Cement Plant is a sister concern of Ramco Industries Limited. REFERENCES 4g L LA Zadeh, ‘Panzy Ss, Ifomnaton and Cone" Academic Pres. New ‘Yori vat 8031985, p 338 2. W'Zadeh ‘Outline of New Approach tothe Analyst of Complex yates and Deguion Poceres” IEEE Trauatuon on Syem Mae & Gioemetcr SMC. 1913-028 3.1 J Asom and B Witeamark.“Adapive Con? Addiom Wesley Pubsning Co. May 1989. 4H Duda. "Cement Data Book vol 2, BAUVERLAG Grit Weisbaen, 198, 36 5.1 King and H Mandan. “The Applicaton o Fuzzy Conul Systems {oIndonral Processes” Automatica, a 3 1976, p23S {6.1 Unbers and PJ King. “An Anat of Human Decision Maing erent Kila Conte! and the Inplicaon fot Automaton” Inert Sura of Man Machine Sade, 0 10 1980p I 7.L-P Holmblad un 1 Osergar, "Coto of « Cent Kin by Fuzzy Logie” PL Smueh Review, 67. FL Smdih & Co, Coerioge, Denmark 8.CCLee. Fury Login Cone! Systems Pr fand It IEEE Transaction ‘Siren Man Chbemeer 012,09 1 1990, p08 9-AX Swainand8 Subd Fury PID Hybrid Corr Scheme for Cement Kin Cones Proceeding of IE Itematonal Conference on intelligent (nd rstramentaton, Singapore. 28 lly, 199.913. 10S Bennet "el Time Computer Contr An ltodaction’ Price Hall Imeretvnl (UR) Li, 1988" ‘COMPOSITIONAL RULE OF INFERENCE, Compositional of ieference canbe iteprted a the matching of two faery subst rations Tazz set and elation). HRs fey ean Aramanivene of dscouses (UD) UieV. heh Puy subse ia V which Induced by be compstion OCR a Aas aor ere stand fo mai composition Atersvely, ican be represented 3 (9) = Supe (0), A) whee we U, ¥eV and A ana abit ft ev and ae fry subeof Und V with Az10210 03} poosason Hence, foo 02 01 03] ReJoo 03 04 07 [00 03 04 05 rat = [01 08 02}, then he fazy subset of V can be caesaed by composition i of nln oo a2 a1 07 = (a1 08 02) |00 03 a1 071 bo a3 ot a3) = 00 03 a1 07 we APPENDIX » ANEXAMPLE OF ONE RULE BLOCK Here INCR snd DECR represents Inreasng and Desretsng respectively ‘bo LDECR. MDECR. SDECR, SINCR. MINCR. LINCR AND BZT ‘presents Lage Decree, Medi Decree, Srl Bers, Small oe ‘ree, Mediu Inreae, Lge Ines tnd Buming Zose Tempera. repecivey FLOW (FREE LIME) ANDINCR (KILN SPEED) THEN SINCR (BZD, IP LOW (FREE LIME) AND OK (KILN SPEED) THEN MDECR (SZT) TPLOW (FREE LIME) AND DECK (KILV SPEED) THEN LDECR (BZ) IFOK (FREE LIME) AND INCR (KILN SPEED) THEN MINCR (227 JF OK (FREE LIME) AND OK (KIL SPEED) THEN ZERO (827) FOX (FREE LIME) AND DECR (KILN SPEED) THEN MDECR (BZT) TF HIGH (FREE LIME) AND INCR (KILV SPEED) THEN LINCR (827) IP HIGH (FREE LIME) AND OK (KILN SPEED) THEN MINCR (BZD) [FHIGH (FREE LIME) AND DECR (KILN SPEED) THEN SDECR (827) 6