European Control Conference (ECC) July 7-9,, Zürch, Swtzerland. Advanced control solutons to ncrease effcency a furnace combuston process S. M. Zanol*, D. Barches*, G. Astolf* and L. Barbon ** *D.I.I.G.A, Unverstà Poltecnca delle Marche, Ancona, Italy (s.zanol@unvpm.t, d.barches@d.unvpm.t, g.astolf@ @d.unvpm.t,) ** affdabltà automazon- I&C,ap raffnera d Ancona, Italy (l.barbon@gruppoap.com) Furnaces are key equpments and major energy consumers petrochemcal ndustry []: furnaces combuston performance drectly affects the qualty fnal products as well as the total energy consumpton the plant. In addton, a not optmal combuston may cause severe ncrease the temperaturee and the concentraton polluton components n the exhaust gasses. For these reasons control furnace combuston has becomes one the top research topcs n the refnery ndustry. The need to enhance the effcency level, to ncrease the prtablty as well as the commtment to meet envronmental requrements has motvated the nvestment n new sensor technologes and the applcatons advanced control logcs. In the present paper, the performances an exstng furnace combuston control system a refnery plant have been frst analyzed. The avalablty a relable oxygen analyzer for the measurement the percentage O n the exhaust gases at the chmney (see fgure ) and the adopton advanced PID control strateges [] allowed mprovng the typcal -fuel rato-control scheme that couples the Abstract In the present work the problem the furnaces combuston optmzaton n petrochemcal envronment s presented. In partcular, the paper s focused on the combuston effcency that drectly affects the operatng costss the plant. A prelmnary study the combuston process has been performed. A model the system has been obtaned by a black-box approach and lmtatons the exstng control archtecture have been analyzed. A new control archtecture, based on advanced PID control archtecture, coupled cross- and mplemented n a Dstrbuted Control System (DCS). The major benefts ntroduced by the new control system can be lmtng control logcs and Fuzzy logc has been developed found n ts relablty and n ts robustness to compensate the measurable dsturbances that affect the furnace. Moreover, the proposed control scheme has been proven to be effectve n the reducton the O content n the exhaust furnace gas as well as n the reducton the fuel consumpton. As a consequence the O reducton a reducton the exhaust gas temperature has been acheved thus further ncreasng the furnace effcency. The total effcency ncrease has been estmated about. wth a sgnfcant energy savng about 5 k /year. Fnally, the reducton ntrogen oxde and carbon monoxde concentratons n the exhaust gases acheved by the new control strategy, allows mnmzng the polluton emssons satsfyng the actual natonal envronmental requrements. I. INTRODUCTION flow and temperaturee combuston regulators the furnace. In partcular, the proposed double cross-lmtng control (DCL-C) strategy ss able to compensate for the large perturbatons the man combuston parameters that n the prevous control scheme were observed n presence varatons the charge feed from the upstream plants. Fnally,, to further ncrease system performances a Fuzzy controller has been developed that allows the fulfllment transent response requrements. The controller, followng a structure proposed by one the author n a prevous work [6], s based on the applcaton Fuzzy control technques to guarantee the suppresson or at least the lmtaton the overshoot n the system response. Fgure : Furnace system The proposed control system has been sutably mplemented n the Dstrbuted Control System (DCS) the plant allowng a great mprovement the control performance the orgnal system and a sgnfcant reducton the energy consumpton and envronmental polluton. The paper s organzed as follows. In Secton I a bref ntroducton the paper s gven. Problems related to combuston n furnaces are summarzed n Secton II. The mathematcal model adopted to descrbe the most mportant dynamcs the process used to develop and test the new controller s presented n Secton III. In Secton IV and V the new control strategy and the proposed Fuzzy controller are descrbed. In Secton VI the acheved results are dscussed. Fnally,, conclusons are presented n secton VII. 978--95-474-8/ EUCA 46
II. PROBLEM DEFINITION Combuston requres a fuel and an oxdant (typcally, oxygen that s present n the ); nsuffcent quantty oxygen n the combuston causes fuel resdues, resultng n ncomplete combuston wth soot. On the other hand, excessve causes problems such as a large amount exhaust gas and unnecessary heatng, resultng n degraded combuston effcency. Fgure shows the relatonshp between the -fuel rato and the thermal losses. The green (dotted) lne descrbes the theoretcal behavor the thermal loss wth respect to the -fuel rato whle the blue (dashed) curve represents the thermal loss due to ncomplete combuston. The red (sold) curve represents the resultng behavor n terms thermal effcency; three dfferent zones can be defned: for low values -fuel raton ncomplete combuston occurs and soot and smoke wth noxous partculate matter are generated (ncomplete combuston zone ) whle for hgh rato values thermal losses due to excessve O concentratons take place (Excessve zone); between the two an Optmum combuston zone s then defned where the maxmum values Thermal effcency are acheved []. THERMA LOSS Incomplete Combuston zone Relatonshp between Ar Fue Rato and Heat Effcency Optmum Combuston Zone AIR-FUEL RATIO Thermal loss due to ncomplete combuston Theoretcal Ar-Fuel Rato Thermal Effcency Excessve zone Fgure : Relatonshp between Ar-fuel Rato and Heat Effcency For combuston furnaces such as heatng furnaces and bolers n plants and factores, small-scale controllers such as sngle loop controllers are typcally employed to optmze the -fuel control rato and mprovng the combuston effcency. In large combuston furnaces, where dstrbuted control systems (DCS) and advanced control technques (multvarable predctve control, etc.) are used, typcally fuel rato and dumper valve (whch affectng furnace nternal pressure nfluences the furnace ventlaton) are controlled so as to prevent CO, CO and NOx (ntrogen oxde) from beng emtted. In ths way, the envronmental requrements are fulflled regardless possble thermal losses wth a consequent economc lost. In fgure the Vacuum Furnace control scheme that was prevously actve n the plant s presented. The drawbacks ths control scheme when employed for the regulaton the furnace combuston can be summarzed as:. O emtted wth the exhaust gasses s characterzed by sgnfcant oscllatons;. Not stable values the furnace output temperature;. Not stable values the Ar/fuel rato durng the transents. THERMAL EFFICIENCY PV check,66 AC O E.U. Fgure : Exstng control scheme Before the engneerng the new control logc, the exstng control system s analyzed n order to detect ts crtcal ponts. The followng man crtcal ponts have been ndvdualzed. Frstly, the regulaton the output temperature (TC) s realzed by the exstng logc through the control the fuel gas flow ( fuel). Hence, controlled varatons fuel gas are taken nto account by the flow controller ( ) guaranteeng an excess thus preventng the ncomplete combuston (see Fgure ). However no assurance optmal combuston effcency s gven. Another crtcal pont s the O regulaton n the smokes whch, n the exstng logc, s realzed through a controller (AC O) that checks the relatonshp -fuel. Ths control scheme allows avodng an excess fuel flow but vce versa t s possble to have an excess flow that nvolves economc lost. Fnally, no explct soluton accountng for the absorpton possble varatons the nlet load has been consdered. III. MODEL IDENTIFICATION The target the new control system strategy s to control the output temperature the furnace and to mnmze O resdue. Before ts actual mplementaton n the plant, the mprovements the new control archtecture have been tested flne on a smulaton envronment. At ths purpose t has been necessary to perform a multvarable model [4] dentfcaton the process under study. In ths way t was possble to desgn and test a new control archtecture safely performng all the necessary tunng the proposed controllers and estmatng the performance the process n closed loop operatons [5]. As t s well presented n lterature [6, 7] the dynamcs many ndustral processes can be well descrbed by a smple frst order transfer functon wth delay. In ths work the generc transfer functon has been chosen wth the followng structure: K Td () s g s e () s where as usual K s the process gan, s the process tme constant and T d s the process delay. From step tests performed on the furnace under study the model the TC fuel fuel gas Burner protecton PC PC 47
7 6 O on output smoke RMSEP (Valdaton) =.O Measure Predcton 86 84 Furnace Out Temperature RMSEP (Valdaton) =.5 C Measure Predcton O 5 4 C 8 8 78 4 5 6 7 8 Sample Fgure 4: O model valdaton process has been developed and smulated n the Matlab/Smulnk envronment. In fgures 4 and 5 the results the model valdaton phase the man controlled varables the process (percentage O and furnace output temperature) are shown. A MISO model for the predcton the O has been adopted. From the step test performed on the plant t results that the O can be well predcted from the Fuel flow, the Ar flows nto the furnace and the Dumper valve poston. The valdaton phase performed on a dfferent dataset has shown a good agreement wth the measurements wth an RMSEP (Root Mean Square Error Predcton) value. percentage oxygen as t can be seen n Fgure 4. Smlarly, the output temperature resulted to be sutably predcted by the Fuel and Ar flows nto the furnace and the Dumper valve poston. As t can be seen from the Fgure 5 the model has a good agreement wth the measurements: the computed RMSEP s.5 C. IV. NEW CONTROL STRATEGY In order to mprove the performance n the regulaton the furnace output temperature and n order to mnmze the percentage O n the waste gas, a new automatc control system [7] [8] [9] [] has been desgned. The new control archtecture has been prevously tested n an flne smulaton envronment,.e. MatLab/Smulnk. The furnace uses gas as fuel and as combustonsupportng tool. The value -fuel rato drectly affects the performance the combuston control system, the qualty gasol producton, and the energy consumpton. Good control the -fuel rato s the key to ensure full combuston, mprovng combuston qualty and control performance the combuston system []. Many factors have been recognzed causng large fluctuaton the optmal -fuel rato n the furnace system. For example, due to gas and pressure fluctuatons and lag the exstng control system, the actual -fuel rato may devates from the system settng values. In addton, fluctuatons the calorfc value mxed gas, whch may serously affects the furnace temperature as well as the thermal effcency furnace, are controlled by the temperature controller (TC, n Fgure ) causng a drect varaton the fuel flow ( fuel) and the flow as a 76 4 5 6 7 8 Sample Fgure 5: Out temperature model valdaton consequence the effects the /fuel rato control scheme. In the proposed new control scheme (see Fgure 6), a double cross-lmtng control (DCL-C) strategy s adopted [] wth the ntent to mprove the combuston effcency, to acheve accurate trackng and dynamc optmzaton /fuel rato. In the DCL-C system n order to prevent excessve surplus gas or n the secondary-loop controller, both gas and flow measurements are consdered and a mutual restrant s performed n case sudden load varatons. Through the DCL-C, the secondary-loop controller mposes, accordng to output the man loop, lower and upper bounds thus preventng, respectvely, ncomplete (lack oxygen) combuston and excessve O concentratons where the heat loss n the excess s larger than the heat provded by more effcent combuston. In the new control logc when the furnace output temperature changes, smultaneous ncrements or decrements the and gas flow are performed, thus achevng best control the furnace combuston. In fgure 6 the cross lmtng logc s hghlghted n red color; t s possble to note that an ncrease or decrease fuel or flow, drectly affects the O controller and the output temperature controller, respectvely. Addtonally, a feed forward logc (FF) [] s ntroduced to compensate nlet feed varatons and nlet temperature varatons as shown n the upper and rght sde fgure 6. The man contrbuton ths feedforward component s that t prevents sudden fluctuatons and perturbaton the combuston due to sudden varatons charge feed. Ths s partcular mportant because n many cases these varatons could otherwse lead to undesrable blocks the process. The feedforward acton has been opportunely nserted at the output the temperature controller, nstead that nsertng t drectly on the actuaton valve. In ths way, rapd varatons n the actuaton valve can be fltered. Fnally, a CO controller s nserted as a guard (see fgure 6) n order to lmt possble excessve decreasng the flow. No other organc compound has been montored snce t has been observed that the CO was the far most sensble varable to varaton O. Furthermore, due to the hgh performances the furnace at ssue the emsson NOx compounds are very lmted thus not representng crtcal element. 48
TC FUEL DEMAND AIR/FUEL SP + + FF Furnace nlet temperature +,66,5 PV check AC O C/D D C,95,95,5 E.U.,5,95,5,95,5,95 PV check,66 AC CO λ B A A/B fuel Burner protecton PC PC FT FT fuel fuel gas Fgure 6: The proposed control scheme In order to assure robustness and the handlng possble process falures and faults, sutable control logcs and data checkng and flterng [4] have been developed such as process varables checks, blockng the new logc n case analyzer fal, frozen values for handlng bad data and spkes detecton ; n order to avod change the constrans operated by the operators, guard controller are mplemented wthout SP-trackng. In case serous CO alarms, AC and TC master controllers are set n manual operaton mode. In fgure 7, the medan blocks are hghlghted: both and fuel gas control setponts, bound respectvely to temperature and fuel gas varatons, are ted to ther rates wth demand changes lmted to plus/mnus 5 for both components. In ths way the control system s able to accurately track correct /fuel ratos over a wde range operatng parameters. In case persstently fuel supply reductons, the mxture, untl the embeddedd overrde logc can assure to keep lmted the drawbacks feed reducton. Fgure 7: The medan logc (lght blue boxes) for the Ar/Fuel Rato control system Inputs to the DFS are: The set-pont error SE = z- zs, where z s the requred set pont and zs s the smoothed subsetwhere z s the actuall value the controlled varable; the pont generated by the DFS; the error E = zs-z, varaton the process varable s DY. In fgures 9 and performances the DFS are compared to a standard PID confguraton and wth a setpontt weghted PID (setpont =.7). In most the cases the adopton the DFS scheme allowed to elmnated the overshoot n the system response (as n fgure 9) or at least, nn the most crtcal cases, to the lmt t. Thss behavor s hghly desrable n the consdered plant snce ths could assure to reach the steady state response wthoutt trggerng oscllatons that could possbly destablze the system. V. FUZZY P PID CONTROL In order to obtan good performances from the pont vew the transent response behavor (lmted overshoot, fast rsng tme, short assessng tme wth lmted control efforts) dfferent control approaches have been evaluated [ 5]. In the followng ths secton the results the applcaton Fuzzy logc technques for the realzaton a sutable smoothng the controller set pont are dscussed. The adoptedd control scheme s based on a dscrete fuzzy smoother (DFS) (proposed n [6] by one the authors) that, workng on-lne, s able to adapt ts actons to the system behavor. The dea s to drve the system wth a sequence steps, so as to keep, at opportune stages, the nput sgnal small. In ths way, snce the absolute value the overshoot depends from the value the step change, t s possble to lmt the overshoot to proper small values. An advantage ths approach s that generally a shorter rsng tme can be acheved. Flow [m/h]..8.6.4. 5 Fgure 8: proposed control scheme Flow control Set Pont Dscrete Fuzzy Smother (DFS) Classcal PID Setpont wegthed PID 5 5 5 4 Tme [s] Fgure 9: Flow controll behavour wth dfferent control solutons 49
Control Effort 85 Furnace Output Temperature.8 Control Effort [m/h].6.4..8.6.4 Dscrete Fuzzy Smother (DFS) Classcal PID Setpont wegthed PID C 8,5 8 77,5 75 // // // 4// 5// 6// 7//. 5 5 5 5 4 Tme [s] Fgure : Control effort wth dfferent control solutons VI. RESULTS The mprovements n the regulaton the furnace output temperature obtaned by mplementng the proposed control logc n the DCS are shown n the followng fgures. As t can be seen comparng, n fgures and, the output error the old (here named exstng ) controller to that the new ( proposed ) controller, the control error has been decreased about 4. Fgure : Output temperature wth proposed control scheme 5 5-5 -4 - - - 4 5 Fgure : standard devaton the error n output temperature wth exstng control scheme 5 Output Furnace Temperature Control Errors Output Furnace Temperature Control Errors 85 8,5 Furnace Output Temperature 5 5 C 8 5 77,5 75 // 8// 5// // 9// 5/4/ /4/ 8// Fgure : Output temperature wth exstng control scheme Fgure shows the results obtaned by the new control system mplemented n the DCS. Comparng these results wth the prevously mplemented control logc shown n fgures,, the benefts n terms ncreased nose rejecton can be seen. Fgures and 4 show, respectvely the standard devaton the controlled output temperature for the exstng and proposed control scheme: comparng the two t s clear that the proposed control scheme allows for a more effectve regulaton the furnace output temperature. In agreement, the control error n the regulaton the percentage O n the waste gas has been largely decreased obtanng a reducton up to about 75 (compare fgures 5 and 6). Smlarly, the standard devatons the controlled percentage O the two control schemes are compared n fgure 7 and 8. Is t clear from the above results that the proposed control scheme allows to obtan hgher economc return; n fact, when adoptng the new control strategy, the O workng pont s kept closer to the mnmum target. Fnally, fgure 9 shows the decrease O emsson as a result the mplementaton the new control scheme. -5-4 - - - 4 5 Fgure 4: standard devaton the error n output temperature wth proposed control scheme 7 6 5 4 // 8// 5// // 9// 5/4/ /4/ 8/4/ Fgure 5: Percentage oxygen on smoker wth exstng control scheme.5.5 O on output smoke O on output smoke.5 // // // 4// 5// 6// 7// Fgure 6: Percentage oxygen on smoker wth the proposed control scheme 4
Percentage O Control Errors 5 O n output smoke 4.5 4 8.5 6 4.5.5 -.8 -.6 -.4 -... 4.6.8 / 9/ /9/ 5/9/ 7/9/ 9/9/ / // Fgure 7: standard dev. the error n O wth exstng control scheme Fgure 9: Decreased percentage O n output smoke obtaned wth the proposed control scheme 4 Percentage O Control Errors 8 6 4 -.8 -.6 -.4 -... 4.6.8 Fgure 8: stand. dev. the error n O wth the proposed control scheme Fgure : DCS snap-shoot hghlghts O reducton on real system In fgure a snapshot drectly taken from the DCS HMI s depcted hghlghtng the same results fgure 9. VII. CO NCLUSIONS An advanced PID controller archtecture desgn that optmzes furnace combuston s presented. Process model dentfcaton has been performed and a new control soluton has been proposed, analyzed and tested n smulaton. As a fnal step the proposed control system has been sutably translated for the mplementaton n the Dstrbuted Control System used n the refnery plant and t s actually operatve. The mplementaton the proposed system confrmed the expected hgh performances mprovements: better stochometrc combuston constrants management; safer transt management; tghter control output temperature and heater excess (O level optmzaton, CO guard controller). The proposed Double Cross-Lmtng control (DCL-C) senstve to smultaneous varatons the gas flow, lmtng effectvely the actual value the excess rate, ensurng the combuston control system to work at optmum condtons ( that s, wthn the optmum combuston zone), meanwhle, avodng burnng gasol at excessve condtons and ensurng good qualty gasol products. After the Double Cross-Lmtng control logc and the DFS set pont controller have been ntroduced nto the furnace combuston control system, the accuracy the gas and flow control has been mproved whle the heat losses caused by excess combuston and lackng-oxygen combuston have been consderably reduced. Thus, the system has been proven to be useful n helpng to mprove product qualty and product yeld as well as reducng energy consumpton. The economc can make the loop and the gas loop gann oxygen optmzaton has been estmated to be up to 5 k /year and even greater. REFERENCES [] Speght J.G., The chemstry and technology petroleum. CRC Press, 6. [] Maples R.E., Petroleum refnng technology and economcs. nd Edton. Pennewell Books,. [] Shell, Practce Worth Replcatng for Automatc Excess Oxygen Control for Natural Draft Fred Heaters. GS.7.547 - ECCN EAR 99 October 9 [4] Coughanowr D.R., Process system analyss and control. nd Edton, McGraw Hll, New York, 99. [5] Seborg D.E., Edgar T.F., Mellchamp D.A., Process dynamcs and control. John Wley & Sons, New York, 989. [6] Zhuu Y.C., Backx T., Identfcatonn multvarable ndustral processes for smulaton, dagnoss and control. Sprnger, London, 99. [7] Shnskey F.G., Process Control Systems. Mc Graw Hll, New York, 996. [8] Anrudda D., Mng-Zu Ho, Shankar P. B., Structure and synthess PIDD controllers, Sprnger,. [9] Luyben W.L., Luyben M.L., Essentals process control. McGraw Hll, New York, 997. [] Stephanopoulus G., Chemcal process control. Prentce Hall Inda, Neww Delh, 995. [] Shell, Control System And Instrumented Protectve Functons For Fred Equpment: Mult Burner Control. DEP.4..7 Jan. 998 [] Skogestad S., Postlethwate I., Multvarable feedback control: analyss and desgn, Wley, 5. [] Astrom K.J., Hagglung T., PID controllers: theory, desgn, and tunng, ISA, Researchh Trangle Par, 995. [4] Morar M., Zafrou E., Robust process control. Prentce Hall, New York, 989. [5] Tann Kok Kong, Wang Qung-Guo, Hang Chang Chen. Advances n PIDD control, Sprnger, 999. [6] Zanol S.M., Conte G., Remotely operated vehcle depth control. Control Engneerng Practce,. [7] Jan Maran Macejowsk, Predctve Control wth Constrant. Practce Hall, 4