Intensification of slow reversible chemical transformation: carboxylation of resorcinol as a case study

2.1 Materials

Resorcinol (Acros, Belgium, 98%) and KHCO₃ (Applichem, Germany, pure Ph. Eur. USP) were used, as supplied without further purification. The solutions were prepared as required with deionized water. Diluted solutions of resorcinol (Acros, Belgium, 98%), β-resorcylic acid (VWR, USA, for synthesis) and γ-resorcylic acid (VWR, USA, for synthesis) in hydrochloric acid 0.1 m [Titrisol or dilution of HCl 25% Merck (Switzerland), for analysis] were used for the calibration of the UV-Vis spectrophotometer.

2.2 Continuous milli-reactor

To demonstrate the feasibility of the concept, the reaction was carried out under high P and T in a milli-reactor containing static mixers (Figure 3). In the present study, the SMXS and SMXL mixer types were used, as they enable efficient mixing and heat transfer under laminar flow [20, 21].

Figure 3

Milli-reactor containing static mixers.

The milli-plant is composed of two main parts. The first part was designed to achieve a rapid heating and mixing of the reaction mixture at the reactor inlet. It is made of 20 jacketed SMXS mixer elements (Sulzer Chemtech, Switzerland) within a tube of 3 mm diameter, for a total length of 137 mm. The second part consists of jacketed SMXL mixer elements (Sulzer Chemtech, Switzerland) within a tube of 6 mm diameter. This part has a length of 1000 mm and provides a good mixing under laminar flow (characteristic mixing time 0.3 s) and a sufficient residence time to complete the reaction. The connection of the different elements is insured by clamping flanges (Franco-Montan, France).

As the reaction is very slow under ambient conditions, premixed reactant solutions were pumped into the milli-reactor with an HPLC pump (Waters 510, USA). The inlet concentrations of resorcinol and KHCO₃ were fixed at 0.4 m and 1.2 m, respectively. Oil was circulating from an oil bath (Haake N3 digital, Germany) into the double mantel and maintained at a constant T of 433 K. After the reaction section, the mixture was cooled down to stop the reaction (quenching section). A P of 21 bar was maintained by a fine metering valve (Swagelok, USA), to avoid solvent and carbon dioxide evaporation from the reaction mixture, in both the reaction and quenching section. In the product separator, P was released and the reaction mixture was collected. The samples were diluted in HCl 0.1 m and analyzed by UV-Vis spectrophotometry.

2.3 Analysis

The quantification of resorcinol, β-resorcylic acid and γ-resorcylic was based on absorption measurements at three wavelengths: 275, 295 and 314 nm [22]. The absorbances were recorded with a UV-Vis spectrophotometer (Perkin Elmer Lambda 20, USA) against a blank solution made of HCl 0.1 m.

After analysis, the concentrations of the three compounds were calculated by solving the following linear equation system.

The mean error of the method is less than ±8% [23]. The main advantage of this method is that the three compounds can be quantified quickly, simultaneously and quantitatively in the reaction mixture.

2.4 Data treatment

The concentrations of resorcinol obtained with the equation system [Eq. (2)] were used to calculate the conversion X (constant fluid density):

where C_1,0 is the initial concentration and C₁ the concentration at time t.

The methodology applied for PI is based on the optimization of the specific performance L_P,V of a continuously operated reactor:

where X=conversion, τ=residence time, S=selectivity

3.1 Determination of the optimal new process window

Recently, we performed a detailed kinetic and thermodynamic study of the carboxylation of resorcinol in aqueous solutions using a micro-plant [23]. In this reaction, resorcinol (1) and carbon dioxide are the two reactants where carbon dioxide is formed in solution by dissociation of KHCO₃. We established that the partial P of carbon dioxide has no effect on the reaction behavior for concentrated KHCO₃ solutions (C_KHCO3,0/C_1,0>1). It was also shown that the formation of γ-resorcylic acid (3) is negligible at short residence times, τ<30 s. Based on these experimental results, for the first time a reliable kinetic and thermodynamic model has been proposed:

where r is the reaction rate, k is the kinetic constant and T is the reaction temperature.

Using this model, the residence time τ needed to reach a given conversion which can be calculated. It is important to note that in this case, the maximum conversion is limited by the equilibrium. Moreover, as the reaction is exothermic, the equilibrium conversion decreases with an increase of the T while the reaction rate increases.

Figure 4 shows the evolution of the specific productivity as a function of T for a conversion of 30%. The star indicates the value obtained at conventional 373 K. As can be seen, the specific productivity goes through a maximum which corresponds to the optimal process window for the chosen diluted conditions. Under these optimal conditions (T=450 K, P=10 bar, t_r=8 s and τ=29 s), the specific productivity increases 100-fold. At the same time, the consumption of KHCO₃ is decreased 4-fold, as compared to the industrial protocol that used at least 5 m of KHCO₃.

Figure 4

Specific performance L_p,v as a function of the temperature (T) representative for the determination of the optimal novel process window at P=10 bar, C_1,0=0.4 m, C_KHCO3,0=1.2 m and X=30%. The conventional batch process is represented by the star.

The optimal conversion for a chemical process can only be determined, based on a detailed economic analysis being out of the scope of this study. Since the equilibrium conversion decreases with T, the same procedure of T optimization has been applied for lower equilibrium conversions as presented in Figure 5. As can be seen, the intensification factor reaches 100,000 for a conversion equal to 10%. Even more impressive PI factors can be attained (up to 10⁷) keeping the concentrations as high as in the conventional industrial process. The prediction could be extended for concentrated solutions, which can be seen in Figure 6, showing the PI factor up to 9·10⁶ at conversion of 10% with C_1,0=1 m and C_KHCO3,0=5 m. It is worth noting that this prediction is based on increased intrinsic kinetics only. These fast reaction transformations could not be easily attained even using micro-reaction equipment, due to heat and mass transfer limitations. Thus, at 530 K, the characteristic reaction time becomes t_r=0.16 s, and at 630 K only t_r=0.004 s. In comparison, under standard industrial conditions t_r=1000 s.

Figure 5

Specific performance L_p,v as a function of the temperature representative for the determination of the optimal novel process window at P=10 bar, C_1,0=0.4 m, C_KHCO3,0=1.2 m for different equilibrium conversions. The conventional batch process is represented by the star.

Figure 6

Specific performance L_p,v as a function of the temperature representative for the determination of the optimal novel process window for different equilibrium conversions at P=10 bar, for concentrated solutions C_1,0=1 m and C_KHCO3,0=5 m. The conventional batch process is represented by the star.

3.2 Validation of the PI prediction in the milli-reactor with static mixers

In order to validate the predicted PI, a milli-reactor equipped with static mixer elements was used (for details see the experimental section). The experimental conditions were set at 433 K, P of 21 bar and the equilibrium conversion of 35%. Under these conditions, the characteristic reaction time of t_r=8 s is more than one order of magnitude higher of characteristic mixing time (0.04 s), ensuring efficient mixing in the first compartment of the reactor. The adiabatic T rise has been estimated as only a few degrees since a relatively low concentration of reactants was used. Therefore, no issue of heat removal has to be considered.

The simulations were performed applying mass and energy balances corresponding to a plug flow reactor. For static mixers, the assumption of an ideal plug flow reactor is justified by the presence of metal baffles, which produce an intense radial mixing.

The mass balance can be expressed as a function of the reactor length (z) as follows [24]:

where F_i is the molar flow rate of compound i, S the reactor section and R_i represents the transformation rate of compound i.

Assuming a constant T over the section of the reactor, a heat balance on a differential volume element can be performed [24]:

where r_j is the rate of reaction j, ΔH_j the enthalpy of reaction j, Q₀ the inlet volumetric flow rate, ρ the fluid density, c_p the specific heat, d_t the tube diameter, T_f the T of the external heat exchange fluid, T the mean T of the reaction mixture and U corresponds to the overall heat transfer coefficient.

In the present study, the external heat transfer and the wall resistance were neglected. Moreover, the internal heat transfer coefficient (h) was estimated using correlations based on the dimensionless Nusselt number (Nu) [25]:

The corresponding definitions for the Reynolds number (Re) and Prandlt number (Pr) are the following:

Re is calculated based on the normalized diameter (D). The Pr_W is the Pr refereed to the wall T. The parameters for the equation corresponding to each static mixer are recapitulated in Table 1.

The estimation of the kinetic constants was done based on the following empirical kinetic model [23]:

By applying this model, the results of the simulations were compared with a set of experimental data obtained at 433 K and 21 bar (Figure 7). Even if the optimal T is 450 K, the experiments at 433 K validate the idea where a residence time in the range of <1 min could be attained with a milli-reactor.

Figure 7

Conversion as a function of time for 21 bar at 433 K, C_1,0=0.4 m, C_KHCO3,0=1.2 m for equilibrium conversion of 35%; the simulation (solid dark line) results are compared with the experimental data (symbols).

A threshold P (21 bar) has to be applied to avoid solvent and CO₂ evaporation from the reaction mixture. The deviation between the experimental results and the simulation is in the order of 15%. The agreement is satisfactory, considering the simplifications that were done in the simulation model. These experiments prove the feasibility of proposed concepts and shows that the milli-scale reactor is a suitable design for industrial scale production.