Cement, Energy and Environment July-Sep 2002

energy conservation eq uations and a large set of nonlinear algebraic equations. The dependent va riables are the calci um carbonate mass fraction in the so Iid phase, the carbon dioxide mass fraction in the gas phase, and the temperatures of the so li d an d gas phases. The dependent variables also include the methane, ethane, water and oxygen mass fractions in the gas phase. The limestone fee d composition was taken to be 98.6% ca lc ium carbonate. The data listed typical va lues for rotary lime kilns of the type found in loca l indu stri es tn Brazil. Simulation Studies The mode l developed was verified in Part I of this study by comparing the simulation results wi th th e experi mental data presented for a pilot rotary kiln by Watkinson and Brimacombe. The aim of the simulation studies in this section is to include a validated model for determination of flame size in that model in orde r to eva luate th e effect of certa in va ri ables on the ge ne ra l performance of an ind ustrial rotary ki ln. Simulation studies into the control variables for kiln operation The four control variables that affect the overall performance of a rotary k iln are th e limes tone feed rate, the speed of rotation, the fuel feedrate and the flame length. By controlling these variables the kiIn operator is able to carry out sta ble lime production. Clearly, there are many other variables that affect the calcina tion proce ss. These incl ude the chemi ca l composition and the granulometric distribution of the limestone feed. and the composition and calori fic value of the fuel. These variables are usually not controll able. The stability of the production process m is heavily dependent on the·correct cho ice of control variables . The flame length is adjusted by controlling the rati o of primary and secondary airflow. A study of the iso lated effects of the co ntrol variables on the performance of the process was carried out in this work. A steady-s tate one- dimensional model was developed to describe li mestone ca lcination in a rotary ki ln. The mode l provides a description of th e ca lc inat ion reaction kinetics, the heat and mass transfer phenomena between the soli d bed and the gas phase, and the rheo logy of the bed. Siinulation results permitted a detailed evaluation of the effects of changes in the operational variables on the performance of the rotary kil n and on the quali ty of the end product. The resu lts showed good agreement with experimental data reported in the literature. As the mode l was also val idated by the experimental results measured in a pilot kiln , the model was used to simul ate variations in the most important control variables of an industri a l rotary k il n. The simul ati on re sults prov ided information which can be used in expert systems, in the des ign of rotary ki lns and also in the analysis of operational condi-tions used in limestone calcination processes. Courtesy: ZKG International No.5 /2002, Pp 7 .J-83. Fax: +-19(0} 6/23700/22 £ mail: ::kg@Bmtl'erlag de. MODEL PREDICTIVE CONTROL Greg Martin and Steve Me Garel Expeti system solutions for the cement industry have been around for 20 years but have not had unqualified acceptance. Cost versus benefit, systems fallen into disuse, the question or viabi lity for mills, and the inabi lity to deal with process drift are generall y ci ted as the reasons for widespread skepticism. Aut omation usi ng the powerfu l technique of mod e l predictive control (MPC) has been used for the past 25 years in the oi l re finin g, chemi ca ls, po lymers, power, and food industries, with more th an 5,000 in s ta ll at ions worldwide. In late 1999, it was implemented on a kiln/cooler and a product ball mill , and both ins tallat ions were accepted as superior to the ir traditional expert system equivalents. In late 2000, a MPC app lication was implemented on a ve rti cal mi ll with simil a r results; and in early 200 I, another was implemented on a finishing mill producing several grades of cement dai ly. MPC represents the process dynamics by step responses. Each s tep re sponse is that of one con trolled variab le (CV) to a postt tve unit step tn one manipulated variable (MV). These responses are determined by plant tests or engineered. The length of the model is such that the response settles at the new steady-state value. This mathematical process is analogous to matrix inversion, as ind icated by th e matr ix S. A properly imp lemen ted MPC application makes moves similar to an experienced operator. This is, when a new setpoint is entered, one or two large moves come out, and then the controll er waits for the process to respond, poss ibly seve ral control intervals, before a smaller corrective move is made. Practically, MPC is the dual of an expe rt system. Almost every aspect is based on a diffe rent principle. More specifically: • MPC uses a model of the process; an expet system uses a model of the operator.