Ethylene Dichloride Cracking Reactor Modelling

2y ago
13 Views
2 Downloads
979.63 KB
10 Pages
Last View : 2m ago
Last Download : 2m ago
Upload by : Esmeralda Toy
Transcription

Ethylene Dichloride cracking reactor modellingSusana Broeiro BentoChemical Engineering Department, Instituto Superior TΓ©cnico, Lisbon, PortugalProcess Systems Enterprise, London, United KingdomARTICLE INFODate:November 2017Key words:ModellingVCMCrackingRadical mechanismState estimationgPROMSABSTRACTVinyl Chloride monomer (VCM) is one of the most important commoditychemicals and it is produced mainly by the cracking of ethylene dichloride (EDC). Byproducts formation is inevitable, creating several inefficiencies, and accurate model of theprocess is essential for its optimizationIn the present work, an EDC cracker model was set-up using the furnace modelfrom gPROMS ProcessBuilder, developed by PSE. The cracking kinetic mechanismimplemented consists of 108 reversible reactions and 47 components, as reported byChoi et al. [1]The model predictions are compared to predictions from another model whichused a cracking kinetic model tuned to plant data. The deviations for the maincomponents were in the range of 1.4-1.9%. The deviations for impurities were moresignificant.A dynamic simulation of a cycle was carried out. The predictions of pressuredrop, VCM flow rate and EDC flow rate over the cycle were compared to plant data.Subsequently, state estimations were performed to assess the feasibility of improving themodel predictions and the initial results are positive.Finally, a study regarding the possibility of reducing the cracking kinetic schemewas initiated. Allowing a deviation of 0.1% from the original results, it was verified that 48reactions could be excluded without compromising the model accuracy. More testsconsidering other impurities in the furnace feed should be done to further validate thispossible kinetic scheme reduction.1𝐢2 𝐻4 2𝐻𝐢𝑙 𝑂2 𝐢2 𝐻4 𝐢𝑙2 𝐻2 𝑂21.IntroductionThe commercial significance of the vinyl chloridemonomer (VCM) can be highlighted by the productionof polyvinyl chloride (PVC), the world's second mostabundant plastic. PVC is used in the most diversesectors, ranging from healthcare to construction andelectronics. [2]Currently, vinyl chloride is mainly producedthrough the thermal cracking of ethylene dichloride(EDC). This is a balanced process, which means that allintermediates and by-products are recycled in a waythat ensures a tight closure of the material balance toonly VCM as the final product, starting from ethylene,chlorine and oxygen. [3] The process begins withchlorination of ethylene to ethylene dichloride:𝐢2 𝐻4 𝐢𝑙2 𝐢2 𝐻4 𝐢𝑙2(3)The overall reaction (eq.4) is exothermal so theVCM plant should be able to cover a large part of itsenergy needs.𝐢2 𝐻4 0.5𝐢𝑙2 0.25𝑂2 𝐢2 𝐻3 𝐢𝑙 0.5𝐻2 𝑂(4)The EDC cracking takes place in a pyrolysisfurnace (figure 1)(1)Followed by EDC dehydrochlorination to VCM, throughthermal cracking according to equation 2.𝐢2 𝐻4 𝐢𝑙2 𝐢2 𝐻4 𝐢𝑙 𝐻𝐢𝑙(2)Figure 1 - EDC cracker furnace diagramThe HCl produced during the EDC cracking isrecycled to the oxychlorination section, where it is usedtogether with ethylene to produce EDC (eq.3)In principle, the complex thermal cracking ofEDC is considered to proceed via free-radical reactions.Rigorous reaction mechanisms have been studied and1

2.3 ReadData Foreign Objectimproved several times by various researchers. [4]Ranzi et al. introduced a reaction kinetic scheme withmore than 200 elementary reactions with more than 40molecular and radical species. Borsa et al. [3] developed the most complex cracking kinetic mechanism forEDC pyrolysis, including 135 compounds and radicalspecies and more than 800 reactions. Choi et al. [1]established a mechanism that involves 108 reversiblereactions and 47 molecular/radical species. Theaddition of carbon tetrachloride as promoter was firstinvestigated by Choi et al. [1] Schirmeister et al. [5]simplified the EDC pyrolysis mechanism aiming thedata accuracy and expenditure optimization for modeladjustment. A total of 31 reactions, 18 compounds, and8 radical species were used to describe all relevantproducts, intermediates, and byproducts. [4] A typicalEDC conversion would be between 50 and 60%, inorder to limit by-product formation and obtainselectivities to VCM around 99%. [5]Even though it is possible to achieve high yields,the formation of by-products is inevitable, causingsignificant inefficiencies in the process.Coke formation is an important reason forconcern, since its deposition inside the reactor coilsdemands periodical shut downs of the unit. Besidescoke, there are other gas phase impurities such aschloroprene and butadiene that cause down-streamdifficulties in distillation columns.Having this in consideration, it is important toaccurately model the process aiming the model basedprocess optimization.With the ReadData FO, it is possible to addinformation to the model, regarding physical propertiesor even data to be use as input to the model variables.This information is obtained from a .txt file, from whichis converted into arrays.2.4 State EstimationState estimation is a widespread and wellestablished technique in control engineering andweather forecasting.If on-line data of some output variables areavailable, a state estimator can adjust the modelprediction using these measurements to obtain a betterestimate of the state. This is the most importantapplication of on-line state estimation according toSimon. [7]In gProms the Extended Kalman Filter isadopted, since it is one of the simplest and mostimportant tools for state estimation purposes. [7] Duringstate estimation, the model receives on-line measureddata regarding the input and output variables. For eachtime unit, the estimator updates the output variableaccording to a prediction/correction approach. Firstly,there is the prediction step, where model equations aretaken into account, followed by the correction step,where available measurements are used to correct thepredicted state estimate. Hence for each instant themodel will give two values for the output, resulting fromeach of these steps.For each output variable and parameter avariance is defined. The variance set to the parameterscan be interpreted as a measure of how much its initialvalue can change during state estimation, in order tomeet the objective. A higher variance will allow largerchange to the parameter value in comparison to asmaller variance. On the other hand, the variance of theplant data can be seen as a measure of the confidencethe model can have on it. A smaller variance indicatesmore accuracy of the measurement and stateestimation would give more importance to suchmeasurements in comparison to those with highervariance.To implement this technique in gPROMS, twofiles are required. The first is a configuration file, whereall the parameters, inputs and output variables to beconsidered are specified, as well as their variances.The second file is a text file that contains all the plantdata regarding the input and output variables.2. Materials and MethodsIn this work, the models were created andsimulated in gPROMS ProcessBuilder software,developed by Process Systems Enterprise. ThegPROMS advanced process modelling platform is apowerful equation-oriented modelling and optimizationtool on which all of PSE's gPROMS products are built.Besides the integral parts of gPROMS, it is alsopossible to use external software components, whichprovide a range of computational services to themodels. These are defined as parameters namedForeign Object (FO) and include physical propertiespackages, external unit operation modules, or evencomplete computational fluid dynamics (CFD) softwarepackages.2.1 MultiflashMultiflash is the standard gPROMS physicalproperties package, supplied by KBC AdvancedTechnologies. It is a highly rigorous properties package,which supports all commonly-used thermo-dynamic andtransport properties, including a wide range ofequations of state and activity coefficient thermodynamic models. This is achieved with a Multiflash inputfile (.mfl), in which all the components, physicalproperties models, among other things that arenecessary to the problem, are defined.3. Model Set-up – Furnace modelThe EDC cracker was set-up using the furnacemodel libraries within gPROMS ProcessBuilder.The furnace comprises several models inside ofitself, consisting in three mains sections (Figure 2): Convection Section – Where the hydrocarbonstream is heated;2

Radiant Section (Coil) – Where cracking reactionsoccur;Transfer Line Exchanger (TLE) – Where the coiloutlet stream is quickly quenched to preventdegradation of the highly reactive product throughsecondary reactions;assumed to behave the same and one representativecoil is modelled.Both the convection section and the coil modelsconsist basically in several cracking tube models.The cracking mechanism implemented was theone reported by Choi et al. [1] , consisting of 108reversible reactions, with 47 components, of which 22are radical species.The radiant section consists of multiple coilsoperating in parallel. In the furnace model, each coil isTLECONVECTIONSECTIONCOILgML to LSKMLSKM to gMLFigure 2- Furnace model from within gPROMS ProcessBuilder3.1 Cracking tube modelIn the cracking tube model, it is assumed a onedimensional plug flow due to the turbulent flow, as wellas low viscosity for the reaction side stream. This modelcalls for other sub-models: cracking kinetic model, cokingkinetic model, fluid properties model, heat transfercoefficient model and friction factor coefficient model.The coking model was also developed during thepresent work. For this model it was considered that cokeis formed through the dehydrogenation of Tar. Tardroplets form at high temperature in the pyrolysis furnaceand are transported through the heat exchanger to thequench tower. When these droplets impinge on the wallsurface they suffer dehydrogenation, originating coke [8].Thus the coking reaction rate is considered to be thereaction rate of tar dehydrogenation.In the coking model, the mass balance for Tar isdone considering that the concentration of Tar is given bythe difference between the Tar that is formed and isconsumed by dehydrogenation.The reaction rate of Tar formation (equation 8) isa function of acetylene and chloride concentrations, sinceit was concluded that the influence of other cokingpromoters were negligible.The cracking kinetic model is used to determinethe reaction rate. For the case when the reaction isconsidered irreversible the equation 5 is used, while theequation 6 applies for the reversible reactions.π‘πΆπ‘Ÿπ‘— π‘˜π‘“π‘— 𝐢𝑖 𝑛𝑓,π‘˜π‘—(5)π‘˜ 1π‘πΆπ‘Ÿπ‘— π‘˜π‘“π‘— πΆπ‘–π‘˜ 1𝑁𝐢𝑛𝑓,π‘˜π‘— π‘˜π‘Ÿπ‘— πΆπ‘˜ π‘›π‘Ÿ,π‘˜π‘—(6)π‘˜ 1In these equations π‘˜π‘“π‘– and π‘˜π‘Ÿπ‘– are the kineticconstants for the forward and reverse reactionsrespectively, 𝑛𝑓,𝑖𝑗 and π‘›π‘Ÿ,𝑖𝑗 are the reaction orders and 𝐢𝑖is the concentration of component 𝑖. The kineticconstants for the forward reactions (π‘˜π‘“,𝑗 ) are calculatedaccording to the equation 7. πΈπ‘Ž,𝑗(7)π‘˜π‘“,𝑗 𝐴𝑗 𝑇𝑏𝑗 exp ()𝑅. π‘‡π‘Ÿπ‘‘π‘Žπ‘Ÿ π‘“π‘œπ‘Ÿπ‘šπ‘Žπ‘‘π‘–π‘œπ‘› (𝑧) π‘˜π‘‘π‘Žπ‘Ÿ π‘“π‘œπ‘Ÿπ‘šπ‘Žπ‘‘π‘–π‘œπ‘› 𝐢𝐢2𝐻2 (𝑧)𝐢𝐢𝑙 (𝑧)(8)π‘Ÿπ‘‘π‘Žπ‘Ÿ οΏ½π‘œπ‘› (𝑧) π‘˜π‘‘π‘Žπ‘Ÿ οΏ½π‘œπ‘› 𝐢𝑇𝐴𝑅 (𝑧)(9)Both kinetic constants for Tar formation (equation8) and Tar dehydrogenation (equation 9) follow theArrhenius equation. The values for the activationenergies and pre-exponential factors were obtained froma previous project done by PSE (EDCM1).Where T is the fluid’s temperature (K), 𝐴𝑗 is thepre-exponential factor, πΈπ‘Ž,𝑗 is the activation energy ofreaction 𝑗 and 𝑏𝑗 is the temperature exponent used tocorrect deviations from the Arrhenius equation. Thekinetic constant for the reverse reaction is determinedusing the equilibrium constant. The equilibrium constantis calculated from the change of standard entropy ( 𝑆𝑗0 )The fluid properties Model is used to determine allthe properties required for the cracking tube model.Multiflash does not support radical species and theirproperties. For this reason, two sub models are calledwhether it is using a molecular based mechanism or aradical one.When using the molecular properties model, theMultiflash FO is used to get the information regarding theand enthalpy ( 𝐻𝑗0 ) during the reaction at system’stemperature (𝑇) and pressure (𝑃)3

position equal to 1 (T6 in Figure 3). By having the sixthtemperature, the model is able to calculate the remainingconstants of the equation 13 and determine thetemperature profile inside the coil. Figure 3 shows thetemperature profile in the coil, where the CIT appears ingreen, the five temperature measurements in blue and inpurple is the sixth temperature determined by the linearregression.Gas temperaturefollowing properties of the mixture: Density, viscosity,thermal Conductivity, heat capacity, enthalpy andcomponents molecular weight.In the case of a radical based mechanism, therequired properties are imported using the ReadDataforeign object. From this file, the model receives thefollowing properties for each component: molecularweight (𝑀𝑀 ), enthalpy of formation ( 𝐻𝑓 ), entropy offormation ( 𝑆𝑓 ) and the parameters for heat capacitycalculation (π‘Ž0 , π‘Ž1 , π‘Ž2 and 𝑏).The enthalpy of the mixture is calculated basedon the components enthalpy of formation, according toequation 10.𝑁𝐢 𝐻 [ 𝐻𝑓,𝑖 𝑀𝑖 ] ̅̅̅𝐢𝑝 [𝑇(𝑧) π‘‡π‘Ÿπ‘’π‘“ ](10)𝑖 10Where π‘‡π‘Ÿπ‘’π‘“ is the reference temperature (298.15K) and ̅̅̅𝐢𝑝 is the average heat capacity of the mixture,given by the weighted average of the heat capacities ofeach component (equation 11).𝑁𝐢̅̅̅𝐢𝑝 𝑖 1𝐢𝑝,𝑖 𝑀𝑖𝑀𝑀,𝑖0,814. Simulation Results(11)Following the EDC cracker set-up, it wasnecessary to test the performance and accuracy of themodel (EDCM2), which has a considerable size.In a previous work, an EDC cracker model wasdeveloped by PSE to simulate a specific industrial unit(EDCM1). Since it was considered that the modelaccurately described the system, at an early stage,EDCM1 was used to validate the results from the EDCcracker model developed in the present work (EDCM2).To make this comparison possible, the requiredinput variables were exactly the same in both models:feed composition and flow rate, coil inlet pressure (CIP)and Process gas temperatures (Temperature ProfileInterpolation Model).It was considered that the furnace feed wasmainly EDC ( 99 wt%) and a small amount of CCl 4.Carbon tetrachloride is known to be an efficient source ofCl radical and it can be used to promote the pyrolysisreaction. However, the Cl radical also acts as a promoterfor undesirable coke formation. [1].Regarding the output, this analysis was donefocusing on the variables considered relevant to describethe good behaviour of the model: Conversion of EDC,outlet composition and pressure drop. The results forEDC conversion and outlet composition will reflect theaccuracy of the cracking kinetic model. Regarding theoutlet composition, besides acetylene, only the two maincomponents were considered (EDC and VCM) since theyalone make 80% of the total. Acetylene was onlyconsidered in this analysis due to its relevance for thecoking model.The pressure is mainly affected by cokingdeposition, thus it can be used as an indication of theefficiency of the coking kinetic model implemented in thiswork.(12)Due to the lack of data, and considering the smallconcentration of radicals and by–products, the remainingproperties (viscosity and thermal conductivity) wereobtained using Multiflash for the main components (EDC,VCM and HCl).3.1 Temperature Profile interpolation modelThe temperature profile along the coil has anextreme importance in the accuracy of the results. Inprevious works, the temperature in the process wasdetermined by the heat balanced based on the flowratesof the flue gas, fuel and air fed to the furnace. [9].In this work, a model was developed where thetemperature profile is determined by polynomialapproximation, according to equation 13. This isachieved by considering as inputs to the model five realtemperature measurements (T1 to T5 in Figure 3Erro! Aorigem da referΓͺncia nΓ£o foi encontrada.) as well asthe respective axial positions. Besides this, the modelalso receives the Coil Inlet Temperature (CIT) from theupstream process simulation that corresponds to axialposition zero.𝑇(𝑧) π‘Ž 𝑏. 𝑧 𝑐. 𝑧 2 𝑑. 𝑧 3 𝑒. 𝑧 4 𝑓. 𝑧 5 𝑔. 𝑧 60,6Figure 3 - Temperature Profile in the coil representationπ‘‡π‘‡π‘Ÿπ‘’π‘“0,4axial positionThe heat capacity of each component isdetermined using a 3rd order polynomial fitting as shownin equation 12.𝐢𝑝,𝑖 π‘Ž0 . 𝑇 3 π‘Ž1 . 𝑇 2 π‘Ž2 . 𝑇 𝑏. 𝑑𝑇0,2(13)Considering only the last two temperaturemeasurements (T4 and T5 in Figure 3Erro! A origem dareferΓͺncia nΓ£o foi encontrada.), a linear regre-ssion isapplied in order to predict the temperature for axial4

Along with the input variables already pointed out,the cracking and coking kinetics are also inputs given tothe model.As mentioned before, in the present work(EDCM2) the cracking kinetics from Choi et al. [1] wereimplemented. However, in the EDCM1 the kinetics usedwere the same but tuned according to the real data fromthe plant. On the other hand, the coking kinetics werestrictly the same in both models. The activation energyand the pre-exponential factor were obtained from aprevious work using the data from the real plant.In Table 1, the deviation between the results ofthe two models is presented.in case A but having the kinetic constant for tar formationfrom acetylene multiplied by 3.5. The model predictionsfor cases A and B were then compared to the availablemeasurements, as it can be seen in Figures 4 to 8. Forthe first 20% of the cycle, real data was not reliable.Table 2 – Description of the two cases considered in this analysis.Table 1 – Deviation between the predictions from EDCM2 and EDCM1(using purely Choi kinetics and Choi kinetics tuned to the real data)Output variableEDCM1 w/Choi kinetics(%)Pressure ase ACase BCrackingkineticsChoi et al.(2001)Choi et al. (2001)CokingkineticsParametersused inEDCM1Same as Case A, but:π‘˜π‘‡π‘Žπ‘Ÿ π‘“π‘Ÿπ‘œπ‘š π‘Žπ‘π‘’π‘‘π‘¦π‘™π‘’π‘›π‘’ 3.5 π‘˜π‘‡π‘Žπ‘Ÿ π‘“π‘Ÿπ‘œπ‘š π‘Žπ‘π‘’π‘‘π‘¦π‘™π‘’π‘›π‘’ (EDCM1)Figure 4 shows the pressure drop predictions forboth cases as well as the real data. In case B due to thescaling factor considered, the coke formation is biggerthan in case A and the pressure drop in the coilincreases. For this reason, the results from case B arebetter matched to the data, presenting an averagedeviation of 1,8% against 9,1% from case A.EDCM1 w/Choi kineticstuned to realdata (%)1.41.91.448Pressure DropThe objective is that the results from EDCM2meet the results from EDCM1 with the tuned kineticssince this is the case that more accurately represents thereality. If in the EDCM1, the kinetics used were purely theones from Choi all these errors would be within 0,2%, asshown in Table 1. Thus the deviations from EDCM1 withtuned kinetics and EDCM2 result from the differencebetween the cracking kinetics.EDCM2 under predicts the acetylene composition in about 50%, as shown in Table 1. As acetylene isrelevant for the coking formation, this discrepancy willhave an impact on the pressure drop predictions.It was then verified that the reaction rate for Tarformation from acetylene in EDCM2 was in average 3.5times lower than the one predicted by EDCM1. Havingthis in consideration, the value of 3.5 was used as afactor to be multiplied by the kinetic constant of Tarformation from acetylene.DataCase ACase B00,20,4Time0,60,81Figure 4 – Coil pressure drop predictions for case A and case B againstreal Data with time being normalised.The effect of scaling factor on outlet flow ratepredictions are not significant ( 0.1%) as presented inthe figures below (Figures 5 to 7). The average deviationbetween model predictions and real data is shown inTable 3.Figure 5 shows the model predictions for EDCoutlet flow rate from both cases and the respective data.The predictions have an average deviation of 4.4% and4.5%, for case A and B, respectively.4.1 Simulation of a cycle (DynamicSimulation)After evaluating the start of run simulation results,simulation of a complete cycle is performed, consideringa period of around 12 months. The model inputs for thesimulation were obtained from plant data and thecracking kinetics from Choi et al. were used.The input variables analysed are the same as theones mentioned before. While for the output variables, inthis validation the outlet flowrates of the maincomponents were considered instead of the outletcomposition.Considering the significant discrepancy in theacetylene concentration previously presented, for thecoking kinetics two cases were considered (case A andB), presented in Table 2. Case A considers the cokingkinetics previously described and already used forEDCM1 and EDCM2. In case B, the scaling factorinfluence is tested so the coking kinetics are the same as5

VCM outlet flowrateEDC outlet flowrateDataCase ACase B00,20,40,60,8100,20,60,81Figure 7 – VCM outlet flowrate predictions for case A and case Bagainst real Data with time being normalised.It is verified that the flowrate predicted by themodel is in general lower than the real data, thus itseems that the model slightly over predicts EDCconversion. EDC conversion was calculated according tothe equation 14, for cases A and B, as well as for thedata. The EDC inlet was obtained from the unit real datawhich was used as an input to the model. The EDC outletcorresponds to the model predictions in each case.The model predictions have an averagedeviation from the real data of 1.8% for both cases. Themodel predicts in average a higher VCM outlet flowrateas it would be expected, since the model over predictsthe EDC conversion.Table 3 presents the average deviation from themodel predictions and the real data for each outputvariable for both cases.(14)Table 3 – Average deviation from real data for each output variablepredictions in case A and B.Figure 6 shows EDC conversion over time. Forboth cases, the EDC conversion is higher than the onecalculated from the data in about 3.3%.Pressure dropDataCase ACase BEDC outletflowrateVCM outletflowrateEDC conversion0,150,4TimeFigure 5 – EDC outlet flowrate predictions for case A and case Bagainst real Data with time being normalised.EDCinlet EDCoutletEDCinletCase ACase BTimeConversionEDC DataCase A (%)Case B (%)9,11,84,44,51,81,85. State Estimation0,45TimeIn state estimation the objective is to adjust themodel predictions to the real data, by changing somechosen related parameters.To apply this technique, it is necessary to setvariances for each adjusted parameter and the datacorresponding to the output variables.The variance in the case of the parameters canbe seen as a measure of how much the model can varyits initial value, in order to meet the objective. A highervariance will allow larger change to the parameter valuein comparison to a smaller variance. On the other hand,the variance of the real data can be seen as a measureof the confidence the model can have on it. A smallervariance indicates more accuracy of the measurementand state estimation would give more importance to suchmeasurements in comparison to those with highervariance.Table 4, shows the parameter to be adjusted bythe model, β€œcoking reaction rate adjustment” (πΆπ‘…π‘…π‘Žπ‘‘π‘— )with the only objective of fit the pressure drop data.0,75Figure 6 – EDC conversion in the outlet of the coil over time for case Aand B and real data with time being normalised.In figure 7 the results for VCM outlet flowrateare presented.6

The adjusted parameter is being multiplied by theexpression used to calculate the coking reaction rate, asindicated by its name (eq.15)π‘Ÿπ‘π‘œπ‘˜π‘’ π‘“π‘œπ‘Ÿπ‘šπ‘Žπ‘‘π‘–π‘œπ‘› π‘Ÿπ‘‘π‘Žπ‘Ÿ οΏ½π‘œπ‘› πΆπ‘…π‘…π‘Žπ‘‘π‘—Figure 9 shows the state estimation results forpressure drop. The model predictions are similar forcases 1 and 2, with average deviations from real data ofabout 2.6% and 2.8%, respectively, as shown in table 5.Both cases show improvements when compared with theresults without state estimation, which presents anaverage deviation of 4.1% (table 5).(15)Pressure DropFor πΆπ‘…π‘…π‘Žπ‘‘π‘— , an initial value of 1 is set, so if stateestimation is not being employed, this value does notchange, hence it will not affect any result. During stateestimation, the model is allowed to change thisparameter over time. If this parameter increases, thecoking formation will also increase, leading to anincrease in pressure drop. The decrease of thisparameter will have the opposite effect.Two cases were considered to be presented inthis analysis (case 1 and 2), that vary from one anotheronly in the variance defined to the coking reaction rateadjustment (table 4).Following the line of thought described above, incase 1 the model is allowed to make more significantchanges in the coking reaction rate adjustmentparameter, since the variance set to this parameter ishigher.SE Case 1SE Case 2Simulation05 10-65 10-7Pressure drop5 10-65 10-6EDC outlet flow rateVCM outlet flow rate1 10201 10201 10201 10SE Case 1Reaction Rate AdjustmentOutputsVariableCase 2πΆπ‘…π‘…π‘Žπ‘‘π‘—0,6Despite the proximity of the results from case 1and 2, the first one is still able to present betterpredictions. This is in agreement with what wasexpected, since in case 1 the model has more freedom tochange the coking rate parameter in order to adjust thepressure drop predictions to the data.Figure 10 shows the evolution of the reaction rateadjustment parameter over time. In the Simulation thereis no change of this parameter from its initial value, sincethe model is not doing any state er/VariableCase 10,2Figure 8 – Pressure drop predictions from state estimation againstreal data with time being normalised.Table 4 – Parameters adjusted in state estimation as wellas the output variables considered and respective variances for cases1 and 2.TypeData20The variances set to the outlet flow rates aremuch higher than the order of magnitude of thesevariables (more than 1 1017 times). These values weredefined to assure the variances are high enough for themodel not to consider its data as relevant since it is onlyintended to meet the pressure drop data.On the other hand, the variance set to pressuredrop is relatively small. Hence the model will adjust theparameter with the only objective of meeting the pressuredrop data.Figures 9 to 12 show the model predictions withstate estimation for both cases. The model predictionswithout state estimation (Simulation) are also presented,so it is possible to verify if state estimation has improved,or not, the results when compared to the real data.Similarly to section 4, the data corresponding tothe first 20% of the cycle was considered to beunreliable. In addition to that, for state estimationpurposes, only the first 70% of the cycle was considered.SE Case 2Simulation00,20,40,6timeFigure 9 – Reaction rate adjustment variation from state estimationwith time being normalised.For the other two cases, it can be seen that theparameter does not vary at the beginning of the cycle.This is because in that period there is no data forpressure drop. Considering the model is only workingtowards the pressure drop adjustment if there isn’t anydata the model does not change the parameter.As it was expected, in case 1 the cokingreaction rate adjustment parameter will have moresignificant oscillations, since the set variance is higherthan in case 2.As the pressure drop increases, the modelintensifies the coke formation, by increasing the cokingreaction rate adjustment parameter. At some point,around 35% of the period, the data shows a lower peak7

VCM outlet flowratein pressure drop. Thus the model responds similarly bydecreasing the parameter. Around 50% of the time, theparameter starts to increase again until another decreasein pressure drop occurs (around 58% of the time). Theparameter continues decreasing since then, evenassuming negative values at the end of the cycle.The fact the parameter has a negative valuemeans the coking reaction rate is also negative, which inreality would represent coke dissipation. Obviously in areal system, after being formed, coke will not disappearunless the operation is stopped and a decoking processis implemented.In case 2, it can be verified the parameter has asimilar behaviour than in case 1, but with oscillations oflower amplitude. Contrary to case 1, the parameter neverreaches negative values.Therefore, it is possible to verify that thevariance settings on measurements and parameters areimportant to get meaningful values of the adjustedparameters.For the remaining output variables, the resultswith state estimation (cases 1 and 2) show improvementswhen compared to the Simulation case.In figure 11, the results for EDC outlet flow rateare presented. Case 1 and 2 have average deviations of5.3% and 5.5%, respectively. Without state estimation,the model predictions present an average deviation of6.3%.SE Case 1SE Case 2Simulation00,20,40,6timeFigure 11 - VCM outlet flowrate predictions from state estimationagainst real data with time being normalised.Table 6 shows the average deviations of themodel predictions from real data, for each caseconsidered and output variable.Table 5 - Average deviations (%) of state estimation predictions fromthe real data for each case and variablePressuredropEDC outletflowrateDataSE Case 1SE Case 2SimulationEDC outlet flowrateDataVCM outletflowrateCase 1Case 6Table 6 - Improvement of the model predictions with state estimationrelatively to the prediction only case.00,20,40,6timeCase 1Case 2Figure 10 – EDC outlet flowrate predictions from state estimationagainst real data with time being normalised.Pressure dropEDC outlet flowrate43,2739,216,112,0Figure 12 shows the results for VCM outlet flowrate. For this component, the model predictions presentan average deviation of 3.4% for case 1 and 3.1% forcase 2. The predictions without state estimation have anaverage deviation of 4.2%, which is, as expected, higherthan the other two cases.VCM outlet flowrate5,0912,336. Kinetic ReductionAs a final step in this work, a study regarding thepossibility of reducing the cracking kinetic scheme wasattempted. The

chemicals and it is produced mainly by the cracking of ethylene dichloride (EDC). By-products formation is inevitable, creating several inefficiencies, and accurate model of the process is essential for its optimization In the present work, an EDC cracker model was set-up using the furnace

Related Documents:

models of the thermal cracking process of Ethylene Dichloride (EDC) to produce Vinyl Chloride Monomer (VCM) at Abu-Kammash Chemical Complex. The developed design was set against the amount of heat saved. In addition, Both EDC cracking units ( old and new design) were . Exergy Analysis of Ethylene Dichloride Cracking Unit at Abu-Kammash Chemical .

OCCUPATIONAL EXPOSURE TO ETHYLENE DICHLORIDE (1,2 - DICHLOROETHANE ) . Medical 2 Section 3 - Labeling (Posting) 3 . Ethylene Dichloride 7 Section 6 - Work Practices 8 Section 7 - Monitoring and Recordkeeping 11 II. INTRODUCTION 14 III. BIOLOGIC EFFECTS OF EXPOSURE Extent of Exposure 16 Historical Reports 18 Effects on Humans 21

Technology Manager Ethylene Furnaces Technip Benelux New Technologies in Ethylene Cracking Furnace Design Presentation for AIChE NL/B Zoetermeer, 31 October 2017. 2 Table of Contents 1. Introduction TechnipFMC 2. Cracking Fundamentals . Supply of 400 Cracking Furnaces Liquid Furnaces GK6 .

Coupled simulation of furnace and cracking reactor in 3D Simulation of butane cracking Ethylene concentration profile in a 2D model. Conclusions Simulations of the COMSOL 1D model for thermal cracking of ethane at plant conditions show absolute deviations less than 5% when you compare them with actual GC analysis.

The cracking reactions occur via free-radical mechanisms, and for the cracking of naphtha, resulting in yields of ethylene stream is made up of (in between 25 35% and propylene between 14 18% [1]. butadiene as well as 2.1 Steam Cracking process A simplified flowsheet of the steam cracking process that be found in Figure 3.

As shown in Figure1, the ethylene cracking furnace is one of the core ethylene production equipment. Ensuring its long-term operational safety and stability is the prerequisite for normal ethylene production. The furnace tube is the main component of the cracking furnace, and it operates in the combustion chamber filled with high-temperature .

3. E/P Feed to the Furnace - for ethylene plants using Ethane/ Propane mix as the feed, this analyzer provides feed-forward data for advanced furnace control in order to maximize cracking furnace efficiency. For ethylene plants using naphtha a the feed to the cracking furnaces, a process FT-NIR is often used instead of a GC. Ethylene Plant 2

Animal Nutrition is a core text for undergraduates in Animal Science, Veterinary Science, Agriculture, Biology and Biochemistry studying this subject. It also provides a standard reference text for agricultural advisers, animal nutritionists and manufacturers of animal feeds. The latest edition of this classic text continues to provide a clear and comprehensive introduction to the science and .