Spyro Resumo Alargado - ULisboa

1

EXTENDED ABSTRACT

Spyro modelling and steam cracking furnaces optimization Fábio Santos, Henrique Matos, André Vilelas & Jorge Fernandes

The ethylene and propylene production worldwide is mostly accomplished through the steam cracking process. In the current work, furnaces have been modelled in Spyro Suite 7 software to perform the tuning to the reality conditions. Initially, the feedstocks susceptible to thermal cracking were thermodynamically analysed to input the correct properties in the model. Relatively to naphtha, which enters the coils in liquid state, an error in the file, containing the right properties, import occurs and to perform Spyro simulations that problem must be fixed by the company propriety, Technip. For the remaining feedstocks entering the coils in gaseous state, the properties contained in the model are suitable. After the convergence of the model and its tuning to calculate the real outside HTC factors, a performance indicator (KPI) including the deviations between the COT and the BFW and fuel gas flow rates obtained by the software and the reality was considered. This indicator proved that the deviations obtained by the reality model were smaller than the normal convergence with the outside HTC factors equal to one. The possibility of revamping the convection section banks, which are authentic heat exchangers, led to a positive conclusion in the balance between high pressure steam (HPI) produced and fuel gas consumed. There was a greater impact on the first bundle of the convection section (HTC) having the lowest outside HTC factor. This was also proved by a simplified economic analysis. This was also proved by a simplified economic analysis. With the designation of ‘’protection steam’’, is added to the furnace in order to protect the coils in case of some operational failures and minimize coke formation at HTC bundle due to premature cracking. After a sensitive analysis concerning the flow rate of this feedstock, it was concluded that this value should be the highest allowed by the coils design specifications, maintaining the steam/hydrocarbon ratio. In addition to these studies, many more can be realized now that Repsol has their furnaces modelled in the software Spyro Suite 7. Keywords: Steam cracking, Spyro Suite 7, Thermodynamic properties, Outside HTC factors, Protection steam, KPI.

1. Introduction

1.1. State of the Art pyro was created in the late sixties with the objective of predicting olefin yields from hydrocarbon pyrolysis. This program was a marriage between a mathematical and kinetic

model by the Professor Mario Dente and Eliseo Ranzi [1]. Since then, Spyro has been tuned and validated not only in kinetic scheme but also in a full steam cracking simulation and optimization tool. For many years, the complexity of the radical reactions and mathematical scheme were studied with a view to find practical methods characterizing all possible reactions (more than 6600 nowadays). To know the initial distribution of pyrolysis products, a Rice theory based on the formation of radicals that by chain mechanism results in olefins production was created [2]. Along limited number of isomerization and decomposition kinetic parameters, it became possible to predict the primary products from decomposition of saturated species. Improving prediction of heavier feedstocks, such as VGOs and HVGOs, pyrolysis product distributions, simplifications were made in order to reduce the number of reactions and related kinetic parameters and also the number of species to be considered. It’s well known that the heavier the feedstock is, more difficulties are present in the mechanistic modelling and for that, correct feedstock characterization is necessary. To improve the program, Technip, before known as KTI (Kinetics Technology International), added new ways to characterize heavy feedstocks based on PINA, density, TBP or ASTM boiling points. In a way to reduce the convergence time, Spyro was reduced in sub-models. Sub-models for entrance, section, nodes and exit of the cracking coil and between the transfer line exchanger and the one before were attach to the existing ones like feed and mixer [3]. With the appearance of larger computers and thermochemical kinetic theory and experimental data, a model with the same mechanistic kinetic scheme for feedstocks from ethane thought gasoils, became usable to solve booth simulation and optimization cases. That model was called Equation-based Spyro [4]. A new distributive reaction-mixing synthesis model, d-RMix, incorporating ideal CSTR, PFR and DSR models was applied for

thermal cracking of ethane. The maximum amount of ethylene per unit mass of ethane was obtained for the plug flow mode in isothermal and isobaric operations with maximum temperature and minimum pressure allowed [5]. At that point, a flowsheet package was ready for industrial uses. Demanding more flexibility to Spyro, a symbolic model definition (SMD) module was integrated. A language similar to the gPROMS one was adapted to the program to supply information during the solution procedure, making easier to find possible errors inside the model [6]. In ethane cracking a maximum ethylene yield of about 67 wt% was revealed by an optimization with a linear-concave temperature profile between 1200 and 1300K [7]. That shape of temperature profile also revealed that successive reactions of product ethylene to larger side-products were minimized. In temperatures above 1300k, dehydrogenation reactions and H-abstraction of ethylene became more important [6].

1.2. Principles of Steam Cracking Steam cracking was first patented in 1913 and the first facilities appeared in the United States in the years near 1920 [10]. Since then, became the main process to produce ethylene and propylene and other valuable petrochemicals. Also known as pyrolysis, this process involves a breaking carbon-carbon or carbon-hydrogen bond within hydrocarbon molecules in the presence of steam and high temperatures. Ethylene and propylene are most valuable unsaturated molecules originated by this process along with other olefins and diolefins [11]. Cracking reactions require high input of heat energy and high temperatures since they are highly endothermic. Inside these extreme conditions, we have cracking temperatures rounding 800 to 850ºC, residence times of 0.1 to 0.5 seconds and low pressures, slightly higher than atmospheric one [11]. These extreme conditions are justified by the stability of the hydrocarbons like for example, at temperatures near 815ºC ethylene start to be more stable than ethane [9]. In the top of the stability career are coke and hydrogen.

s

2

To reduce the amount of coke which is a product of secondary reactions, residence times must be very low increasing the yield of primary products like ethylene and propylene represented in table 1. Those short residence times with very rapid cooling or quenching of the cracked products favours the yields of the desired hydrocarbons [12].

Table 1 - Steam cracking products [12].

Feedstocks/ Steam

Primary Reactions Secondary Reactions Ethylene C4 products

Propylene C5 products Acetylene C6 products Hydrogen Aromatics Methane C7 products

- Heavier products (coke, etc.) These hydrocarbons are mainly produced by the primary reactions that prevail over secondary ones due to the conditions said before and the reduced hydrocarbon partial pressure by steam present on the system. Secondary products result from reactions where the number of the molecules decrease, so decreasing the pressure favours the formations of the desirable ones. Steam also reduces the partial pressure of high molecule-mass aromatics, reducing the tendency to deposit and form coke in the radiation coils [12]. The amount of steam added to the hydrocarbon feed depends on the feedstock used and is usually expressed in a mass/volume ratio. The chemistry of steam cracking (figure 1) is extremely complex since it includes primary and secondary cracking of C-C bonds leading to hydrocarbons with shorter chains, dehydrogenation reactions yielding unsaturated species and addition reactions leading to formation of aromatics and coke. To have a clear vision of the chemistry of the steam cracking process the scheme below show the cracking net of n-heptane.

Figure 1 N-heptane cracking reaction steps (adapted by [15]).

Summing up, to have maximum ethylene and propylene production, the feedstocks must be highly saturated, the coil outlet temperatures must be high, steam is required in the system and the residence times must be low. A variety of different feedstocks can be used in the steam cracking process starting from the lightest like ethane and propane, passing through butanes and naphthas and finishing in the heavy ones like diesel fuels and distillates which the costs are lower than the light ones. Ethane, being associated with natural gas and commonly used in American and Middle East steam crackers, is the feedstock that gives higher yields of ethylene and lower of propylene, C4 hydrocarbons and aromatic gasolines. Naphtha is the main feedstocks used in the Europe steam crackers as they supply a large range of products like 25 to 30% of ethylene, 13 e 17% of propylene and about 20% of gasoline. The heavier feedstocks produce more fuel oil and les ethylene then the lighter ones. The amount of each product depends on the severity of the process. The propylene to ethylene mass ratio is the most common measure of severity among

the coil outlet temperature and others. When low-severity cracking conditions are present, the P/E ratio is higher than the one in the high-severity conditions.

1.3. Process Description Steam crackers (figure 2) are industrial facilities that can vary since the state of feedstocks to the product specification products required as the end of the chain. They are usually divided in three main zones with different equipment and functions. Starting with the hot zone including the cracking furnaces, the quench exchanger and the columns of the hot separation train, passing through the compression zone including a cracked gas compressor, dryers, purification and separation columns and finishing with the cold zone comprising the cold box, methanator, separation columns, C2 and C3 converters and the gasoline hydrogenation [11].

Figure 2 - Steam cracking process diagram

Starting in the storage area where the feedstocks are available, they are then pumped and preheated by heat recovery from effluents of the unit and finally fed separately in the coils of the furnace. The hydrocarbons enter the convection section where they are pre-heated with dilution steam by the heat released from the flue gas burned at the radiation section where the cracking reactions occur (figure 3).

Figure 3 - Cracking furnace scheme (Drum instead of Pocket)

In the coils outlet, the effluent gas passes by an indirect quenching in the transfer line exchangers (TLX or TLE) where high pressure steam (HPI) is generated and by a direct quenching by injection of quench oil. These two brutal quench from temperatures of 830/860ºC to 380-470ºC at TLE outlet and 200ºC after quench oil injection immediately stops the chemical reactions and avoid formation of coke in the coils [12]. The heavier fractions, mainly fuel oil, are then separated by the bottom of the primary fractionator tower going out through the top, gasoline and lighter fractions. The fuel oil can be stored or can be

3

used as quench oil in the effluent direct quench, after being cooled in heat exchangers where feedstocks are pre-heated and steam generated. From the top leaves the gas effluent which, with water, fed a scrubber. In here the water and heaviest gasoline condense coming out from the bottom of the column going the last one to the hydrogenation section. The remaining water, after removed the hydrocarbons left, becomes process steam by passing through heat exchangers [12]. At this point finishes the hot zone of the process and starts the compression zone. After the first two separations, the effluent gas is compressed by a centrifugal compressor that generally is composed by five stages. In the inter-stages, the gas is cooled, which, by the increase in pressure, leads to condensation of water and hydrocarbons. The liquid condensates are separated from the gaseous phase in drums situated between the stages of the compressor. Heavier hydrocarbons, mainly gasoline, from the first, second and third stages goes to the hydrogenation section. From the last two stages, the condensates after being cooled, feed the deethanizer to separate the C2

– from C3+

fraction going the last one to the depropanizer. To generate the amount of energy necessary for the compression a steam turbine is necessary. That energy is provided from the high-pressure steam (HPI) produced in the transfer line exchanges (TLX) situated at the furnaces as mentioned before [9]. The effluent gas has acid compounds in his composition like hydrogen sulphide (H2S) and carbon dioxide (CO2) which are removed in a washing tower situated between the fourth and fifth stage of the compressor. Normally is a caustic washing scrubber where that compounds are eliminated in the form of sodium sulphide, soluble in a soda caustic solution, and sodium carbonate respectively [12]. Before entering the deethanizer the gas effluent is cooled and dried in order to remove the water, avoiding the risk of hydrate formation and water icing [12]. Starting at this point the cold zone, by the top of the deethanizer leaves the C2

- fraction containing acetylene (C2H2) that is further converted to ethylene in adiabatic reactors. Another drying step is done to insure no water to the rest of the process. Now hydrogenated, C2- fraction, needs to be cooled to separate methane (CH4) liquid and hydrogen (H2). With help of Joule-Thompson effect and cold fluids like methane, hydrogen and ethylene, hydrogen is removed by condensations of methane. The hydrogen stream, before going to the hydrogenation reactors, needs to suffer a removal of carbon monoxide (CO). Methane and C2- fraction, with no hydrogen left, are separated in the demethanizer. From the bottom comes out a mixture rich in ethylene and ethane and from the top methane which can be used as fuel gas in the furnaces. Ethylene and ethane are separated in the C2 splitter. As said before, ethylene is used as cold utility. The refrigeration is possible due to the Joule-Tompson effect and a centrifugal compressor. The same happens to propylene supporting higher temperatures than ethylene [11]. The C3+ fraction from the deethanizer is the feed to the depropanizer. Here, C3 fraction is separated from C4 and C5 fractions. The lighter passes through a hydrogenation reactor to convert components like propyne and propadiene to propylene. Propylene is finally separated from propane in the C3 splitter. Finally, C4 is separated from C5 fraction in the debutanizer and the last following to the gasoline hydrogenation. This C4 fraction is rich in butadiene which is a valuable product like ethylene and propylene. The valuable products can be storage or can be used directly in other processes like for example, Repsol’s case which the high and low density polyethylene plant has as main feedstock the ethylene produced in the same site.

2. Spyro Used by almost all steam crackers in the world, Spyro main use is to predict the yields of the products in the conditions implemented

by the user. Besides that, it can be used for global simulation of furnace along with the steam drum and the transfer line exchanger representing the cracking itself. It is the main tool for the tuning of the furnace as well as for the study of revamping cases viability. Repsol has 8 furnaces operating in Sines. A typical scheme of the Repsol’s furnaces is shown in the figure 4. All feedstocks before entering the coils in the convection section are mixed with steam, named “protection steam”. This steam protects the coils from thermal shocks caused for example when the feedstocks flow became unstable or when the furnace goes down. That protection steam comes from the dilution steam which is mixed with the hydrocarbon feed before the feed preheater (FPH) and the dilution steam super heater (DSSH) banks. After the mixture’s made, it enters the last bundle of the convention section, high temperature coil (HTC), and the goes to the radiation section. In the economizer (ECO) bank the boiler feed water is pre-heated and fed to the steam drum, where one liquid and one vapour phase are present. That liquid water flows from the steam drum to the TLX where high pressure steam is generated by the cooling of the effluent to stop the chemical reactions. The HPI goes back to the steam drum being then superheated in the high pressure steam super heater banks (HPSSH I and II). Between these two banks there is an injection of boiler feed water to control the final temperature of the HPI leaving the furnace. The heat transferred from the flue gas going out of from the radiation section at high temperatures allows the required temperatures for the process side fluids. That steam produced is then fed to a turbine to generate work, moving the centrifugal gas compressor being the two coupled.

Figure 4 - Furnace scheme at Repsol's site

2.1. Feedstocks For the cracking process, the feedstocks used are naphtha, butane, propane and ethane. The choice of which one will be used in the process is made with an evaluation based on the product yields obtained and their price in the market.

2.1.1. Propane/Ethane Feedstocks need to be well defined for the start of the process. Spyro Suite 7 check the vapour fraction in the process conditions by the properties inside the model. For propane and ethane, the property table selected must be the gas feedstock with the initial conditions, namely the vapour fraction should be set equal to 1. Composition of the feed must be defined with the components and their weight or volume fractions as well as flow rate and inlet temperature with the help of real data (figure 5). Pressure values of feed sources are not fixed, this allows Spyro to calculate them by fixing the coil outlet

4

pressure of the effluent generated by the centrifugal compressor after the furnace.

Figure 5 - Propane's feed characterization in Spyro

Besides ethane being the most used feedstock for ethylene production, it is a product of other common feedstocks. In Europe, mainly, this compound is recycled and continuous fed to the furnace after being separated from ethylene in the C2 splitter. Ethane is the main feedstock in the north of USA and Middle East because they extract it from the currently shale gas at lower temperatures.

2.1.2. Naphtha/Butane When detailed composition is not available to user, other ways of defining the feed are included in Spyro like for naphtha, gasoil and air. That characterization can be made with a PINA analysis, the specific gravity and the boiling curve percentages (figure 6). PINA analysis is made in mass or volume and it appears as normal paraffins, iso-paraffins, naphtenes and aromatics fractions. The boiling curve can be TBP, ASTM-D86 or ASTM-D2887 type according with the test made to that feed. ASTM-D86 is the more common one because it can be determined in a laboratory. In case of naphtha feed, the initial boiling point must be at temperature below 130ºC and 95% vaporized point below 180ºC. The temperatures from these two points with the one from 50% vaporized are required in the curve characterization. The 95% point avoid adding additional components due to very long tails in the ASTM data meaning that if the last 5% of the feed boils over a wide temperature range Spyro feedstock characterization will not be skewed in favour of the heavier components [8].

Figure 6 – Naphtha's feed characterization in Spyro

In Repsol case, naphtha enters the convection section at liquid state where is preheated and vaporized. For that reason, to describe the feed, properties like fraction, enthalpy, viscosity, thermal conductivity and specific heat of booth liquid and vapour phase need to be calculated.

With the help of Aspen Hysys V8.4 a stream analysis was made with the intuition to have the table including all the properties described before. After the analysis, the results can be exported to excel not forgetting that the values of enthalpies given by Hysys tool comprises the enthalpy of formation of the mixture called naphtha in the case study. The enthalpy of formation need to be calculated in a way of deducting from the enthalpy of each state given by Hysys (table 2) [13]. To calculate the enthalpy of formation to naphtha liquid and gas, a weighted average of the majority compounds present his composition provided by Repsol’s laboratory.

Table 2 - Enthalpy of formation of naphtha in gas and liquid phase

Finally, with the correct properties, a txt file (figure 7) is made and saved as ppf. extension readable by Spyro Suite 7.

Figure 7 - Naphtha properties written in a txt file

In this txt file, begin (LIQ 1 VAP 0) and end (LIQ 0 VAP 1) vaporization points need to be well defined as in the figure 7. The heattype equal to 2 indicates the presence of mixed vapour and liquid properties, the spectype equal to 0 indicates tabulated properties and for last the zero after unit means a system with metric units. The initials represent each property and the last letter indicates the phase like for example, VISL means viscosity for the liquid. At last that file is imported to the simulation and naphtha uses that properties in the bundles before the high temperate convection (HTC) one. In this last one naphtha is already all vaporized so the property table used is the same as for propane and ethane. In a continuous basis, naphtha is mixed with a C5 fraction coming from gasoline before entering the furnace. This fraction can be

Compound Fraction ΔHf(l) (KJ/mol)

ΔHf(l) Mixture (Kcal/Kg)

n-pentane 0.327 -173.5

-533.7 n-hexane 0.205 -198.7

cicloheptane 0.174 -156.4

2Mpentane 0.294 -204.3

Compound Fraction ΔHf(g) (KJ/mol)

ΔHf(g) Mixture

(Kcal/Kg) n-pentane 0,327 -146.8

-476.7 n-hexane 0.205 -167.1

cicloheptane 0.174 -124.8

2Mpentane 0.294 -174.3

5

mixed or can continue in the gasoline if the market indicates it always respecting the reid vapour pressure specification. Therefore, if the price of naphtha is higher than the gasoline, the recycling compensates like the opposite if the price is lower. Remind that the gasoline as final product have to be within specifications. That flow rate is small comparing to naphtha so it is not necessary to add new properties only for C5 fraction. The same procedure of adding properties to the model happens with butane.

2.2. How to obtain Spyro’s convergence Spyro can be simulated as full furnace (figure 8) with radiation coil, TLX with steam drum, firebox and convection section models or only for yield simulations containing only the radiation coil model.

Figure 8 - Full furnace flowsheet simulation

To obtain reliable results in both simulations, the advised list for normal convergence is the same (figure 9).

Figure 9 - Spyro convergence advised fixed list

The more unfixed estimation parameters inputted in Spyro, the better and faster the convergence behaviour. Starting with the process gas side, hydrocarbon and dilution steam flow rates and temperatures need to be fixed (figure 6). Other way to fix dilution steam flow rate is by fixing the dilution ration in mixer model where this compound is mixed with the hydrocarbons. The vapour fraction of the hydrocarbon feed must be verified along with the correct properties for Spyro to perform the calculations. At last the pressure can be fixed at coil outlet, in transfer line volume or in the product exit. Pressure drop for all junctions like mixer, coil inlet, outlet and more must be fixed too. One of the cracking severity parameters must be fixed. It can be the coil outlet temperature, key component conversion or the P/E ratio in the effluent.

In the high-pressure steam side, BFW inlet temperature is fixed leaving the flow rate unfixed. Pressure can be fixed in the HPI effluent or in the steam drum along with the blow down factor or flow rate, the heat loss fraction or enthalpy loss and external heat. If the TLX shell side is connected to the steam drum the mode selected is steam drum operation and if it works as BFW heater the mode selected is BFW heater operation. All the properties tables selected are VDI steam tables and estimated pressures, flow rates and temperatures are important for the convergence behavior. If it is the case, BFW quench added to temperate the HPI produced must have flow rate, temperature and pressure fixed and the mixer operates in minimum pressure mode with only pressure drop fixed. In the firebox model, the pressure and the excess air for radiant wall burners are fixed unfixing the flow rate of fuel gas and air. In the other way, temperature and pressure of both compounds are fixed not forgetting to give proper estimate parameters for the flow rates. The banks inside the convection section need to have all tuning parameters fixed and unfixed real values for the inlet and outlet temperatures of the process and flue gas side. With all parameters correctly inputted in Spyro and all connections made (appendix IV), becomes possible to converge the full furnace simulation and compare the model output calculations with real cases. Values for the output of the model normal convergence are represented in table 3. The output is different from reality, representing this simulation the ideal operating conditions. The percentage of deviation was estimated by the equation 2.2.1.

%𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 = (-./0123456)3456

∗ 100 (2.2.1) A key performance indicator (KPI) including the sum of COT, BFW and fuel gas flow rates percentage of deviation was defined to compare the normal convergence and the tuning results described in the next chapter. Table 3 – Deviation from the ideal convergence of Spyro from real operating values.

F1003 COT (ºC) BFW (kg/h) Fuel (kg/h)

Real 851.60 29616.29 3444.49

Spyro 838.00 28380.34 3018.03

% Deviation 1.60% 4.17% 12.38%

F1006 COT (ºC) BFW (kg/h) Fuel (kg/h)

Real 852.00 29764.43 3199.08

Spyro 836.00 27403.75 2961.49

% Deviation 1.88% 7.93% 7.43%

Besides the deviations shown in table 3, there are more like for example in temperatures leaving the convection section bundles. That deviations became insignificant when the tuning to the real furnace is made. The KPI indicator, for the normal convergence is shown in table 4. Table 4 - KPI indicator for normal convergence of the model.

F1003 COT (ºC) BFW (kg/h) Fuel (kg/h) KPI

% Deviation

1.60% 4.17% 12.38% 18.15%

F1006 COT (ºC) BFW (kg/h) Fuel (kg/h) KPI

% Deviation

1.88% 7.93% 7.43% 17.24%

6

3. Model Tuning The main task of this thesis was to evaluate all the feedstocks used for the steam cracking at Repsol Polímeros situated in Sines as well as to converge all the real furnaces in Spyro convection model. After the correct properties given by Aspen Hysys v8.4 for hydrocarbon feedstocks entering the furnace in liquid state and the analysis made by Repsol laboratory (appendix II), characterization in the Spyro model can be made. After the normal convergence, the model had to be tuned to real condition operation values. To achieve this, the outside HTC factors were calculated by the model by fixing operation values from a period of time where they were constant (appendix V). Having all the furnaces converge to real operating values, the outside HTC factors were fixed and then two sensitive analysis were made. First the hypothesis of revamping bundles by changing the factors to 1, one by one and evaluate the gains and losses. The second was to analyse the impact of adding or reducing 1 metric tonne of ‘’protection steam’’ in the hydrocarbon coils maintaining the ratio steam/hydrocarbons. In other words, if the 1 metric tonne is reduced in the ‘’protection steam’’, one is added to the dilution steam. A problem occurred in the furnaces with liquid feedstocks entering the convection section. Spyro recognized the ppf. file but when the import was made the properties didn’t appear in the model. With the help of Technip, Eng. André Vilelas and Eng. Jorge Fernandes, we reached the conclusion that the problem was in the software installed in Repsol and for that reason the goals were made only for the furnaces with propane as feedstock (F1003 and F1006).

3.1. Outside HTC Factors Determination In a way to achieve the state of each bundle from the convection section an estimation of the outside Heat Transfer Coefficient (HTC) factor was made. This outside HTC factor is related with problems in the bundle coatings like corrosion and deposition of dirt present in the flue gas which is drown by the negative pressure induced by the fan. Some of these bundles are composed by fins (figure 10) to increase the heat transfer area. They are spaced between each other and with the passing days, months and years of operation, dirt particles accumulate between the fins which difficult the heat transfer from the flue gases. Notice that in bare tube, deposition of dirt particles is more difficult to occur being the correction the main parameter in the outside HTC factor.

Figure 10 - Cut fins and full fins (Repsol's case)

After the normal convergence of the model, temperatures of the process side and temperature of the flue gas leaving the convection section must be fixed (appendix V). In this case, operating values of the temperatures leaving the bundles have to be inputted in the model and fixed, unfixing the outside heat transfer coefficient factors. After the simulation run of the full furnace, Spyro gives the factors for each bundle (figure 11).

Figure 11 - Outside HTC factors calculated by Spyro

In this task two obstacles appeared. First, the operation don´t have values of the process temperature leaving the feed pre-heater (FPH) and dilution steam super heater (DSSH) bundles due to the absence of local temperature indicators. To overcome this, a sensitive analysis was made with this two factors ranging from 0.6 to 0.9 in both bundles. The boundaries of the gap were chosen by typical values and having in account that these factors cannot be higher than 1 and lower than 0. The other obstacle was the fact that by fixing the temperature of the flue gas leaving the convection section to the atmosphere, we cannot fix the amount of oxygen in this feed. By fixing the two variables, Spyro delivers a message to the user saying that the model has one additional variable fixed and cannot perform the run. Because the furnaces aren’t fully isolated, parasite air enters through cracks between the burners and pipes with the housing due to the negative pressure induced by the fans. However, the values of that percentage of oxygen given by Spyro were not far of the real ones, the temperature of the leaving flue gas seems the variable more important to fix in the outside HTC factors test. A detailed scheme to overcome this task is represented in the figure 12.

Figure 12 - Tuning scheme for real outside HTC factors calculation

Values of fuel gas and boiler feed water flow rates as well as COT, were compared to the real operation ones to choose the outside HTC factors who minimize that differences between the Spyro and the real ones (table 5). That deviation was estimated by the equation 2.2.1.

Table 5 - Outside HTC factors calculated by Spyro for furnace F1003.

DSSH/FPH

Deviation HTC HPSSH-II HPSSH-I ECO

0.6/0.6 8.363% 0.702 1.073 0.955 0.982

0.6/0.7 8.364% 0.658 1.036 0.913 0.982

0.6/0.8 8.365% 0.617 1.002 0.876 0.982

0.6/0.9 8.366% 0.509 0.924 0.794 0.978

0.7/0.6 8.363% 0.686 1.059 0.939 0.982

0.7/0.7 8.364% 0.641 1.022 0.897 0.982

0.7/0.8 8.365% 0.601 0.989 0.861 0.982

7

0.7/0.9 8.366% 0.564 0.960 0.829 0.982

0.8/0.6 8.363% 0.671 1.047 0.925 0.982

0.8/0.7 8.364% 0.627 1.010 0.884 0.982

0.8/0.8 8.365% 0.587 0.978 0.848 0.982

0.8/0.9 8.366% 0.550 0.948 0.817 0.982

0.9/0.6 8.363% 0.658 1.036 0.913 0.982

0.9/0.7 8.364% 0.614 1.000 0.873 0.982

0.9/0.8 8.365% 0.601 0.989 0.861 0982

0.9/0.9 8.366% 0.537 0.939 0.806 0.982

Given the gap with the possible values for the outside HTC factors and analysed the sum of the errors for the parameters related before, the factors chosen for furnace F1003 were 0.8 to DSSH bank and 0.8 to FPH bank. By the same method, the outside HTC factors chosen for furnace F1006 were 0.7 to DSSH and 0.9 to FPH banks. That factors along with the ones for the other banks are represented in the table 6. After the analysis was made and having in account the feasible values we can conclude that the error differs in the third decimal digits (table 5). Note that some furnaces which had recently revamps in the HPSSH I and II bundles have the outside HTC factors close to 1 like in furnace F1006 for example.

Table 6 - Real outside HTC factors for furnaces F1003 and F1006

HTC HPSSH-II

HPSSH-I

DSSH FPH ECO

F1003 0.587 0.978 0.848 0,8 0,8 0.982

F1006 0.744 1.000 0.763 0,7 0,9 0.985

To compare the real tuning case with the normal convergence of the model with all outside HTC factors equal to 1, the KPI used in chapter 2 was compared to the one calculated to the real operation case (table 7). Besides the KPI indicator only includes COT, BFW and fuel gas flow rates, the deviation in the temperatures leaving the bundles from the convection section turn zero in the tuning case. The gap between the KPI’s with the total deviation including all these parameters becomes larger and more significant.

Table 7 - Difference between the KPI's from the real and the ideal simulations.

F1003 COT (ºC) BFW (kg/h)

Fuel (kg/h)

KPI

% Deviation 1.60% 1.56% 7.69% 10.84%

F1006 COT (ºC) BFW (kg/h)

Fuel (kg/h)

KPI

% Deviation 1.76% 7.91% 4.65% 14.32%

3.2. Revamp Case-Study After the tuning of the Repsol’ cracking furnaces model using the simulation by Spyro, a new problem was posed: What would be the benefits in a possible revamp of each bundle from the convection section? To achieve an answer to that question, the outside HTC factors were changed to 1, one by one, maintaining all the others equal to the real furnace case (figure 13). Besides unfixing the process outlet temperatures and fixing the real factors described in table 4, in this case in more correct to fix the percentage of oxygen in the flue gas given by Spyro for the real simulation instead of fixing the temperature of the flue gas leaving the furnace. By that, it is

guaranteed that no temperatures are fixed letting Spyro to calculate them.

Figure 13 - Revamp scheme by changing the outside HTC factors one

by one to 1.

After this test, differences between fuel gas consumed and HPI produced from revamping cases and real furnace where measured to see which bundle causes more impact. In the furnace F1006, the case of revamping the high-pressure steam super heater II (HPSSH-II) bank has no deviations corresponding to the real outside HTC factor described in table 8. Table 8 - % of deviation from the real case parameters of revamping

cases

F1006 Fuel Gas (kg/h) HPI (kg/h)

Real Case 3049 31236

Revamp Case % Deviation HTC=1 -1.744 -2.656

HPSSH-II =1 - - HPSSH-I =1 0.352 -0.439

DSSH = 1 -0.347 -0.419 FPH = 1 -0.376 -0.453 ECO = 1 0.002 0.115

F1003 Fuel Gas (kg/h) HPI (kg/h)

Real Case 3179.729 33307.384

Revamp Case % Deviation HTC=1 -4.674 -3.400

HPSSH-II =1 0.029 -0.055 HPSSH-I =1 0.210 -0.312

DSSH = 1 -0.328 -0.258 FPH = 1 -0.766 -0.978 ECO = 1 -0.004 0.133

In table 5 the deviation is calculated with the deviation between the revamped cases from the real case (equation 3.2.1). The minus signal represents less consumption/production of the revamped cases for both parameters respectively. With these values, it’s hard to take conclusions from which bank causes more impact when revamped.

%𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 = (-./0123456)3456

∗ 100 (3.2.1) This test along with an economic evaluation make possible to conclude if a revamp to the bundles is viable. That economic evaluation was made in a simple way comparing only the fuel gas consumed and the high-pressure steam (HPI) produced prices. Values of €/kg for HPI produced and for fuel gas consumed were taken for this task [14]. Summing up, the test was made with the view to give the user the earnings or expends compared to the real case (table 9). The values represent with minus signal represents losses.

8

Table 9 - Earnings and expends from the revamp cases relative to the

real case.

F1006 €/h €/day €/month

HTC=1 10.01 240.17 7205.11

HPSSH-II =1 - - -

HPSSH-I =1 -7.03 -168.65 -5059.47

DSSH = 1 -0.25 -6.01 -180.19

FPH = 1 -0.26 -6.17 -185.24

ECO = 1 0.95 22.80 683.92

F1003 €/h €/day €/month

HTC=1 14.45 346.75 10402.36

HPSSH-II =1 -0.78 -18.81 -564.25

HPSSH-I =1 -4.87 -116.94 -3508.30

DSSH = 1 -0.40 -9.68 -290.49

FPH = 1 -1.22 -29.26 -877.66

ECO = 1 1.24 29.66 889.89

Looking at both tables 5 and 6 it is possible to take conclusions about which bundle revamp case causes more impact. Having the lowest real outside HTC factor in both furnaces F1003 and F1006, it was expected to predict that the HTC bank was the one causing more impact. This simple analysis gives Repsol the idea of a possible revamp case, which before being taken further, more detailed studies have to be done. Notice that this was only a simple evaluation test looking only for the furnace as the whole process. In the real steam cracking process, the HPI produced is not sold and the fuel gas not bought. The first one feds the turbine to generate the energy necessary to move the gas centrifugal compressor and the second one comes from the demethanizer and from the hydrogen leftover from the hydrogenation section (chapter 1).

3.3. Protection Steam Another aspect that came out was the ‘’protection steam’’ flow rate, which is added to the hydrocarbon feedstocks before entering the convection section of the furnaces. The steam is controlled manually and don’t having flow rates in the control panel, this value doesn’t enter for the calculi of the mass ratio steam/hydrocarbons. It is mixed with the hydrocarbon feedstocks to protect the coils when occurring for example a shut down. If the furnaces go down, the hydrocarbon flow rates stop automatically and the coils can be fragile be the thermic shock increasing the probability to break. Ensuring that protection steam continue to flow into the coils, that probability decreases exponentially. The flow rate in all furnaces is about 1 metric tone and that value comes by the difference of the total dilution steam fed to each furnace and the sum of dilution steam entering the coils of DSSH bundle given by the control panel. The impact of this steam was measured by a sensitive analysis (table 10), increasing 1 metric tone in the hydrocarbon coils before entering the FPH and decreasing it in the dilution steam flow rate entering the DSSH bundle and vice versa.

Table 10 - Impact on adding or reducing the flow rate of ''protection steam''.

F1006 Fuel Gas (kg/h) HPI (kg/h)

Real Case 3049.51 31236.74

Protection Steam % Error

+ 1 ton/h 0.26 0.32

- 1 ton/h -0.24 -0.29

F1003 Fuel Gas (kg/h) HPI (kg/h)

Real Case 3179.73 33307.38

Protection Steam % Error

+ 1 ton/h 0.24 0.31

- 1 ton/h -0.20 -0.28

With the information from table 7, it isn’t easy to distinguish the most favourable case. The minus signals indicate less consumption/production of fuel gas and HPI respectively. When making the same economic analysis referred in the sub chapter before, the task becomes much easier (table 11). Table 11 - Earnings and expends of adding or reducing 1 metric tonne

of ''protection steam'' relative to the real case.

F1006 €/h €/day €/month

+ 1 ton/h 0.17 4.11 123.35

- 1 ton/h -0.16 -3.93 -117.92

F1003 €/h €/day €/month

+ 1 ton/h 0.42 10.11 303.33

- 1 ton/h -0.53 -12.72 -381.6

The results given by the economic evaluated indicates that the more steam used as ‘’protection steam’’, maintaining the mass ratio steam/hydrocarbons, the more HPI is produced compensating more fuel consumed. In this economic evaluation, no impact calculation was done by electricity generation lost at Power Plant and the extra feed to the Cracker. Being above the DSSH bank, the FPH one in the favourable case has more flow rate circulating in the coils. To pre-heat that hydrocarbons/’’protection steam’’ mixture, more fuel gas needs to be consumed producing more high pressure steam. Besides the protection criteria, the more steam into the coils, more flow rate passing in the transfer line exchangers, more HPI produced. By feeding more high pressure steam to the turbine, more power is available to rotate the compressor shaft. Of course, if you have more flow rate in the coils you consume more fuel gas in the burners to achieve cracking temperatures. The coils were design with a maximum flow rate allowed and Repsol Polímeros work in their maximum to produce more HPI in the transfer line exchangers and generate more power to the compressor insuring the protection of the coils and the facility to evaporate the liquid feedstocks in the FPH bundle.

4. Conclusions In this work, a steam cracking simulation model called Spyro Suite 7 was used, which is a necessary toll used by the majority of steam cracking producers worldwide. With the option of yield prediction, using only the radiation coils model, it is a fundamental tool to predict which are the required operational conditions to be taken in real time cases. It can be used to simulate the full furnace models. Before the convergence of the full furnace, a thermodynamic evaluation of the feedstocks had to be made in order to characterize them properly in Spyro. For gaseous feedstocks it was concluded that with the detailed mass or volume composition and selecting the

9

gaseous feedstock properties, the model was correct and ready to be connected to the others to perform the full furnace simulations. Some difficulties were found with the feedstocks entering the furnace in liquid phase. Spyro properties don’t describe well naphtha’s used at Repsol’s steam cracking. New properties had to be added with the help of Aspen Hysys v8.4 and correctly written in a ppf. file readable by Spyro. The ppf. file was recognized but the properties didn’t appear in the model enabling to simulate furnaces with liquid feedstocks. Having all feedstocks thermodynamically analysed, it was only possible to converge the furnaces (F1003 and F1006) using as feedstock propane to the real operating conditions. Englobing the corrosion and deposition of dirt particles in the fins of the bundles of the convection section a factor designated as outside heat transfer coefficient factor was taken in account (chapter 4.1). The chosen outside HTC factors for furnaces F1003 and F1006, besides being in the gap of possible values, describes well the real state of each bank of the convection section as well as the behaviour of the real time operating conditions. Being, the HTC bank, the first bundle after the radiation box, it was expected to have the lowest outside HTC factor in both furnaces. This is due to the first bank of the convection section being more able to suffer from dirt deposition and corrosion than the higher ones. The real outside HTC factors from furnace F1006 were higher than the ones from F1003 as a result of suffering recently a revamp in the HPSSH I and II banks and cleaning process in the others. The main key performance indicator (KPI) containing the sum of the deviations from COT and BFW and fuel gas flow rates between the simulations and the real case was used to prove the convergence of the model to real operating values. The values of this parameter in the normal convergence of the furnace was 18.15% and 17.24%, decreasing when the furnace in tuned to the real operating values to 10.84% and 14.32% for furnaces F1003 and F1006 respectively. To study a possible revamp case of each bundle, a sensitive analysis was made to calculate the savings in fuel gas and HPI flow rates. From those analyses, the high temperature convection (HTC) bundle, being the one with the lowest outside HTC factor, represented the highest savings when revamped. In other words, less HPI compensates less fuel consumed turning it into profit. Besides the high temperature convection (HTC) bundle, the economizer (ECO) bundle also represented profits when revamped. To take decisions in the remaining bundles of the convection section, more detailed studies must be done to prevent having financial losses. The economic evaluation was made in a simplified way, establishing the balance only in the furnace and not for the whole process. Furthermore, the ‘’protection steam’’ flow rate was taken into question. Maintaining the mass ratio steam/hydrocarbons, another sensitive analysis was made by increasing/decreasing one metric tonne per hour of ‘’protection steam’’ and by decreasing/increasing it in the dilution steam. The benefits were analysed along with an economic evaluation like in the revamp case study. The favourable scenario was reached when one metric tonne per hour was increased in the ‘’protection steam’’ flow rate. The high-pressure steam produced compensates the higher amount of fuel consumed, leading to a higher flow rate into the turbine generating more power to the effluent centrifugal compressor. It indicates that working in the maximum flow rate allowed by the specifications of the coils is the better scenario to operate. To sum up, Repsol now has their furnaces F1003 and F1006 simulated in Spyro for other studies and yield simulations. Relatively to the revamping cases, more detailed studies need to be done to have all economic parameters and differences in the real operation and Spyro parameters to decide further actions.

4.1. Future Work To have all of Repsol’s furnaces simulated in Spyro and able to converge, the properties, describing feedstock entering in the convection section in liquid state, import problem has to be resolved internally with Technip’s help in a future study. When resolved, the

furnace tuning sequence is the same that the one described in chapter 4. For different liquid feedstocks, new properties have to be simulated with Aspen Hysys v8.4 and written after in a new ppf. file readable in Spyro. A global economic evaluation must be made to prove the investment in a possible bank revamp case in the convection section, as well a more detailed comparison between real case and simulation case parameters.

5. References [1] – ‘The history of Spyro Simulation Software’, www.spyrosuite.com/pyrotec/hystory/ , (online, Accessed: 2017-05-15) [2] – E. Ranzi, M. Dente, S. Plerucci and G. Blardl, ‘Initial Product Distributions from Pyrolysis of Normal and Branched Paraffins’, in Industrial & Engineering Chemistry Fundamentals, 1983 [3] – Marco W.M. van Goethem, Florian I. Kleinendorst, Cor van Keeuwen and Nils van Velzen, ‘Equation-based Spyro Model and Solver for the Simulation of the Steam Cracking Process’, in Computers & Chemical Engineering, 2001 [4] – Marco W.M. van Goethem, Florian I. Kleinendorst, Cor van Keeuwen, Nils van Velzen, Mario Dente and Eliseo Ranzi, ‘Equation Based Spyro Model and Optimiser for the Modeling of the Steam Cracking Process’, 2002 [5] – M.W.M. Van Goethem, S. Barendregt, J. Grievink, J.A. Moulijn and P.J.T. Verheijen, ‘Towards Synthesis of an Optimal Thermal Cracking Reactor’, in Chemical Engineering Research and Design, 2008 [6] – Marco W.M. van Goethem and Peter J.T. Verheijen, ‘Integration of Symbolic Modelling within an Equation Based Flowsheet Package for Steam Pyrolysis’, in Simulation Modelling Practice and Theory, 2013 [7] – Marco W.M. Van Goethem, Simon Barendregt, Johan Grievink, Peter J.T. Verheijen, Mario Dente and Elisio Ranzi, ‘A Kinetic Modelling Study of Ethane Cracking for Optimal Ethylene Yield’, in Chemical Engineering Research and Design, 2013 [8] – Pyrotec division of Technip Benelux BV, ‘Spyro Suite 7 Training manual’, Repsol archive, 2017 [9] – Oliveira Gomes, ‘Considerações Gerais Sobre Pirólise’, Repsol archive, 2017 [10] – ‘Olefins Production: olefins by steam cracking’ , https://mol.hu/images/pdf/A_MOL_rol/tvk-rol/tarsasagunkról_roviden/egyetemi_kapcsolatok/debreceni_egyetem/oktatasi_anyagok/bemutatok/OLEFINS%20PRODUCTION.pdf , (online, Accessed: 2017-06-10) [11] – Heinz Zimmermann and Roland Walzi, ‘Ethylene’, in Ullmann’s encyclopedia of industrial chemistry, 2012 [12] – António Almeida, ‘Descrição simplificada do processo’, Repsol archive, 2017 [13] – ‘Enthalpies of formation’, http://webbook.nist.gov/chemistry/ , (online, Accessed: 2017-08-09) [14] – Jorge Fernandes, ‘Consumers prices’, Repsol archive, 2017 [15] – Ifp, ‘Petrochemicals Olefins and Aromatics, Repsol archive,

2017