The first stripping stage of a SWS unit consists of a distillation column ‒ a H₂S stripper column ‒ fed by pre-heated sour water stream at the top. A reboiler provides energy to generate stripping steam to remove the pollutants. An appropriate operation pressure of the stripper column ensures that H₂S is preferably removed in the first stripping stage while NH₃ is maintained in the liquid stream at the column bottom. The effluent gases at the top of the column are cooled in a condenser and the liquid and vapor phases are separated in the top drum. The H₂S rich gas from top drum is sent to SRU, and the liquid phase returns to the sour water feed tank. To reduce steam consumption in the reboiler, a heat exchanger is used to heat the sour water stream and to cool down the bottom stream from the stripper column.

SWSU-1 was evaluated by simulation with Aspen HYSYS^®. A flowsheet of the first stripping stage is shown in Figure 2. This work consider sour water stream to be composed of water, NH₃ and H₂S. Considering a sour system, involving equilibrium of ionic species, the thermodynamic model Sour PR of the simulator was adopted, which combines the Peng-Robinson equation of state and the Wilson’s API Sour Model^† for handling sour water systems. In the Sour PR model, the K-values for the aqueous phase are calculated using Wilson’s API-Sour method to account for the ionization of H₂S, CO₂ and NH₃ in the aqueous water phase.

Figure 2.

Flowsheet of the first stripping stage simulated in Aspen HYSYS^®

Simulation runs considered a H₂S stripper column, with the following variables defined: sour water feed composition, pressure and temperature, stripper column operating pressures, top drum pressure and temperature. Reboiler duty (Q_reb) was varied and process responses-H₂S recovery and NH₃ content in the acid gas – were analyzed. Table 1 shows the simulation inputs of the base case. Streams and equipment items refers to Figure 2.

Example response curves are shown in Figure 3a and Figure 3b, evidencing the separation limit point of the H₂S stripper column, which represents the maximum H₂S recovery without exceeding the NH₃ limit in the acid gas to SRU. In Figure 3a, H₂S recovery is defined as the percentage of the H₂S originally in the sour water stream that is found in the acid gas stream to SRU. H₂S recovery increases linearly with reboiler duty increase, until a limit where it assumes an asymptotic behavior. Parallelly, the NH₃ content in the acid gas stream is very low and nearly independent of the reboiler duty, in low to moderate duties, until a threshold is reached where a steep ascent occurs. The change in behavior of both curves occurs at the same value of reboiler duty. Below this point, it is possible to obtain an acid gas stream rich in H₂S and with very low NH₃ contents. Above this duty value, a large amount of NH₃ is sent to the acid gas stream with increased amount of water. This point is hereinafter defined as the separation limit point of the H₂S stripper column and identified by the “^*” symbol.

Figure 3.

H₂S recovery, NH₃ content in the acid gas (a) and water flowrate in the top stream (b)

The separation limit point also presents a significant change in the behavior of the equilibrium stages temperatures of the H₂S stripper column, as shown in Figure 4. As the equilibrium stage 5 exhibits the major temperature increase at the separation limit point, it was chosen as a sensitive stage to characterize the transition from normal operation to the operational limit. Equilibrium stages 10 and 11 present the reboiler inlet and outlet temperature, respectively. The temperature of stage 11 is constant along the evaluated range of reboiler duty, while the reboiler inlet temperature increases progressively with increasing reboiler duty. Thus, the difference between the inlet and outlet reboiler temperatures, and the temperature of the sensitive stage could be used for monitoring process performance.

Figure 4.

H₂S stripper column – outlet stage temperatures

This study was developed applying Response Surface Method (RSM) to obtain empirical models to characterize the separation limit point of the H₂S stripper column. RSM consists of a set of mathematical and statistical techniques that are based on fitting empirical models to experimental data obtained from an experimental design [23]. For this purpose, the following Factors (variables that can be changed independently of each other) and Responses (measured values from experiments with the H₂S stripper) were considered.

• Operating pressure of the H₂S stripper column (P);

• Molar ratio H₂S:NH₃ in the feed of the sour water stripping unit (H2SNH3);

• NH₃ content in the feed of the sour water stripping unit (xNH3);

• Bottom and feed temperatures difference of the H₂S stripper column (dT_bf).

• Reboiler duty divided by the molar flow rate of the sour water unit feed at the separation limit point (Q^*_reb), herein named ‘reboiler duty/feed ratio’, which represents the amount of energy required to reach the operating limit point. This response has a practical importance for monitoring the unit and delimits the region considered as normal operation;

• Maximum H₂S recovery (RecH2S*), defined as percentage of the H₂S originally in the sour water unit feed that is found in the acid gas stream to SRU, at the separation limit point;

• NH₃ molar fraction in the acid gas to SRU at the separation limit point (yNH3*);

• Difference between the inlet and outlet reboiler temperatures at the separation limit point (dT^*_reb);

• Stage 5 temperature at the separation limit point (T^*_sens).

A three level factorial experimental design was carried out considering the levels of factors. Combining the three levels of each factor listed in Table 2, a set of 81 experimental points was generated. For each point, the curve of the response variables versus reboiler duty was constructed using Aspen HYSYS^® process simulator, according to the flowsheet simulation presented in Figure 2, to obtain the responses at the separation limit point of the H₂S stripper column. As the experimental data was obtained from a process simulator, the experiments are referred to as pseudo-experiments and the results as pseudo-experimental. Also, because of the use of a process simulator to obtain experimental data, experimental replicates were not considered.

First and second order RS models, RS O(1) and RS O(2), respectively, were tested specifying a 95% confidence level for the estimation of model parameters. The use of logarithmic function in the model factors and responses was also evaluated. The general form of the models tested is given in eqs. (1-5).

RS O(1) model:

RS O(1) model with logarithmic function in the factors:

RS O(2) model:

RS O(2) model with logarithmic function in the factors:

RS O(2) model with logarithmic function in the factors and responses:

After model parameters estimation, selection of the best RSM for each response was based out considering the total quadratic Sum of deviation (SoS) between observed (Ȳ_i) and predicted values (Y_i), as shown in eq. (6). If the logarithmic function is applied to the response, the total quadratic sum of deviation assumes the form shown in eq. (7):

The first stripping stage of a SWS Unit (SWSU-1) was dynamically simulated in Aspen HYSYS^® process simulator, with flowsheet shown in Figure 5.

Figure 5.

Flowsheet of the first stripping stage simulated in Aspen HYSYS^® dynamic

The objective of the reboiler duty control is to ensure that the H₂S stripping column operates with H₂S recovery greater than 90%, but without exceeding the operational limit. Typically, the steam flow to the reboiler is manipulated by a ratio control between the reboiler duty (i.e. steam flow rate) and the flowrate of sour water to the H₂S stripping column, as shown in Figure 6. The H₂S recovery is not a real-time measured variable, and the set-point adjustments of the ratio control are performed by operators, based on laboratory analysis. However, this control is not efficient when disturbances occur in other input variables, such as the stripping column temperature.

Figure 6.

Ratio control for H₂S stripping column reboiler duty [R_m = reboiler duty (u)/sour water flowrate (d)]

Alternatively, the proposed RS model of maximum reboiler duty was used in the dynamic simulation for generating the set-point for the ratio control, as shown in Figure 7. The advantage of this control is that it considers the main disturbances in the input variables, which are the RS model factors. The factors ‘pressure and difference between the bottom and feed column temperature’ are measured online, while the factors ‘NH₃ content in the sour water stripping unit feed and H₂S/NH₃ in the sour water stripping unit feed’ could be measured online or made available by laboratory analysis. However, since the unit has a feed tank with high residence times, usually at least 24 hours, the composition changes are very slow. Thus, the RS model can be used even when the feed composition is not analyzed online.

Figure 7.

RS model generating a set-point to the ratio control for reboiler duty

Table 3 shows the SoS values of RS models tested for each response. Considering SoS values shown in Table 3, the first order RS models did not adequately fit pseudo-experimental data. The pseudo-experimental data from responses Q^*_reb and dT^*_reb were best fitted by second order RS models, while, for responses RecH2S* and T^*_sens, logarithmic function in the factors yielded better results.

Graphics in Figure 8 show the pseudo-experimental data versus data obtained from RS models selected for these responses, where the selected RS models satisfactorily fit the pseudo-experimental data. For the response yNH3*, logarithmic function was applied to the factors and response for better results (Figure 9). In this case, the SoS calculated for RS O(2) used eq. (7) and could not be compared with the SoS obtained for the others RS models tested [SoS from eq. (6)].

Figure 8.

Pseudo-experimental data versus RS models predictions: Q^*_ref (a); RecH2S* (b); dT^*_reb (c) and T^*_sens (d)

Figure 9.

Pseudo-experimental data versus data obtained for yNH3* RS models: RS O(2) model (a) and RS O(2) with logarithmic function in the factors and response (b)

Thus, the RS model for responses Q^*_reb and dT^*_reb assumes the general form shown in eq. (3). Eq. (4) shows the general RS model for responses RecH2S* and T^*_sens , while eq. (5) refers to response yNH3*. The equation parameters are presented in Table 4.

The RS models obtained were validated using a separate data set composed of 21 points for each response, also generated by simulation. The responses obtained by the RS models are presented in a normalized form and plotted against the values obtained by simulation in Aspen HYSYS^® (Figure 10). The points of all the responses are very close to the diagonal line, indicating that the RS models show an excellent predictive capability.

Figure 10.

RS models responses versus responses from Aspen HYSYS^® simulation

RS models allow the estimation of process responses as a function of the selected factors, and permits a sensitivity analysis of the effect of each factor on the responses studied.

Figure 11 shows the response surfaces for Q^*_reb. Each surface is presented with two factors, while the remaining factors are maintained at their mean values. Figure 11a shows that the effect of the (H₂S/NH₃) factor on the reboiler duty at the operation limit point depends on the NH₃ content in the sour water (xNH3) and vice versa. For lower xNH3 values, the (H₂S/NH₃) practically does not affect Q^*_reb. However, increasing the NH₃ content in sour water, as (H₂S/NH₃) increases, more energy is needed in the reboiler to achieve the separation limit point. Nevertheless, in both cases, the variation observed in Q^*_reb is lower than 5% in the ranges evaluated for all factors. Figure 11b shows a small effect of the factor P over Q^*_reb, where an increase in the operation pressure leads to an increase in Q^*_reb. The factor dT_bf presents the most important effect on Q^*_reb. This is due to the strong impact of the column feed temperature on the column’s energy balance. The higher the factor dT_bf, the lower the column feed temperature and, consequently, the larger the amount of energy required to reach the separation limit point. In this case, the variation in Q^*_reb is about 20%.

Figure 11.

RS for Q^*_reb: factors xNH3 and (H₂S/NH₃) (a) and factors dT_bf and P (b)

In Figure 12a, the influence of dT_bf on RecH2S* is negligible, whereas lower NH₃ contents in the sour water feed lead to higher maximum recovery of H₂S. In Figure 12b, an increase in the factors (P) and (H₂S/NH₃) increases the RecH2S*, and the effect of these factors is much stronger than those observed in Figure 12a.

Figure 12.

RS for RecH2S*: factors xNH3 and dT_bf (a) and factors (H₂S/NH₃) and P (b)

In Figure 13a, (H₂S/NH₃) has no impact on the NH₃ content in the acid gas at the separation limit point (yNH3*). On the other hand, dT_bf has a direct effect, where higher values of dT_bf leads to a reduction in the NH₃ content in the acid gas. Indirectly, a lower H₂S stripper column feed temperature leads to lower concentrations of NH₃ in the acid gas. Figure 13b shows that the factor operation pressure has small influence on the response yNH3* and the most important factor on this response is the concentration of NH₃ in the sour water.

Figure 13.

RS for yNH3*: factors dT_bf and (H₂S/NH₃) (a) and factors xNH3 and P (b)

Figure 14a and Figure 14b shows that the factors (P) and (H₂S/NH₃) have almost no influence on response dT^*_reb . Otherwise, the higher the dT_bf and the higher the xNH3, the higher the value of dT^*_reb.

Figure 14.

RS for dT^*_reb: factors dT_bf and P (a) and factors xNH3 and (H₂S/NH₃) (b)

Figure 15a and Figure 15b shows that the effect of the factor (H₂S/NH₃) on the response T^*_sens is negligible and the influence of the factor dT_bf is very small. Variations of the NH₃ content in the sour water over the studied range can lead to variations up to 10 °C in T^*_sens. Nevertheless, the operating pressure is the most important factor for this response, as the higher operating pressure leads to higher sensitive stage temperatures, reaching variations up to 20 °C in the evaluated range of pressure.

Figure 15.

RS for T^*_sens: factors xNH3 (a) and P and factors dT_bf and (H₂S/NH₃) (b)

The results of the RS model for responses Q^*_reb, RecH2S* and dT^*_reba were compared against data from a SWS unit installed in a Brazilian oil refinery. As the responses refer to the separation limit point, Q^*_reb is the maximum value of reboiler duty when the H₂S stripper column operates in a normal condition.

Data period of the SWS plant corresponds to 65 hours of typical operation. As shown in Figure 16a, the reboiler duty from SWS plant (Q_{reb_plant}) was under the maximum reboiler duty for normal operation provided by RS model (Q^*_reb). Yet, the difference between the inlet and outlet reboiler temperatures (dT_{reb_plant}) is supposed to be larger than its value at the separation limit point obtained by RS model dT^*_reb. This behavior is also observed in Figure 16a. Moreover, as previously shown in Figure 4, at normal operation, the SWS top temperature is very close to the feed temperature (dT_{bf_plant}).

Figure 16.

RS model results compared to plant data – typical operation

Figure 16b shows that the H₂S recovery (RecH2S_plant*) is minor than the maximum H₂S calculated by the RS model (RecH2S*), as expected. The NH₃ content in the acid gas to SRU is not measured in the plant, but the RS results (yNH3*) shows a low and stable value.

Figure 17 shows another sample of plant data achieved from a period of 14 hours when disturbances to the H₂S stripper column leads to an operating condition that crosses the separation limit point. As an evidence of this situation, the difference between the top column temperature and the feed column temperature (dT_{bf_plant}) reaches values up to 50 °C, against values around a few degrees in normal operation. During this period, the difference between the inlet and outlet reboiler temperatures (dT_{reb_plant}) assumes nearly the limit value calculated by RS model (dT^*_reb) (Figure 17a). It is interesting to note that it occurs when reboiler duty from plant data became higher than the maximum reboiler duty for normal operation provided by RS model (Figure 17b). Clearly, RS model has a potential use as software-based sensor to support plant operation.

Figure 17.

RS model results compared to plant data – disturbance

As described previously, the reboiler duty control of the H₂S stripping is typically performed by a ratio control of the vapor flow to the reboiler and the sour water flow to the column (R_df). In the dynamic simulation, the variable reboiler duty was adopted instead of the vapor flow to the reboiler, for simplicity.

To evaluate the predictive capacity of RS models in the dynamic simulation, a sequence of two steps was performed in the set-point of the traditional ratio control (total simulation time horizon of 309 minutes). Figure 18 shows the RS model factors and the R_df ratio. Graphics in Figure 19 show the behavior of the RS model responses and the dynamic response of the process. In the first step, Q_reb is less than Q^*_reb, indicating that the H₂S stripper column is operating under the operating limit point. In the second step, Q_reb is greater than Q^*_reb, and the process goes through the operating limit point. At this point, the NH₃ content in the acid gas (yNH3) obtained by simulation is higher than the limit value predicted by the RS model (yNH3*), as expected. The H₂S recovery (RecH2S) reaches the maximum value predicted in the RS model (RecH2S*) and the difference between the inlet and outlet reboiler temperatures (dT_reb) reaches the minimum value, predicted by the RS model (dT^*_reb). Thus, it is observed that the RS models present a great capacity to predict the main responses of the process.

Figure 18.

RS model factors and ratio between reboiler duty and H₂S stripper column feed

Figure 19.

RS model responses and dynamic simulation responses: reboiler duty per molar flowrate of the sour water feed (a); H₂S recovery (b); NH₃ in the acid gas stream (c) and difference between the inlet and outlet reboiler temperatures (d)

Next, a sequence of three steps was performed in the set-point of the H₂S stripper column feed temperature controller, +1 °C, +1 °C and −2 °C. An increase in the feed column temperature leads to a reduction in the variable dT_reb. Figure 20 compares the process response in two situations, with the conventional ratio control (a) to (e), considering a total simulation time horizon of 600 minutes, and with the response surface model generating the set-point for the ratio control (f) to (j), in a total simulation time horizon of 1,000 minutes.

Figure 20.

Traditional ratio control: RS model factors (a); steam to feed ratio (b); H₂S recovery (c); NH₃ in acid gas (d); differential reboiler temperature (e), RS model giving a set-point for the ratio control: RS model factors (f); steam to feed ratio (g); H₂S recovery (h); NH₃ in acid gas (i) and differential reboiler temperature (j)

It can be observed that, as expected, the disturbance in the column feed leads to changes in the process response variables, which are not corrected by the conventional ratio control. There is no change in the variable reboiler duty (Q_reb). In the second step, the process goes to a condition over the operating limit point. Therefore, a great increase in the NH₃ content in the acid gas (yNH3) occurs. It could lead to disturbances in the SRU process or to deviation of the acid gas stream from SRU to flare, increasing SO_x plant emissions. There was also a large increase in the temperature of the sensitive column stage.

On the other hand, the use of the RS model to generate a set-point for the ratio control was very efficient, keeping the process stable throughout the all period of disturbances in H₂S stripper column temperature. The variable Q_reb was adjusted according to the value calculated by RS model (Q^*_reb), multiplied by a factor of 0.993, to keep the system slightly below the separation limit point. There was no significant increase in the NH₃ content in the acid gas, and H₂S recovery remained stable.

The new control strategy developed in this work has the potential to minimize SO_x emissions in two ways. Firstly, by allowing the H₂S stripper column to operate more steadily, it makes possible to operate closer to the separation limit point, with higher H₂S recovery. The second way is by reducing the deviations of the acid gas from SRU to flare.

As an example of this potential of emission reduction, consider as a basis of comparison, the normal operation of a SWSU, where 90% of the H₂S in the sour water feed is sent to SRU and only 10% of H₂S, which is stripped in the second stage, resulting in SO_x emission. The sour water flow of the base case previously described is also considered. Table 5 presents the potential of reducing SO_x emissions, calculated as Sulfur dioxide (SO₂), in these two situations.

The increase of one percentage point in H₂S recovery implies in a 10% reduction in SO_x emissions. When the deviation of the acid gas to flare occurs, the entire H₂S inventory of the SWSU becomes SO₂ emission. Thus, a control strategy that allows more stable operation of the plant has the potential for reduction of SO_x emissions.