Real-time operation guide system for sintering process with artificial intelligence

Vol.12 No. 5J. CENT. SOUTH UNIV. TECHNOL.Oct. 2005Article ID: 1005 - 9784(2005)05 - 0531 -05.Real-time operation guide systemfor sintering process with artificial intelligenceFAN Xiao- hui(范晓慧), CHEN Xu-ling( 陈许玲), JIANG Tao(姜涛)， LI Tao(李桃)( School of Resources Processing and Bioengineering ,Central South University, Changsha 410083, China)Abstract: In order to optimize the sintering process， a real-time operation guide system with artificial intelligencewas developed, mainly including the data acquisition online subsystem, the sinter chemical composition controller，the sintering process state controller, and the abnormal conditions diagnosis subsystem. Knowledge base of the sin-tering process controlling was constructed, and inference engine of the system was established. Sinter chemical com-positions were controlled by the strategies of self-adaptive prediction, internal optimization and center on basicity.And the state of sintering was stabilized centering on permeability. In order to meet the needs of process change andmake the system clear, the system has learning ability and explanation function. The software of the system was de-veloped in Visual C+ + programming language. The application of the system shows that the hitting accuracy ofsinter compositions and burning through point prediction are more than 85%; the first grade rate of sinter chemicalcomposition, stability rate of burning through point and stability rate of sintering process are increased by 3%，9%and 4%，respectively.Key words: sintering process; process control; artificial intelligenceCLC number: TP182Document code: A1 INTRODUCTIONequipment has been equipped in large and middlescale sintering plants since 1980s. So next the atFor a long time， sintering process has beentention should be focused on investigating thecontrolled mainly depending on operators' experi-method of total control for sintering process andence. Because different operators have differentdeveloping the optimum process control system tooperation knowledge，decision- making ability ancrealize the standardization and automation of sinte-responsibility, manual operation inevitably resultsring production.in the fluctuation of operation control， and thenthe production is influenced. Especially， with the2 CONTROL FUNCTION OF SYSTEMdevelopment of large- scale sintering equipment ，the influence is greater. The general mathematicalThe real-time operation guide system for con-models can only be used to predict some of thetrolling sintering process mainly has three parts:process parameters and control part of the sinte-1) sinter chemical composition control ;ring process because of complexity of sintering2) sintering process state control;process and the related influential factors, thus it3) diagnosis of abnormal conditions duringis difficult to control the whole sintering process.sintering.The rapid development of artificial intelligence pro-vides a new way for the control of sintering2.1 Control of sinter chemical compositionprocess.2.1.1Self- adaptive prediction of chemical com-Artificial intelligence system has been appliedposition in sinterin the sintering plants abroad, such as Japan, Aus-Chemical composition of the sinter is mainlytralia，and good results have been achieved-1-1.affected by the raw materials (see Fig. 1). ThereBut in China, though there are more than 310 setsare more than several hours from proportioning toof sintering machines whose total sintering area isanalyzing sinter, and the time lag is long. So it isbeyond 21 300 m2，the automation level is low,necessary to predict the sinter composition. Thewhich limits the improvement of sinter quality and中国煤化工al network can avoidquantity. The process monitoring and controllingestamathematics model,MYHCNMHG①Foundation item: Project (5037 4080) supported by the National Natural Science Foundation of China; project(030609) supported by theInnovation Project of Postgraduate Education of Central South UniversityReceived date: 2004- 11 - 20; Accepted date: 2004 - 12 -24Corresponde8sAN Xiao hui, Professor, PhD;Tel: + 86-731- 8830542; E mail: JOEW @ mail. csu. edu. cn●532●Journal CSUT Vol. 12 No. 52005and can realize the nonlinear mapping of system.interval optimization is suitable for the sinteringThe self-adaptive and self-learning characteristicsprocess control. The reasons are as follows.of neural network can track the dynamic change of1) The sintering process control is based on athe system!large quantity of real-time production data. Because of the influence of noise and unknown dis-(C) (FeO) (TFe) (SiO2)1 (CaO) (MgO)\turbance，there is error between the actual dataand measurement. So the strict point optimizationis impossible. But the interval optimization can decrease the influence of measurement error.(FeO)2 (TFe)2 (SiO2)2 (R)2 (MgO)22) The point optimization results in the fre-quent adjustment of process， which affects thenormal production. But the interval optimization(C)3 (FeO)3(TFe)s (SiO2)3 (CaO)3 (MgO)3can avoid adjusting the process frequently.The production parameters can be divided intoFig. 1Main factors influencing chemicaloptimal interval (indicated by 0)，normal intervalcomposition of sinter(indicated by十1 and一1)，abnormal interval (in-( * )1一Raw material composition;dicated by十2 and - -2). The aim of interval opti-( * )z一Sinter composition;mization is to make the parameters kept at optimal( * );一Return fines composition;A- +B一 Influence of A on B .interval. The parameter interval division is shownin Fig. 2.2.1.2 Control strategy based on basicityChemical compositions of sinter chiefly includeAbnormalNormsOptimalR(basicity)，TFe, SiO2 CaO, MgO. It is ideal thatintcrval.pointintervintervalall chemical compositions reach optimal conditionssimultaneously，but it is almost impossible to opti-mize every chemical composition at the same timeFig.2 Parameter interval divisionbecause of the high relativity and random variationof chemical compositions. So there must be a focal2.1.4 Control of chemical composition in sinterpoint in controlling chemical compositions.For the sake of realizing the control strategyAt present, most blast furnace burden is madeup of major high-basicity sinter and a few acid pel-centred on basicity， avoiding the great fluctuationlets or coarse ore. The optimal basicity or optimalof production，and reducing the influence of pre-diction error， the chemical compositions of sinterbasicity ratio of sinter to pellet is determined bywere controlled in accordance with the basicityexperience. The mass ratio of sinter to pellet is de-state and its change tendency which is determinedcided in accordance with blast furnace slag basici-by the past, now and future values of basicity.ty，equipment capacity and raw materials supply.So the fluctuation of sinter basicity indirectly af-2. 2 Sintering process state controlfects the normal production of blast furnace. As a2.2.1 Control strategy centred on permeabilityresult, the control strategy of sinter based on baThe sintering process state is popularly con -sicity was proposed for stablizing the blast furnacetrolled by stabilizing the location of burningproduction.through point (Xrp)，which is mainly affected byPriority is given to sinter basicity control inbed permeability. XETp will be advanced if the per-control strategy based on basicity， and when sintermeability is good. On the contrary， it willbasicity other than other compositions satisfies thelagged behind. So only with the stabilization andproduction requirement, measure may not beimprovement of bed permeability, can XBTP andtaken; and when sinter basicity is not contented, .sintering process be stabilized. Therefore the sin-even if others are satisfactory，the measures musttering process control should be made centred onbe taken[8.9].perr中国煤化工siving priority to per-meal2.1.3 Control strategy of interval optimization2. 2MYHCN MH Glility .The point optimization aims at optimizing thThere are a number of variables to reflect per-point，and the interval optimization aims at optimi-meability, such as wind flow through sintering bedzing the optimal interval around the optimal point.(Q)，vertical sintering speed (U⊥ )，location ofThe characteristics of sintering process controlburning through point (XBrp)， waste gas tempera-make it d妇而教据o realize point optimization, buture (T) and negative pressure of main duct (Op).FAN Xiao-hui, et al: Real- time operation guide system for sintering process with artificial intelligenceQ and T are easy to be disturbed by other factors，the reason for abnormality， and at last gave thesuch as leakage rate，seasonal change, coke level,operation guidance. Knowledge base and inferenceand XBTP is a time- delay variable. So the permeabil-engine are the principal components of the system.ity was judged by the variables of V⊥，Op and the2.3.1 Organization of knowledge basepredictive value of XBrp.The knowledge for diagnosing abnormal con-2.2.3 J udgment and prediction of Xerpditions includes four categories， i. e. production"he location of XBTP was calculated by wastedata, facts，heuristic knowledge and meta knowl-gas temperature curve of wind box. Because it isedge. Facts include dynamic facts， which reflectaffected by the leakage of discharge end, the wasteproduction state, and static facts that reflect thegas temperature value may be lower than its actual .technological characteristics， production require-value，and the curve may present an extreme valuement and sever for system inference. Heuristicwhile the bed is not burned throughly. In order to .knowledge is accumulated by sinter expert in long-ensure the judgment accuracy, XBTP was correctedtime production practice. In this system，it is ap-using the waste gas temperature of large chamber.plied mainly to judging production state, analyzingThe waste gas temperature of large chamber, whenreason and guiding option. This judging, analyzingthe thermal state was normal and XBTP was suit-and guiding process is productive rule in total. Me-able, was regarded as a normal value. The differ-ta knowledge is mainly applied to determining theence△T was calculated by comparing the mea-solution order of every subtask， and gives eachsured value with the normal value and the△T wassubtask solution strategies and required knowledgethen multiplied by a weight factor a，whose valuebase name. It is applied to tuning in the initialwas chosen to be 0. 02 winbox per C by experi-facts before solution of subtask, storing results af-ence. So XBTPO was corrected to XBTPm a11.12]1ter solution, and coordinating the solution processXrETPm = XgTpo一aOTof every subtaskl13.14] .Based on long- time sintering production expe-The multiplicity of the knowledge type andrience，when the sintering production is normalthe variance of each type' s express methods deter-and XBrp is stable, the inflexion point of wind boxmine the multi- base structure of knowledge base,temperature curve lies in certain wind box, and thenamely a three-base structure consisting of data-inflexion point changes with XBtp. So XBTP can bebase，fact base and rule base. This structure dpredicted by the location of the inflexion point ofcreases its search space and enhances solution effi-the waste gas temperature curve.ciency and reliability of the system.2.2.4 Control of sintering state2.3.2 Design of inference engineThe sintering state can be controlled accordingThe expert system for diagnosing abnormalto the result of permeability judgment. If the per-conditions was based on the real time productionmeability is good, strand speed and bed height willdata and realized through judging state, analyzingbe adjusted to stabilize XBTp. But if the permeabili-causation and guiding operation. At first, the sys-ty is bad, the raw parameters such as raw mixtem judged the state of every parameter accordingmoisture, mix size, ignition temperature and coketo the real time production data, and then analyzedproportion, will be first controlled to improve sin-these states and found the cause, finally offeredtering permeability.the controlling guidance according to those statesThe input variables of BTP fuzzy controllerand causation. Thus the inference of the system iare the difference between predicted value of XBrpa multi- goal one. The back goal is the condition ofand its setting value， the trend and velocitythe next one. And the larger the goal grade， thechange. The fuzzy control rules were based on the .nearer the final goal. The inference process is notexperience of operator. For instance, if XBrp is ad-only the inference of each grade goal, but also a-vanced and has the forward trend, the strand speedmong them，that is to say, there is inference inwill be increased. These experiences were decross and wire direction[15]. The multi-goal infer-scribed by fuzzy sets and constituted the rules tableence mode is shown in Fig. 3.of BTP fuzzy controller.2.3 Diagnosis of abnormal conditions中国煤化工oF SYSTEMIn order to make the sintering process stableand smooth，the abnormities should be diagnosedfHCNMHGand eliminated in time. According to the produc-The general structure of system is shown intion experience, expert system of diagnosing andFig.4.eliminating abnormities was built. At first the sys-The data acquisition system attained appro-tem judged the, type of abnormal conditions by ana-priate information from the distributed control sys-lyzing theshep插g production data, then analyzedtem(DCS). The data gained were transferred to●534.Journal CSUT Vol. 12 No. 52005reasons or control guidance for some production，or the inferring conclusions are wrong，the systemNo.T iferenceshould seek help to the expert intuitively to modifyNo.l grade goalengineerNo.2 infercmce盆|the knowledge base. In order to raise the transpar-管, No.2 grade goatcngineer .ence of system，the explanation module explained2the inferring process of every subsystem.“Non interemceNo.n grade goalAPPLICATION OF SYSTEMThe system was developed based on object-Fig.3 Multi-goal inference modeloriented prototype development method with Visu-al C+十programming language[16l. The applica-Sintering processtion of this system shows that the hitting accuracyof sinter compositions and BTP prediction are moreDistributedthan 85%; the first-grade rates of sinter chemicalcontrol systemcomposition，stability rate of BTP and stabilityrate of sintering process are increased by 3%，9%Data acqusitionand 4%，respectively. It realizes optimum processcontrol.If this system is applied to 50% of sinter plantin China，whose annual capacity of sinter is aboutSinter.chemicallPrediction2.5X108 t, the sinter production will be increasedmoduleas follows: 2% X2.5X 10 X50%=2. 5X 10° t,BTPwhich is equivalent to the annual capacity of a 200 .Process state controlcontrolm2 sintering machine. At the same time, the ener-gy consumption and cost of each ton iron will beAbnormal conditionsdecreased.diagnosis5 CONCLUSIONS .Learing| Explanationl|Inference |K nowledgemodule| modulebase1) A real-time operation guide system for con-trolling sintering process has 3 parts， which areFig.4 General structure of system .sinter chemical composition control based on self-adaptive prediction， sintering process state controlthe blackboard.centred on permeability, and diagnosis of abnormalOn the basis of present production data，theconditions.predictive module modified the parameters of the2) The control strategies of sinter chemicalmodels to suit the change of sintering process, andcomposition control centred on basicity and processpredicted the sinter chemical composition andstate control centred on permeability are proposed,XpTp. The predictive information was transferredwhich overcome the difficulty for controlling sinte-to the chemical composition subsystem and thring process accurately. And predictions of chemi-process state control subsystem separately.:al compositions in sinter and XBTP are investiga-The chemical composition control subsystem,ted, which offers the solutions to long time delays.the process state control subsystem and the abnor-Interval optimization is put forward, which is themal condition diagnosis subsystem chose the suit-basement of steady control for sintering process.able knowledge base， respectively. According to3) The general structure of expert system forthe production data and knowledge, they chose thecontrolling the sintering process is studied. Thesuitable inference engine， judged the productionproc中国煤化工-izes the judgment ofstates, analyzed the reasons, and gave the qualita-iscreasons and con-tive and quantitative control guidance.trolYHCNMHGKnowledge base for any expert system cannotinclude all the knowledge required, and with theREFERENCESchange of sintering process， experiences andknowledge in the past may be inapplicable to the[1] Minoru W，Yutake S, Minorn s, et al. Developmentnew state? 5教据en the knowledge cannot give theof operation guide system and its application to ChibaFAN Xiao-hui, et al: Real- time operation guide system for sintering process with artificial intelligence .No. 4 sintering plant[ A].4th International Symposium. [10] JIANG Bo. Study of Operation Guide System andon Agglomeration[C]. Toronto, Canada, 1985. .Multilevel Fuzzy Integrated Judgment of Sintering[2] Unaki H, Miki K. Sakimura H, et al. New controlPermeability[D]. Changsha: School of Mineral Engi-system of sinter plants at Chiba works[A]. IFAC Au-neering， Central South University of Technology,tomation in Mining， Mineral and Metal Processing1999. (in Chinese)[C]. Tokyo, Japan, 1986.[11] Straka G. Process control model for sinter machine[3] Kouichi M，Naoki T，Kazuo N，et al. Modeling forspeed[J]. Metallurgical Plant and Technology Inter-control knowledge in sintering process using neuralnational， 1992， 8(5): 78 - 80.network and fuzzy inference[J]. Iron and Steel, 1992,[12] FAN Xiao-hui, HUANG Tian- zheng, XHEN Jin, et78<7): 1045 - 1052. (in Chinese)al. Expert system for sinter processing control (II )[4] Dawson P R. Recent developments in iron ore sinte-[J]. Journal of Central South University of Techno-ring: (Part 4) [J ]. Ironmaking and Steelmaking ，logy( Naturnal Science), 1998, 29(6): 535 - 537. (in1993, 20(2): 150 - 159.Chinese)[5] SHAO Jun-li, ZHANG Jing, WEI Chang-hua, et al.[13] ZHENG Yao dong, FAN Xiao-hui, CHEN Xu-ling,The Base of Artificial Intelligence[ M]. Bejing: Elec-et al. Establishing of expert system for sinteringtronics Industry Press, 2000. (in Chinese)process control based on database technology[J ].[6] CAI Zi- xing， XU Guang you. Artificial IntelligenceSintering and Pelleting, 2004，29(1): 1 -3. (in Chi-and its Application[ .M]. Beiing: Tsinghua Universitynese)Press, 1996. (in Chinese)[14] WU Xian-yi, QU Hong-tao, XIE Ying. Knowledge[7] SHEN Bing xin, FAN Xiao-hui, CHEN Xu-ling， etrepresentation and its structure of expert system .al. Adaptive prediction system of sinter chemistrydevelopment tool [J ]. Information Technology,based on artificial neural network[J ]. Sintering and2003， 27(5): 85- 88. (in Chinese)Pelleting， 2002, 27(5): 1- 3. (in Chinese)[15] CHEN De-yun. Study of visual inference of expert[8]FAN Xiao-hui， WANG Hai-dong， HUANG Jiao-system[J]. Computer Engineering and Application,zheng，et al. Expert system for sinter chemical compo-2000，36(7): 57- 60. (in Chinese)sition control centred on basicity[J ]. Sintering and[16] ZHU Dong-hai, LIU Liang-hua, DONG Hai-ning, etPelleting， 1997， 22(4): 1-3. (in Chinese)al. Thorough Explain of Visual C++ 6. 0[M]. Bei-[9] FAN Xiao-hui, W ANG Hai-dong. Mathematical Mod-jing: China Machine Press, 1999. (in Chinese)el and Artificial Intelligence of Sintering ProcessC M].(Edited by CHEN Wei-ping)Changsha: Central South University Press. 2002. (in .中国煤化工MYHCNMHG