Chemical Plant Flare Minimization via Plantwide Dynamic Simulation

Ind. Eng. Chem. Res. 2009, 48, 3505–3512

3505

Chemical Plant Flare Minimization via Plantwide Dynamic Simulation Qiang Xu,* Xiongtao Yang, Chaowei Liu, Kuyen Li, Helen H. Lou, and John L. Gossage Department of Chemical Engineering, Lamar UniVersity, Beaumont, Texas 77710

Flaring is crucial to chemical plant safety. However, excessive flaring, especially the intensive flaring during the chemical plant start-up operation, emits huge amounts of volatile organic compounds (VOCs) and highly reactive VOCs, which meanwhile results in tremendous industrial material and energy loss. Thus, the flare emission should be minimized if at all possible. This paper presents a general methodology on flare minimization for chemical plant start-up operations via plantwide dynamic simulation. The methodology starts with setup and validation of plantwide steady-state and dynamic simulation models. The validated dynamic model is then systematically transformed to the initial state of start-up and thereafter virtually run to check the plant start-up procedures. Any infeasible or risky scenarios will be fed back to plant engineers for operation improvement. The plantwide dynamic simulation provides an insight into process dynamic behaviors, which is crucial for the plant to minimize the flaring while maintaining operational feasibility and safety. The efficacy of the developed methodology has been demonstrated by a real start-up test. Introduction g r 9 o . 1 2 s c 6 a 1 . s 0 b 8 e u i p / / 1 / : p 2 t t 0 . h 1 0 | 1 9 i : 0 o 0 d 2 | , 7 9 r e 0 b 0 2 o , t c 7 O 2 r n y a o u r V I b e N F U ) : R b A e W M ( A t e L a y D b n d o e i t d a a c o i l l n b w u o P D

Chemical Chemic al plan plantt star start-up t-up oper operatio ations ns can be cons consider idered ed as plant pla ntwi wide de dyn dynam amic ic ope opera ratio tions, ns, by wh which ich the wh whole ole pla plant nt operating status is transferred from one steady state to another. Apparently, the start-up operation is a highly nonlinear, complex operation that usually involves discontinuous and/or parallel operat ope ratin ing g pr proce ocedur dures, es, as we well ll as a wi wide de ch chang angee of ma many ny controllers’ set points. Such complex operations would inevitably generate huge amounts of off-spec product streams that have to be sent to flaring systems for destruction, if the plant start-up could not be completed promptly. Flaring can protect chemical plant personnel and equipments, as well as protect the local environment from direct emission pollutions. Thus, flaring is crucial to the chemical process industry (CPI). Excessive flaring, however, will also cause negative environmental ronment al and social impacts and result in tremendous material and energy losses.1 The flaring emissions during chemical plant start-up operations generate huge amounts of CO, CO2, NO x, volatile vola tile org organic anic com compoun pounds ds (VO (VOCs) Cs),, high highly ly reac reactive tive VOC (HRVOCs) (defined in Texas air quality regulation as ethylene, propylene,, isomers of butene, and 1,3-buta propylene 1,3-butadiene), diene), and partiall partially y oxygenated hydrocarbons (such as formaldehyde). It has been estimated that an ethylene plant with a capacity of 1.2 billion pounds of ethylene production per year will easily flare about 5.0 million pounds of ethylene during one single start-up.2 Given the 98% flaring efficiency (destruction efficiency), the resultant air emission will include at least 40.0 klb CO, 7.5 klb NO x, 15.1 klb hydrocarbon, and 100.0 klb HRVOCs. The flaring emission will cause highly localized and transient air pollution events, which are harmful to people’s health. For instance, the industrial flare emission of HRVOCs mixed with NO x has been identified identified wit with h high concentrati concentrations ons of ozon ozonee observed in the Houston/ Houston/Galveston Galveston area of Texas, which violates the Nat Nationa ionall Amb Ambient ient Air Qual Quality ity Sta Standar ndards ds (NA (NAAQS AQS)) for 3-7 ozone. Note No te tha thatt the fla flari ring ng em emiss ission ion not on only ly ca cause usess dangerouss environm dangerou environmental ental pollution but also results in tremend tremendous ous raw material and energy loss that could generate much needed products from the industry. As a result of the increas increasingly ingly strict environm envi ronmenta entall regu regulat lations ions and econ economi omicc comp competit etition, ion, flar flaree * To whom correspondence should be addressed. Telephone: 409880-7818. Fax: 409-880-2297. E-mail: [email protected].

minimization has become one of the major concerns for the chemical process industry. Current practice of flare minimization in CPI plants is not standardized standardi zed due to the complex complexity ity of the different process and operating procedures. This causes flare minimization to depend almost exclusively exclusively on the experienc experienced ed and well trained operators, 8,9 engineers,, and adminis engineers administrators. trators. Th Thee foc focal al poi point nt fo forr fla flare re minimization is to reduce the number of instances when the plant has to flare and the quantity of the materials to be flared. However, the industrial-experience based methods are often limited, when they have to confront complex plantwide dynamic operations (e.g., start-up) with critical control and safety issues.10 Consequently, virtual models are employed to study the startup behaviors. Some studies have tried steady-state simulation (SS) for start-up operation, based on a set of predicted steadystate operating points to project the system dynamic response.11 The methods have inherent deficiency because they could not reveal the real dynamic behaviors between two adjacent steady states and thus lack the capability to guide critical process control and operation.12 To provide more accurate dynamic information, plantwide dynamic simulation (DS) methods have recently become popular.13 It is use used d to vi virtu rtual ally ly tes testt pla plant nt sta start rt-up -up ope operat ration ionss according to the operating strategies that will be undertak undertaken en by the plant operating personnel.14-17 It examines critically the potential process operational risks and infeasib infeasibilities. ilities. Based on the DS results, the feedback will help the plant improve the start-up operating strategies and thus reduce flare emissions. The methodology is cost-effective and proved very successful in real application. Based on the previous studies, this paper for the first time generalizes a systematic methodology for flare minimization during CPI plant start-up operations via plantwide dynamic simulation. It covers the modeling of recent practice for start-up operations with total recycles. Meanwhile, modeling experiencee and required industrial experienc industrial data are also summar summarized. ized. A field test for flare minimization during an ethylene plant startup is present presented ed to demonst demonstrate rate the efficacy of the method methodology. ology. Scope of the Dynamic Simulation Model

Thee am Th amoun ountt of fla flarin ring g em emiss ission ion dur durin ing g a pl plant ant sta start rt-up -up normally increases with the start-up duration. To shorten the

10.1021/ie8016219 CCC: $40.75  2009 American Chemical Society Published on Web 02/27/2009

3506

g r 9 o . 1 2 s c 6 a 1 . s 0 b 8 e u i p / / 1 / : p 2 t t 0 . h 1 0 | 1 9 i : 0 o 0 d 2 | , 7 9 r e 0 b 0 2 o , t c 7 O 2 r n y a o u r V I b e N F U ) : R b A e W M ( A t e L a y D b n d o e i t d a a c o i l l n b w u o P D

Ind. Eng. Chem. Res., Vol. 48, No. 7, 2009

Figure 2. General methodology framework. Figure 1. General sketches of normal operation and the initial status for start-ups.

start-up time, the important plant units are usually commissioned commissioned at some designated steady states before initiating the start-up, whic wh ich h is cal calle led d the ini initia tiall sta state te for sta startrt-up up.. Fo Forr ins instan tance, ce, distillation columns need to run with total reflux, and reactors need to be controlled at certain temperature temperature levels. Recent flare minimization minimi zation practices have shown the total-re total-recycle cycle start-up may significantly reduce the flare emission. This suggests that the plant should start with full-recycled streams from downstream process as system input, which can capture, recycle, and reuse large amounts of off-spec products that would otherwise be sent to the flare system as waste streams. To model the total-recycle total-recycle start-up without losing the generality, the proposed methodology willl addr wil address ess the dyna dynamic mic sim simulat ulation ion of the proc process ess syst system em defined in Figure 1. For comparison, the plantwide dynamic simulation model at normal operations is shown in Figure 1a, where normal feed means the raw material input at normal operating conditions. Note that Figure 1a includes all the internal recycles at normal working conditions. Figure 1b shows the initial state of the dynamic simulation simulation model for start-ups. Based on a general case, the DS model contains all the designated outer and inner loops of recycled streams, which are only used for plant start-up. Here, a recycled stream is called an outer-loop stream because it recycles the off-spec materials to the very beginning of the system input; otherwise, otherwise, it is called an inner-lo inner-loop op stream. With the help of Figure 1, the DS model for a plant start-up can be defined as simulating the transient behaviors from the initial condition of start-up (Figure 1b) to the normal working condition (Figure 1a). Note that the topology of the simulation model

changes when full-recycle start-ups are considered, which makes the simulation task extremely complex. General Methodology for Plantwide Dynamic Simulation

To conduct the plantwid plantwidee dynamic simulation for total-re total-recycle cycle start-ups, a systematic modeling methodology has been developed in this paper, which considers more general and complex situati situ ations ons than prev previous ious met methods hods.. The plan plantwi twide de dyna dynamic mic simulation is performed based on the integration of rigorous process models, plant design data, P&ID, DCS historian, and industrial expertise. Generally, the developed methodology is the integration of modeling activities among three interactive stages as shown in Figure 2. Steady-State Modeling and Validation. In the first stage, model development starts with the setup of the SS model for the plant process system that needs to be investigated during the start-up. The sequential modular approach is used in this methodology, by which each subsystem will be modeled and solved solv ed inde independ pendentl ently. y. The reas reasons ons for sele selectin cting g sequ sequent ential ial modular approach instead of the equation oriented approach are mainly because the whole plantwide simulation task is more convenient to be decomposed into small subsystems for validation and troubleshooting. Although the modeling process will be sl slow ower, er, it doe doess not influenc influencee too much as the startstart-up up simulation task is conducted offline. The developed SS model is usually validated by plant design data first, which are collected from plant design documents. Then the model will be further validated by normal steady-state operating conditions, where DCS DC S (di (distr stribu ibuted ted co contr ntrol ol sys system tem)) his histor torian ian wi will ll be use used. d. Sometimes, the real plant data are not in mass or energy balance due to unpredicted reasons such as sensor drifting, malfunction, or the negligence of some input/output streams. Under such



3507

start-ups. Figure 3. General model for plant start-ups.

conditions, data verification, data reconciliation, and the support from industrial expertise will be extensively involved. Typical model mod el tur turning ning parameter parameterss in this stage incl include ude colu column mn tra tray y efficiency, efficien cy, heat transfer efficiency, reaction kinetic parameters, etc. Dynamic Modeling and Validation. If the plantwide SS is satisfied, the validated SS model is then transferred to the DS model. Three types of informa information tion must be involved to support such a model transition. transition. First, equipment dimension data (such as initial equipment hold-up level, vessel type and geometry, hydrau hyd rauli lics, cs, an and d tra tray y geo geome metr try y and de detai tails) ls) fro from m upd updat ated ed mechanical drawings is required, which provides the process capacity information. Second, control strategy and controller parameters from current plant P&ID (piping and instrument diagram) are needed, which provides process control information. Third, process and equipment heat-transfer methods should be given for the dynamic simulation simulation model, which provides the thermodynamic information. When the plantwide DS model has been built, the plant DCS historian of some process upset scenarios will be used again for further model validation. This is a nontrivial task, because lotss of dat lot dataa pre prepro proces cessi sing ng and pos postp tproc rocess essin ing, g, as we well ll as modeling troubleshooting activities, are induced. Meanwhile, it is also unrealistic to expect the DS results will exactly match the real measurements even if all possible modeling efforts have been bee n ma made. de. Th Thee suc succes cesss of th thee DS val valida idati tion on dep depend endss on whet wh ether her or not th thee tim timing ing and am ampl plitu itude de of th thee dyn dynam amic ic responses of each subsystem will match the real DCS historian. Sometimes, the plant may lack some transient concentration, tempera tem perature ture,, or pres pressur suree info informa rmatio tion, n, whi which ch real really ly prov provides ides difficulties difficul ties for DS model validation. validation. Therefore, it is always better to collaborate with experienced plant engineers to judge the validation results, so as to improve the modeling quality. Note that this is the last chance to validate the plantwide DS model. Identification of Start-up Initial Status. Before the validated plantwide DS model is applied for start-up simulation, the whole model should be adjusted to run at the initial state of start-up. Thiss step actu Thi actuall ally y high highligh lights ts the mos mostt imp importa ortant nt diff differen erence ce between normal DS and start-up DS. The initial state of startup usually involves low-load running equipment, zero inflow and outflow rate, and full reflux of distillation columns, columns, as well as tem tempora porary ry mul multipl tiplee recy recycle cless and auxi auxilia liary ry str streams eams at the

beginning of start-up. Meanwhile, the transfer procedure from the nor norma mall st statu atuss to it itss ini initia tiall st start art-up -up st statu atuss oft oftens ens lac lacks ks supporti supp orting ng inf informa ormation tion,, such as real plan plantt data data,, para paramet meter er estimati esti mations, ons, and oper operatin ating g guid guidance ance.. Thu Thus, s, the acti activiti vities es of model status adjustment present the most challenging step for plantwide start-up simulation, where sufficient care is needed to prevent the divergence of the modeling process. Figure 3 shows a general model system to illustrate the transformation, where the initial feed for start-up represents the feed used for commissioning the system to its initial state for start-up. Note that the transition of model status requires not only system model input changes but also process topology changes (recycle and auxiliary streams) and operating status changes (e.g., temperature, pressure, concentration, and control parameters changes). A general algorithm to accomplish the status-adjusting task, which was completely ignored in previous studies, is presented below. Step 1. Make the plantwide DS model run at the normal steady state. At this step, the flow rates of the entire inner and outer loop streams for start-up are zero. The normal feed (stream A in Figure 3) will be exactly equal to system input (stream C in Figure 3), while the flow rates of initial feed for start-up (stream B in Figure 3), system recycle (stream D in Figure 3), total recycle (stream E in Figure 3), and purge (stream F in Figure 3) are all zero. Step St ep 2. Ac Acco cordi rding ng to the pla plant nt sta start rt-up -up pr proce ocedur dure, e, add additional dynamic models into the plantwide DS model. The additionall models include mixers, splitters, additiona splitters, and outer and inner loop streams (heat exchanger models may also be needed). Step 3. Identify the operation conditions for the key units at the initial status for start-up, such as the composition of stream B, controller set points, reflux ratios, and heating and cooling duties. Step 4. Gradually reduce the flow rate of A and meanwhile gradually increase the flow rate of B, which will make the system input of C gradually transform from A to B. During the input transform, the flow rates of the entire inner and outer loop streams for start-up should be gradually increased from zero to the designated value. The controller set points for all the key units should be gradually adjusted accordingly. At the end of this step, the flow rate of A will be zero; the flow rate of B will

3508


Table 1. Major Information for Plantwide Plantwide Dynamic Simulation

modeling level

DS validation

targets

model capability check set point identification

model capability check

major supporting data from plant

plant PFD and P&ID plan pl antt de desi sig gn da data ta nomi no mina nall op oper erat atin ing g da data ta indu in dust stri rial al ex expe pert rtis isee Aspen Plus Hysis Pro II 1

equipment capacity cont co ntro roll st stra rate teg gy an and d pa para rame mete ters rs proc pr oces esss up upse sett da data ta indu in dust stri rial al ex expe pert rtis isee Aspen Dynamics Dynsim Hysis 2 3

typical simulators

normalized time expense


SS validation

reach the desired value, which will make the system run close to the initial status for start-up. Step 5. Gradually increase the splitter ratio of D and make its flow rate increase. Meanwhile gradually decrease the flow rate of B. At the end of this step, the flow rates of B and purge stream F will be zero. Thus, the total recycle flow rate of E will exactly equal the system recycle flow rate of D, which also equals the system input flow rate of C. Step 6. Check the entire DS model to see if every unit is running at the expected initial state of start-up. If yes, the model system has reached the initial state of start-up. Otherwise, the model system needs further fine-tuning. fine-tuning. First, make sure all the controller set points are fixed at the specified values in step 2. Then increase the flow rate of B from zero to some small value, and meanwhile reduce the flow rate of D by the same value by reducing its split ratio, and go to step 5. Note that in steps 4 and 5, the system input transformation needs sufficient care to avoid computational divergence, divergence, because such transformation influences the dynamic simulation result of every unit of the model system. Except the small changes at every ever y inpu inputt adju adjustme stment, nt, the mos mostt imp importa ortant nt is to make the plantwide system reach the steady state at some intermediate transformation transfor mation points (e.g., 50%, 70%, and 90% completi completion on of full recycles), and then continue the model tuning. Thus, the system input transformation in steps 4 and 5 actually suggests the acquisition of a series of steady states. Because of that, the identification of start-up initial status is considered as the most critical and time-consuming task during the dynamic simulation. Also note that this general algorithm is a hypothetical modeling procedur proc eduree wit without hout any real imp implem lementa entation tion ass assumpt umption. ion. It prese pre sents nts a sys system temat atic ic wa way y to obt obtai ain n th thee in initi itial al sta state te of the plantwide DS model for the start-up simulation. After er the Dynamic Dyna mic Simul Simulation ation for Flar Flaree Minim Minimizati ization. on. Aft initial state of a start-up is obtained, the DS model is ready to virtually test the start-up operating procedures provided by the plant. The final modeling activity is to schedule all the dynamic operating procedures as the input of the DS model and “let it run”. Note that a chemical plant start-up undergoes a number of st stage ages. s. At eac each h sta start rt-up -up st stage age,, sp speci ecific fic obj object ective ivess an and d constraints must be obtained, which usually involves process inputs, temperature, pressure, flow rate, and stream components of each unit changing nonlinearly.18-20 Thus the controller set points and even control strategy should be changed accordingly. To model the dynamic process inputs and dynamic start-up operation operati on procedure, a set of program scripts, scripts, or so-call so-called ed startup “tasks”, have to be developed and embedded into the DS.21 Then, the DS model is ready to run, while the simulation results will be fully repeatable. Suppose Sup posedly, dly, the dyna dynamic mic sim simulat ulation ion resu results lts wil willl iden identify tify unexpected or unsafe operation conditions. This will be fed back to the plant operation group to recheck their previous operating operating procedures. The modified operating procedures will be virtually

∼

DS for start-up operation feasibility test safety check start-up procedure improvement turnaround operating data indu in dust stri rial al ex expe pert rtis isee

Aspen Dynamics Dynsim Hysis 3 4 ∼

tested again by rerunning the DS model with modified input. Similarly and iteratively, the DS model will help the plant identify viable or even optimal operating operating strategies for its startup operations. Note that when simulation meets an infeasible problem,, two possibi problem possibilities lities may exist: either the model itself has some problems, problems, which are not detected detected in prev previous ious model validation, or the plant operating procedures do have uncontrollable operations. Thus, troubleshooting troubleshooting should be conducted with the help of both theoretical analysis and industrial expertise. Also note that all the developed methodology is based on firstprinciples models. Therefore, commercial simulators such as Aspen Plus, Aspen Dynamics, PRO/II, Dynsim, or Hysys are recomme reco mmended nded.. The deve develope loped d meth methodol odology ogy is appl applicab icable le in general and is not limited to one particular plant or process. It is also a cost-effective approach for flare minimization. Table Ta ble 1 giv gives es a su summ mmary ary abo about ut th thee ma major jor sup suppor porti ting ng information needed from a real plant. It is worth noting that although alth ough the foc focal al poin pointt of the developed developed met methodo hodology logy is dynam dyn amic ic DS DS,, SS is ab absol solute utely ly nec necess essary ary.. Th This is is no nott onl only y because DS models are generated from the SS models but also because SS can identify the controller set point information, which whic h dete determi rmines nes the sett settling ling point of a dyna dynamic mic response. response. There Th erefor fore, e, th thee con contro trolle llerr se sett poi point ntss at the en end d of sta start rt-up -up operat ope ration ionss sh shoul ould d be ide identi ntifie fied d thr throug ough h SS bef befor oree the DS validation and application. Duee to th Du thee inh inhere erent nt com compl plexi exity ty of pla plantw ntwide ide dy dynam namic ic simulation simula tion and the possible data incom incompletene pleteness, ss, on-site industrial expertise is required at every stage. It helps validate the simulation result and facilitate troubleshooting. Two groups, operation group and simulation group, usually work together for the DS based flare minimization project. The operation group from the plant includes experienced operators and engineers. They prov provide ide alte alternat rnative ive sta start-u rt-up p proc procedur edures es base based d on thei theirr planning and experience. Then the simulation group virtually tests the proposed procedures with the plantwide DS model to check the operational feasibility, reliability, and safety issues. The plantwide DS model will be validated or modified based on the simulation results and joint discussions. The time cost generally increases as the modeling activity goes to the next stage. Based on the experience, if time expense for steady-s steady-state tate modeling is normalized as one unit, dynamic modeling activity will cost two to three units. The final start-up simulation for flare minimization will take three to four units, which is because considerable troubleshooting and improvement efforts will be involved at this stage. Case Study

Ethylene plant start-ups emit huge amounts of VOCs and HRVOCs that may cause highly localized and transient air pollution pollutio n events and also result in tremendous raw material and energy loss. Thus, one flare minimization project based on the


3509

Figure 4. Sketch of the entire modeling system for the case study.


developed methodology has been conducted. A general ethylene plant starts with thermal cracking of raw feedstock in furnaces. Thee ef Th efflu fluen entt cr crac acke ked d ch char arge ge ga gass is co cool oled ed an and d pa part rtia iall lly y condensed and then sent to a charge gas compression system. Charge gas from the final stage compression is dried prior to the chilling train. After the majority of hydrogen is removed from the chilling train, the cold charge gas is then fed to the fractionation section, which consists of a demethanizer (DeC1), deethani deet hanizer zer (De (DeC2), C2), depr depropan opanize izerr (De (DeC3) C3),, and debu debutani tanizer zer (DeC (D eC4) 4).. In the de deme metha thaniz nizer, er, me metha thane ne an and d hyd hydro rogen gen ar aree separated from the charge gas mixture as overhead vapor. The bottom stream with the enriched mixture of C2+ (ethylene, ethane, propane, and heavier hydrocarbons) is then sent to the recovery section for further separation. The recovery section employs a deethanizer, depropanizer, and debutanizer to separate the products of C2s, C3s, C4s, and C5s plus heavier components, respec res pecti tivel vely. y. Al Alll th thee pro produc ducts ts hav havee to me meet et the pr produ oduct ct specifications or purity at normal operation conditions. To reduce flare emission, the start-up with total recycle has been implemented. Figure 4 presents a sketch of the entire modeling system for the case study, where some subsystems are simplified for easy illustration. illustration. It should be noted that since the furnace operations are not within the total-re total-recycled cycled start-up process, the charge gas flow rate and component concentrations from furnaces only functioned as the simulation inputs. Meanwhile, because the charge gas concentrations from a specific furnace are fixed (for either heavy or light naphtha cracking), in th this is ca case se st stud udy y on only ly th thee fu furn rnac acee flo flow w ra rate te is ch chan ange ged d dynamically according to increased number of activated furnaces. nac es. Th Thus, us, the fu furna rnace ce si simu mulat lation ion is unn unnece ecessa ssary ry to be dynamic. To reduce the computational load without the sacrifice of sim simulat ulation ion accu accuracy racy,, the furn furnace ace sim simulat ulation ion is cond conducte ucted d independently independe ntly and is not included in the dynamic simulation simulation of the case study. Note that four major outer recycles are considered during start-up, which include the recycles of H2 from the chilling train: H2 and methane from the top of DeC1, C2s from the top of DeC2, and C3s from the top of DeC3. A major inner recycle is the stream of C4s from the top of DeC4 fed back to the bottom of De DeC2 C2.. Wi With th th these ese re recyc cycles les,, the sta start rt-up -up em emiss issio ions ns are expected to be greatly reduced due to two reasons: one is that the huge amounts of components from H2 though C4 will be reused instead of being flared during start-up; the other is that with the help of these recycled streams, the key units of DeC1

Depropanizer anizer temperature temperature profile at normal steady state. Figure 5. Deprop

through DeC4 can gear toward their normal operation conditions quickly, which means the start-up flaring time will be reduced compared with the start-up without recycles. This daring start-u start-up p indeed presents considerable challenges for mod modelin eling g and dyna dynamic mic simulation simulation.. With the deve develope loped d methodology, methodolo gy, the plantwid plantwidee dynamic simulation has successfu successfully lly helped the plant validate the conceptual start-up procedure. In accordance with the developed methodology, the simulation results at different modeling stages are shown below. Model Development and Validation. After the SS model has been developed, model validation is needed. If the SS test is not satisfied, it needs to be calibrated with plant data by adjusting different model parameters to match the steady-state model results with plant data measurements. For instance, Figure 5 shows the comparison between SS model prediction and real measurements for DeC3 tray temperature profile, which are key factors for evaluatin evaluating g separatio separation n performa performance nce of these distillation columns. It is clear that the simulation results match well with the real measurements. As examples of dynamic model validation, Figures 6-8 show the flow rate dynamic response for DeC2, DeC3, and DeC4 product streams under a recorded disturbance from the DCS historian. The disturbance was caused by a cracking furnac fur nacee shu shutdo tdown wn for dec decoki oking, ng, whi which ch cau caused sed ups upstre tream am process upset. As a result, the disturbance was propagated to the downstream process. The response time and trend

3510


Dynamic ic respon response se for the deetha deethanizer. nizer. Figure 6. Dynam


Figure 9. Start-up procedure for the case study.

Dynamic ic respon response se for the depro depropanize panizer. r. Figure 7. Dynam

Dynamic ic respon response se for the debut debutanizer. anizer. Figure 8. Dynam

predicted by the DS model match the DCS historian quite well, except for some system biases. Through troubleshooting, it is identified that this is because the DCS data does not satisf sat isfy y mas masss bal balanc ancee whi while le the dyn dynami amicc mo model del doe does. s. It actually reminded the plant that the flow rate sensors might have problems during that period. The simulati simulation on results and

analysis are also acknowledged by the plant engineers based on their industrial expertise. DS for Fla Flare re Min Minimi imizat zation ion.. Onc Oncee the stea steady-s dy-stat tatee and dynamic model validation has been completed, the plantwide DS model needs to be transformed to its initial state for startup. At this initial state, the whole process system runs stably with the total recycle. A light hydrocarbon mixture containing hydrogen, methane, ethane, ethylene, and propylene with the composition ratio of 1.5:15:1.5:73:9. in mole percent is circulated in the system. There are no system inputs and outputs at the initial state of start-u start-up, p, but only the inner and outer recycles are fully open. Based on the plant operation strategy, the startup is made up of two different feeds: heavy naphtha and light naphtha. One studied start-up procedure in terms of feed input is shown in Figure 9, which is also the input of the DS model. It shows the charge gas feed from heavy naphtha cracking furnaces (the first two furnaces) is incremented in two ramps for each of the first two furnaces with a half-hour duration in between. The feeds from light naphtha cracking furnace (the third through the seventh furances) are also incremented incremented in two ramps for each furnace but without any idle waiting between two furnaces. Based on this start-up procedure, procedure, the dynamic simulation simulation for product streams of DeC1, DeC2, DeC3, and DeC4 are shown in Figures 10-13. The overall dynamic responses show the startup wi will ll ta take ke ab about out 14 h to rea reach ch the normal normal st stead eady-s y-stat tatee operati oper ation, on, duri during ng whi which ch all the prod product uct stre streams ams are wit within hin specification. specifica tion. Historically, Historically, the plant start-up took about at least one day. Based on the simulation prediction, the new start-up procedure will help the plant save more than 44% of the startup ti time me.. Th Thee pre predi dicte cted d sta startrt-up up tim timee has bee been n rea reall lly y ac ac--


Figure 10. Dynamic response of demethanizer top products during the startup.

3511

Figure 13. Dynamic response of debutanizer bottom products. Table 2. Flare Minimization Results Results for the Case Study


start-up duration (h) shortest start-up in the pastb DS assisted start-up saved percentage (%)

25 14 44

amount of flared raw materials (klb) C1

C2

C3

2163

5569

3017

904 2237 1001 58.0 59.8 66.8

major emitted pollutantsa (klb) C4+

NO x

HC

19.5

270.6

918 7.4 67.0 6 62 2.1

101.2 62.6

2782

a Assumi Ass uming ng 98% flaring efficienc efficiency y and based on U.S U.S.. EPA Flare Efficiency Study Efficiency Study,, 1983 1983.. 22 b The amounts amounts of flar flared ed raw materials materials are estimated based on the plant start-up time and feedstock flow rate.

Figure 11. Dynamic response of deethanizer top products.

Figure 12. Dynamic response of depropanizer top products.

complished on site with the help of the developed plantwide DS model. The dynamic simulation provided an insight into process proc ess dyna dynamic mic beha behavior viors, s, whi which ch real really ly help helped ed the plan plantt beforehand start-up material preparations and process operation, and meanwhile increased the plant engineers’ confidence level. As an example, Figure 10 shows that during the start-up DeC1 top stream temperature will decrease continuously, while flow rate will generally decrease first and then increase. This is because the temperature of DeC1 feed from the chilling train will continuously decrease during the start-up, which makes the top temperature also decrease from -150 to -250 °F. Because of that, the top flow rate of C1 will decrease initially. initially. However,

as the system feed from cracking furnace increases, the total amount of methane in DeC1 feed increases with each furnace feeding in. The combined effects result in the obtained flow rate response. It helps operators avoid potential inappropriate control to increase the top flow rate, as the flow rate will come back automatically to the expected level after some time. As another example, Figure 13 shows that the DeC4 bottom flow rate will be zero during the first hour of start-up. The reason is that at the initial start-up state, C5s or heavier components do not exist in the system. As the charge gas is fed in the system, the C5s or heavier components will gradually accumulate in the sump of DeC4. The simulation predicts that the bottom flow ratee wi rat will ll in incre crease ase af after ter 1 h of sum sump p ac accum cumul ulati ation. on. Th This is informa info rmation tion helps the plan plantt prep prepare are the dow downstr nstream eam unit unitss operation for start-up. With the plantwide dynamic simulation, the flare emission data is summarized in Table 2. For simplification, the flared raw materials are aggregated and classified as C1 (methane), C2 (ethane and ethylene), C3 (propane, propylene), and C4+ (butane, butylene, butadiene, etc.). The flaring emissions are aggregated aggregate d as NO x and hydrocarbons, the calculation for which is based on 98% flaring efficiency and the U.S. EPA Flare Efficiency Study in 1983.22 For comparison, the emission data for the plant best start-up in the past are also calculated and shown in Table 2. The simulation shows C2 are the major flaring materials for both cases. The DS assisted start-up can save 58.0%-66.8% of flared raw material materialss under different categories. categories. As a result, the total NO x and hydrocarbons emissions of the DS as assis siste ted d sta start rt-up -up are est estim imate ated d as 7.4 and 101 101.2 .2 klb klb,, receptively. receptive ly. These are significa significant nt reduction by 62.1% and 62.6%, respectively, compared with 19.5 and 270.6 klb of emissions from the historical best start-up. It sho shoul uld d be in indic dicate ated d tha thatt the fla flari ring ng sou source rcess and thu thuss emissions in the case study have been significantly reduced because of the novel start-up procedure with total recycles (see

3512


Figure 4) and operatio operation n strategies (see Figure 9). The DS plays the important role to virtually verify the applicability of the startup procedures and the operation conditions, which helps reduce flaring during the plant start-up. The DS shows, due to various recycles during a plant start-up, that the majority of off-spec product streams (recycled H2 /C1, C2, C3, and C4 streams in Figure 4) from DeC1 through DeC4 have been reused instead of flared. Meanwhile, start-up time is significantly reduced (11 h less compared with the shortest start-up in the past). Note that the reduction of flaring time means the increase of normal manufacturing time. It also means the saving of raw materials from flar flaring ing for futu future re prod producti uction. on. Fur Further thermor more, e, the sali salient ent environmental and societal benefits from flare minimization are priceles pric eless. s. The predictions predictions of the sta start-u rt-up p tim timee and dyna dynamic mic responses are in good agreement with reality. Conclusions


Plant-wide dynamic simulation serves as a powerful tool in aiding flar aiding flaree min minimi imizati zation on for CPI plants today. Thi Thiss pape paperr developed a general methodology on flare minimization for chemica chem icall plan plantt star start-up t-up oper operati ations ons via plan plantwi twide de dyna dynamic mic simulation. simulat ion. The plantwi plantwide de dynamic simulation simulation is performed based on the integration of rigorous process models, plant design data,, P&I data P&ID, D, DCS historian, historian, and ind industr ustrial ial expe expertis rtise. e. It can predict accurately the dynamic behavior of a process prior to any real plant changes, which gives insight into the process behavi beh avior or th that at is not ap appar parent ent th throu rough gh SS or op opera eratio tional nal experience. experien ce. It helps test plant operating procedures and support decision making during plant start-up. It is also a cost-eff cost-effective ective approach for industrial emission source reduction. The benefits from this study are not only for environment and society but also for the econ economi omics cs and sust sustaina ainabili bility ty for the chem chemical ical process industry. Acknowledgment

This work was in part supported by the Texas Commission on Environmental Quality (TCEQ), Texas Air Research Center, and Texas Hazardous Waste Research Center. Literature Cited (1) Pohl, J.; Lee, J.; Payne, R. Combustion Combustion efficiency efficiency of flares. Combust. Sci. Technol. 1986, 50, 217.

(2) Xu, Q.; Li, K. Y. Dynamic simulation for chemical plant plant turnaround operation. Integr operation. Integrated ated Enviro Environment nmental al Manag Management ement Conso Consortium rtium Meetin Meeting, g, Houston, TX, June 25, 2008. (3) Nam, J.; Webster, Webster, M.; Kimura, Y.; Jeffries, H.; Vizuete, Vizuete, W.; Allen, D. T. Reduct Reductions ions in ozone concentration concentrationss due to contro controls ls on variab variability ility in industrial flare emissions in Houston, Texas. Atmos. EnViron. 2008, 42, 4198.

(4) Vizuete, Vizuete, W.; Kim Kim,, B.; Jeffries, Jeffries, H.; Kim Kimura, ura, Y.; All Allen, en, D. T.; Kioumourtzoglou, M.; Biton, L.; Henderson, B. Modeling ozone formation from industrial emission events in Houston, Texas. Atmos. EnViron. 2008, 42, 7641. (5) Webste Webster, r, M.; Nam, J.; Kimura, Y.; Jeffries, H.; Vizuete, Vizuete, W.; Allen, D. T. The effect of variability in industrial emissions on ozone formation in Housto Houston, n, Texas. Atmos. EnViron. 2007, 41, 9580. (6) Murph Murphy, y, C. F.; Allen, D. T. Hydrocarbon emissions emissions from industrial industrial release events in the Houston -Galveston area and their impact on ozone formation. Atmos. EnViron. 2005, 39, 3785. (7) EPA, National Ambient Air Quality Standards. http://www.epa.gov/ http://www.epa.gov/ air/criteria.ht air/crit eria.html, ml, access accessed ed March 8, 2008. (8) Doe, K.; Coppard, D. Process simulation simulation and ethylene cracker for operation training. Proceedings of the 13th Ethylene Producers Conference , Houston, TX, April 23 -26, 2001; p 656. (9) Chenev Chenevert, ert, D.; Harry, C.; Walker, J.; Unterbrink, Unterbrink, B.; Cain, M. Flare minimization practices improve olefins plant start-ups. Oil Gas J. 2005, 103, 54. (10) Greg, D. The use of dynamic simulation simulation to develop control strategies strategies for emergency situations. Compu Comput. t. Chem. Eng. 2000, 24, 1181. (11) (1 1) Vinod Vinod,, P.; Je Jeffr ffrey ey,, Y. Y.;; Su Suraj rajit it,, D.; Pr Prak akas ash, h, R. R.;; Ji Jiho hong ng W. Application of dynamic simulation in the design, operation, and troubleshooting of compressor. Systems Proceedings of the 36 Turbomachinery Symposium, 2007. (12)) Mar (12 Mario, io, R. E.; Arne, K. l.; Louis, Louis, H.; Sten, B. J. Dynamics Dynamics and control during start-up of heat integrated distillation column. Comput. Chem. Eng. 2000, 24, 1091. Plantw ntwide ide Dyn Dynami amicc Sim Simula ulator torss in Che Chemica micall (13) (1 3) Luybe Luyben. n. W. L. Pla Processing and Control , 1st ed.; CRC Press: Boca Raton, FL, 2002. (14) Bernard Bernard,, A. Start-up planning planning and support Following Following a plant revamp, 2007. AIChE Spring National Meeting, Houston, TX, 2007. (15)) Sin (15 Singh, gh, A.; Li, K. Y.; Lou, H. H.; Hopper, Hopper, J. R.; Golwala, Golwala, H.; Ghumare, Ghumar e, S. Flare minimization minimization via dynam dynamic ic simul simulation. ation. Int. J. EnViron. Pollut. 2007, 29, 19. (16) Li, K.; Xu, Q.; Lou, H. H.; Gossage, J. L.; Singh, Singh, A.; Vragolic, Vragolic, S.; Kelly, T. Flare minimization during plant start-up via dynamic simulation. PSE ASIA 2007, Xi’an, China, 2007. (17) Gore, S., Plant start-up in CPI via dynamic simulation: Case study. Master Thesis, Lamar University, Beaumont, TX, 2005. (18) Rajagopalan, S.; Pradeep, V.; Hiranmayee, V.; Ananthanarayanan, N. A framework for managing transitions transitions in chemical plants. Comput. Chem. Eng. 2005, 29, 305. (19)) Wan (19 Wang, g, L.; Makita, Makita, H.; Kano, M.; Has Hasebe ebe,, S. Dyn Dynami amicc sta start-u rt-up p model of heat integrated distillation column. PSE ASIA 2007, Xi’an, China, 2007. (20) Joao, A. F.; Arrude, L. V. R.; Flavo, N., Jr. Startup of a distillation distillation column using intelligent control techniques. Comput. Chem. Eng. 2005, 30, 309. (21)) Rut (21 Rutherf herford ord,, P.; Pers Persad, ad, W.; Laur Laurits itsen, en, M. Con Consid sider er dyn dynami amicc simulation tools when planning new plant start-up. Hydrocarbon Process. 2003, 82, 75. (22) U.S. Environmental Environmental Protection Agency, Flare Efficien Efficiency cy Study, EPA-600/2-83-052, Cincinnati, OH, 1983.

ReceiVed for reView October 24, 2008 ReVised manuscript receiVed January 19, 2009 Accepted January 22, 2009

IE8016219

Chemical Plant Flare Minimization via Plantwide Dynamic Simulation

Recommend Documents