Process intensification

2.3.1 Internally heat integrated distillation columns

In a standard distillation column, heat is supplied in a reboiler and taken out in a condenser at the top of the column. Owing to the temperature difference between the heating and cooling media employed in the reboiler and condenser, respectively, the separation of components is always accompanied by energy losses. Common distillation processes have an overall thermodynamic efficiency of about 11% (Cussler and Dutta 2012) and are therefore important targets for PI, as they consume a large fraction of energy in chemical plants. Freshwater (1951) introduced the first extensive thermodynamic analysis of distillation processes. He suggested transferring heat from the rectifying to the stripping section of a single distillation column, in order to reduce the temperature lift by a heat pump. This idea was worked out further by Flower and Jackson (1964) using a series of simulations based on the second law of thermodynamics (see also Null 1976). The concept of heat-integrated distillation column (HIDiC) was further investigated by Fitzmorris and Mah (1980) and Nakaiwa et al. (1997, 2001). Nakaiwa et al. (2003), Kiss and Obujić (2014), Kiss (2013), and Górak and Stankiewicz (2011) reviewed the early developments of the HIDiC technology and other recent separation systems. The HIDiC is an improvement of heat-pump operated distillation. HIDiC technology offers several very attractive features (Nakaiwa et al. 2003):

• High energy efficiency. The highest degree of internal heat integration within HIDiC offers higher energy efficiency than other heat-integrated distillation columns do, and a large number of extra degrees of freedom for the process design, thus making it possible to utilize the maximum potential energy savings within a distillation column.

• Zero external reflux and boil-up operation. In HIDiC, the internal heat integration plays the role of the reboiler and condenser employed in common distillation columns, such that in an ideal case, neither a condenser nor a reboiler is necessary.

• Enhanced potential of internal heat integration techniques. Using internal heat integration, many applications within distillation processes are possible, such as multicomponents separation, batch distillation, pressure-saving distillation, and RD.

The principle of a HIDiC system is presented in Figure 15.

Figure 15:

(A) Principles of a heat-integrated distillation column; (B) schematic drawing of a pilot plant (Gorák and Stankiewicz 2011).

In a HIDiC, heat is transferred from a hotter rectification section (operated at higher pressure) to the colder stripping section (see Figure 15A), therefore leading to gradual evaporation and condensation along the length of the stripping and rectifying section, respectively. This concept can also be used if the rectifying and stripping sections are of different sizes. Thus, a short stripper can be connected only to the top or bottom part of a long rectifier, and similarly, a short rectifier can be connected to the top or bottom part of a long stripper (Kiss and Obujić 2014). In order to establish the required temperature driving force, the vapor leaning the top of the stripping section is directly recompressed such that the rectifying section is operated at a higher pressure than the stripping section. The main advantage of HIDiC is based on the fact that the compression effort in this device is limited to bridging the temperature difference over the height of the stripping section only. In order to maximize the thermodynamic efficiency of HIDiC, the compression ratio (P_out/P_in) should be minimized (de Rijke 2007). However, minimizing the compression ratio implies minimization of the temperature difference, which can be compensated for only by a corresponding increase in the heat transfer area. Mah et al. (1977) were the first who described a HIDiC model (see also Tondeur and Kvaalen 1987, Nakaiwa 1988). This model is based on a modified tridiagonal matrix method for the mass and energy balances proposed by Wang and Henke (1966). A good equipment design is essential for industrial implementation of HIDiC. A systematic design procedure for HIDiC was proposed by Gadalla (2009), who divided the design into a thermodynamic and hydraulic part. The design process is initiated by simulating a conventional column or heat pump configuration for the given design problem. Based on the common design parameters, like feed flow rate and the feed temperature/pressure and composition, column pressure, pressure drop, and product specifications, one calculated the number of separation stages in each column section reboiler and condenser heat duties. In case of heat-pump-assisted distillation, the column configuration is simulated to calculate also the compressor power and working pressure ratio. Next, the complete HIDiC configuration, including heat exchange or integration, is simulated employing the basic design. Here, the first step considers the thermodynamic capabilities of the design and then the hydraulic capacities of the stages. The procedures are completed by optimization procedures. An interactive design approach employing a modified Pouchon-Savarit diagram was proposed by Wakabayashi and Hasebe (2013). Shahandeh et al. (2014) and Gutierrez-Guerra et al. (2014) introduced a genetic algorithm, respectively, for optimization of HIDiC configurations. As yet, almost all investigations on HIDiC were simulations and laboratory-scale experiments. One exception is the so-called SuperHIDiC introduced by the Japanese Toyo Engineering company (see Figure 16) (Chementator 2012) (see also Long et al. 2016).

Figure 16:

SuperHIDiC configuration developed by Toyo Engineering Corp. in collaboration with the National Institute of Advanced Industrial Science and Technology (Tokyo, JP).

The company expects energy reductions up to about 50% in refineries employing this column. Examples of application of this column are propylene fractionation, xylene column, and aromatics disproportionation unit stripper. Maruzen Petrochemical Co. Ltd. (Japan) installed the first SuperHIDiC in a methyl-ethyl-ketone unit.

The features of this column are the following:

• Side-exchangers can be provided at composition where enthalpy change should be provided.

• Heat transfer area can be adjusted so as to meet desirable heat duty.

• Pairing of stages between rectifying section and stripping section can be determined in consideration of desirable heat duty and temperature difference.

• Overall heat transfer coefficient can be predicted on a high confidence level.

The rectifying section located below the stripping section is characterized by the following properties:

• Side-exchanges can be accomplished by means of thermos-siphon/gravity.

• Normal trays or packing is utilized.

• Side-cut to obtain intermediate product or multiple-feed can be carried out.

• Feed stage can be optimized.

SuperHIDiC enables the highest energy saving performance in distillation systems developed so far. The system is colicensed by the National Institute of Advanced Industrial Science and Technology and Toyo Engineering.

2.3.2 Dividing-wall column

A useful and efficient concept for the separation of mixtures with three or more components in one column is the dividing-wall column (DWC) (Kaibel 1987, Olujić et al. 2009, Asprion and Kaibel 2010, Dejanović et al. 2010, Niggemann et al. 2010, 2011, Kiss and Obujić 2014). The DWC is an implementation of a thermally coupled distillation design already investigated in the years 1935 (Monroe), 1946 (Wright), and 1965 (Petlyuk). Because the vertical partition in the column shell separates the column into a feed and a side stream section, DWCs are thermodynamically equivalent to Petlyuk columns. The vertical walls avoid radial mixing of vapor and liquid streams and enable the withdrawal of three products with any thermodynamically feasible purity in one column (see Figures 17 and 18).

Figure 17:

(A) Dividing-wall column; (B) basic types of dividing-wall columns (Yildirim et al. 2011).

Figure 18:

(A) Dividing-wall column; (B) integrated three-column configuration; (C) decomposed three-column configuration (Schröder et al. 2016).

Compared to conventional columns-in-series and/or in-parallel configurations, a DWC requires much less energy (up to 30% less), capital, and space. Dejanović et al. (2010) give a complete overview of the work done until 2010 on the research on implementation of DWCs, from early ideas on thermal coupling of distillation columns to practical issues that needed to be solved for their successful implementation. The design of DWCs is more difficult than that of a conventional distillation column, as there are more degrees of freedom to be considered. As yet, design models for DWCs are not available in commercial simulators (Aspen^+®, ChemCad^®, ProSim^®, etc.). Approaches to simulation, optimization and control are presented in that paper (see also Grossmann et al. 2005, Kraemer et al. 2009, Niggemann et al. 2010, Dejanović et al. 2011, Kiss and Bildea 2011, Yildirim et al. 2011, Barttfeld et al. 2004, Caballero 2015, Segovia-Hernández and Bonilla-Petriciolet 2016, Franke 2017, Waltermann and Skiborowski 2017). Particular focus is directed on column internals and dimensioning. The patents literature is also surveyed by Dejanović et al. (2010). Yildirim et al. (2011) focused on current industrial applications of DWCs and related research activities. Long and Lee (2014) reviewed some energy-efficient distillation technologies that can be used in a retrofit design. The research on retrofits and implementation of retrofits using a thermally coupled distillation sequence and a DWC were reviewed. The paper proposed coupled schemes as alternatives to a Petlyuk column that do not have the classical problem of controlling the vapor transfer from one column to another and have not previously been proposed for retrofits. The authors evaluated also a solution for retrofitting azeotropic and extractive distillation systems. Various issues such as the constraints, techno-economic analysis, controllability, and operability were discussed (see also Stankiewicz and Moulijn 2004, Kiss 2013, Long et al. 2016, Rangaiah 2016). DWCs are used for a broad spectrum of systems, e.g. hydrocarbons, alcohols, aldehydes, ketones, acetals, amines, and others (Kaibel et al. 2006). Important is a comparison of modeling results with experimental data for stationary and instationary conditions. The instationary results are needed to cover also start-up and shut-down conditions. Additionally, a concept of the respective control systems is required (Donahue et al. 2016). Niggemann et al. (2010) carried out an extensive analysis of DWCs based on both experimental and simulation studies. Several different feed compositions of a ternary mixture of fatty alcohols have been examined in a purpose-built pilot plant while maintaining product specifications of ~99 wt% at steady-state conditions. Their process model consists of various elements, such as equilibrium stages, collectors and distributors, a reboiler, and a condenser. These elements can be arbitrarily combined to simulate any desired column setup. The model uses ordinary differential equations obtained from mass and energy balances of each element and a set of algebraic equations that are used to predict the physical properties, the vapor-liquid equilibria, and the column hydraulics. The MESH equations for the equilibrium stage model, in which the theoretical stages are numbered from top to bottom, have been set up. Similar model equations have been employed to describe the other elements (reboiler, condenser, collectors, and distributors). The highly nonlinear differential algebraic system of equations has been implemented in the commercial software tool ASPEN Custom Modeler. The model is capable of describing the process characteristics of DWCs, such as the heat transfer across the dividing wall (DW), the liquid distribution above the wall, and the self-adjustment of the vapor distribution below the wall. The heat transfer across the vertical partition may not be neglected owing to its impact on the temperature profiles and the required reboiler duty. The model calculates a self-adjusting vapor split according to the condition of equal pressure drop on both sides of the DW. An extensive validation of the model showed very good agreement between experimental and simulation results for pressure drops, product compositions, and temperature profiles. In addition, it was found that heat transfer across the DW and pressure drop in the nonseparating column internals are extremely important quantities that must not be neglected in laboratory-scale columns. The vapor split values vary significantly in range 0.3–0.5 if the heat transfer across the DW is taken into consideration. If this heat transfer is neglected, the vapor split remained virtually constant around 0.52. Ehlers et al. (2015) confirmed that heat transfer across the DW can actually increase or decrease the minimum energy demand of a DWC, but a heat stream crossing the DW will never change the minimum energy demand of a DWC by more than the amount of heat that is transferred. This implies that the effect of heat transfer across the wall on the energy demand of a DWC can be kept low, if the column is operated close to the point of minimum energy consumption. Niggemann and Fieg (2012) carried out an extensive validation of the transient behavior of a DWC. First, the mathematical process model was parameterized for existing DWCs on the laboratory and production scales. Then, appropriate disturbance scenarios were identified based on systematically conducted simulation studies. Subsequently, these disturbance scenarios were realized on laboratory- and production-scale columns. A comparison of the experimental and simulated results has shown that the simulation model describes the experimental disturbance scenario well in all column segments. Different propagation velocities and different amplitudes of temperatures and concentrations were successfully described. Thus, the transient process behavior of the simulation model is able to describe both the steady-state and transient behavior of DWCs. The great advantage of this validated model is that it “replaces” a real plant. Hence, several types of analysis can be efficiently conducted by means of the process model, such as developments in the fields of process control, optimization, troubleshooting, and debottlenecking. This contributes to a safe and cost-efficient transfer to a corresponding production-scale column. Adrian et al. (2004) have shown that model predictive control for DWCs is superior to single-loop PI controllers, in particular when constraints for operating conditions have to be taken into account. Ling and Luyben (2009) proposed a new DWC control structure that controls the purities of product streams and also minimizes energy consumption (see also Kiss and Bildea 2011). This was achieved implicitly by controlling a composition of the heaviest component in the prefractionator. Disturbances in feed flow and feed composition were used to demonstrate the effectiveness of the proposed control structure. A comparison of the dynamic controllability of the DWC with a conventional configuration was also provided. The authors discussed the problem of defining control structures for DWCs. Zavala-Guzmán et al. (2012) discussed the tuning of PI controllers for DWCs within a pole placement framework to compute values of controller gains, in terms of well-known and identifiable parameters for each control loop, such as static gains and time constants. The goal was to establish a methodology that reduced the number of tuning freedom degrees that inclusively allowed a simultaneous tuning of all controllers. An extension of this approach based on discrete measurements was also published by Zavala-Guzmán et al. (2016). Rodriguez et al. (2017) presented the control of an extractive and reactive DWC (R-DWC). The authors established a decentralized structure as well as a model predictive control and compared both approaches. The decentralized control showed a very oscillatory response as the result of the interactions being difficult to control; on one hand, the multivariable predictive control showed a smooth response and good control being able to handle larger disturbances than the decentralized option. Buck et al. (2011) have introduced decentralized temperature control systems by a systematic approach. Different control systems have been designed and compared by means of simulation and experiment. The methodology is a general framework, applicable to a wide range of control problems in DWCs. Errico et al. (2008) discussed a retrofit of an industrial distillation whereby a large part of the existing equipment should be kept. Various modifications of the existing plant (elimination of all reboilers/condensers or some of the reboilers/condensers) were compared with respect to the energy savings and investment costs. The optimal solution was found to be the alternative that has the minimum modifications of the existing plant. A saving of 23% on energy consumption could be achieved.

Some other models for DWCs have been reviewed by Dejanović et al. (2010). As different column configurations in a distillation network can achieve the same ultimate products, with different amounts of energy to do so. Therefore, systematic generation and evaluation of all possible complex column configurations for given multicomponent separation problems are desirable. Linninger (2009) reviewed advances to tackle the challenges of synthesizing complex column networks with computer-aided design algorithms. Linninger, in particular, discussed his temperature collocation approach, which replaces the column height or equilibrium stages as an independent variable with the bubble point temperature of the liquid mixture on a given equilibrium tray. The composition trajectories are expressed as functions in the independent variable, T, instead of the height of the column. A function was formulated that describes the composition profiles, x_i, as a function of the temperature-dependent equilibrium constant, K_i, column section difference points, and generalized column section reflux. This transformation has been found to approximate column profiles for various solution models very accurately. A novel hybrid genetic algorithm deploying stochastic search in a reduced design space and Newton-methods to satisfy state equations were shown to solve optimal structure and design variables simultaneously. This is an effective tool for modeling and optimization complex column configurations while accounting for nonideal thermodynamics. Grossmann et al. (2005) reviewed work on the optimal design of distillation of individual columns using tray-by-tray models. A combination of novel representations for individual columns and superstructures, combined with disjunctive programming (Yeomans and Grossmann 2000) and robust initialization schemes, has made it possible to solve these problems with reasonable computational effort. Shah and Agrawal (2010) described a simple-to-use six-step matrix method for obtaining all the basic distillation configurations and additional thermally coupled configurations that separate a zeotropic multicompoment feed into essentially pure product streams. The configurations obtained were ranked for a given application subject to criteria of interest. The only information needed to generate the configurations was the number of components in the feed. The authors successfully enumerated all the configurations for feeds containing up to eight components. The approach can also be used to generate nondistillation and hybrid separation configurations and even easy-to-retrofit configurations. Caballero (2015) proposed a novel algorithm for the rigorous design of distillation columns that integrates a process simulator in a generalized disjunctive programming formulation. Operational conditions (reflux ratio, internal and external flows, etc.) as well as structural parameters (number of trays in each column section and, consequently, location of feed and product streams, etc.) were simultaneously optimized. The model was solved using a hybrid simulation optimization approach by taking advantage of the effective and reliable numerical methods included in process simulators for converging distillation columns as well as the thermodynamic packages, property estimation, etc., and at the same time the flexibility of equation-based environment (see also Caballero and Grossmann 2014). Caballero and Grossmann (2013) also demonstrated how to systematically identify all the sequences of separation tasks that can produce configurations that include at least a DWC. Rong (2011) has shown that DWCs can be systematically generated from conventional simple column configurations. Because the simple column sequences with sharp splits are the simple and widely studied conventional schemes for multicomponent distillation, the purpose of Rong’s work is to formulate a procedure for systematic synthesis of DWCs for such simple conventional schemes. A four-step procedure was formulated that systematically generated all possible DWCs from the simple column sequences. Starting from the subspace of the column sequences, the subspace of thermally coupled configurations is first generated, from which the subspace of thermodynamically equivalent structures is then generated. Finally, the subspace of the DWCs was generated. The procedure is easy to use and can systematically generate the possible DWCs from the simple column sequences of an n-component mixture.

Design optimization can be performed based in either sequential design procedures, whereby unit operation models may be taken from commercial simulators, or a simultaneous optimization employing mathematical programming procedures. In general, the sequential design starts with shortcut calculations that are mostly based on Fenske-Underwood-Kirkbride equations, followed by rigorous MESH models (see, for example, Seader et al. 2010). Halvorsen’s (2001)V_min approach can provide good initialization values for liquid and vapor splits. The rigorous models may be taken from a commercial simulator. For example, ASPEN Plus^® contains a so-called MULTIFRAC model for a Petlyuk configuration that is thermodynamically equivalent to a DWC without heat transfer across the wall. Luyben (2013) describes the steps to initialize the model and achieve the product specifications. DWC can be decomposed into several columns (see Figure 18 for a four-column decomposition) in such a way that the configuration is thermodynamically equivalent to a DWC. Dejanović et al. (2011) introduced a systematic procedure to DWC design employing V_min graphs for evaluating the minimum energy demand, followed by a rigorous simulation based on the four-column configuration (see Figure 19).

Figure 19:

(A) Dividing-wall column (DWC); (B) four-column configuration of (A) (Waltermann and Skiborowski 2017).

Waltermann and Skiborowski (2017) introduced a tailored superstructure for simultaneous simulation and optimization of DWCs. The formulation of their DWC model exploits the decomposition approach of Carlberg and Westerberg (1989) for problem initialization, whereas the DWC is represented by three thermally coupled columns. The mathematical model of the columns is based on the MESH equations and employed the decomposition of the optimization problem into a generic superstructure model and an implicit model for equilibrium and enthalpy calculations (Skiborowski et al. 2015), which is integrated into the optimization problem by means of an external function. Different phases and equilibrium solutions were determined by a reliable homotropy continuation algorithm (Press et al. 2007). The introduced external function overcomes the implicit discontinuities at the interface of possible miscibility gaps by means of a reformulation of the three-phase column model. The optimization was implemented in a general algebraic modeling system (Brooke et al. 2005) environment. The optimization of the discrete design degrees of freedom, i.e. the number of equilibrium trays in each section, was accomplished by means of variable locations for the feed stream, the side-streams of the subsequent columns, the top vapor stream of the lower and the bottom liquid stream of the upper subsequent column, as well as the reflux from the reboiler and the condenser. Each potential location of streams was modeled by binary decision variables. The resulting MINLP problem (Fletcher 2009) was solved in a series of relaxed NLP problems, according to an approach by Kraemer et al. (2009). Franke (2017) has presented a MINLP optimization algorithm that considers the nonconvexities of the rigorous distillation model and is able to find a high-quality solution for complex distillation problems. The developed modified Generalized Benders Decomposition algorithm was applied to the optimization of DWCs. Some calculated examples of DWCs showed significantly lower energy savings than reported in the literature. The author pointed out that it is of utmost importance to compare the optimized DWC with the optimized conventional sequence, based on the same cost function and physical property data, because otherwise, the wrong process option may be favored. Further optimization approaches have been presented by Waltermann and Skiborowski (2017).

Recent developments in equipment for DWCs have been described by Olujić et al. (2009). The authors have mentioned that a balance between pressure drop in column packings and avoidance of vapor and liquid maldistribution has to be paid attention to. CFD turned out to be a useful tool for detecting flow maldistributions. Asprion and Kaibel (2010) discuss DWCs for four and more component mixtures. Recent developments in intensified distillation-based separation processes in China have been reviewed by Li et al. (2016). Extensive reviews of distillation have been presented in edited handbooks by Huang et al. (2008), Gorák and Schoenmakers (2014) and Gorák and Olujić (2014).

An R-DWC combines a reactor and a separation unit in one column shell (see Figure 20).

Figure 20:

Schematic showing the path from a distillation column to a reactive dividing-wall column (gray area represents the reactive zone) (Yildirim et al. 2011).

Such a configuration is an example of an extensive process integration. As can be observed from Figure 20, R-DWCs are a combination of RD (see Section 2.3.4) and DWCs. The advantages of this integrated process, i.e. high conversion, high selectivity, and product purity as well as cost and energy savings, have been demonstrated. These features are based on the following properties (Mueller and Kenig 2007):

• Overcoming chemical and thermodynamic equilibrium limitations, leading to an increased yield.

• Suppression of undesired consecutive reactions, resulting in a higher selectivity.

• Direct heat integration in the case of exothermic reactions, leading to a reduced energy consumption.

• Simultaneous liquid evaporation and reaction avoids hot spots.

• Components with close boiling points may be separated.

The difference between a DWC and an R-DWC is the reactive zone in the prefractionator at the feed side of the column (Figure 20). In this reactive zone, the reactants of the feed stream are transformed into the products. Multiple feed streams might be used. For heterogeneously catalyzed reactions, catalyst particles are placed into the reactive zone, which might be structured packings with catalyst bags. Kaibel published the first paper on R-DWCs in the year 1984 (Kaibel 1984). Mueller and Kenig (2007) issued the first paper on R-DWCs employing a rate-based approach. Special attention was given to both the heat transfer through the DW and the vapor flow rate split below the DW. Schröder et al. (2016) analyzed the fundamental mechanisms of the R-DWCs, and a profound process understanding was deduced based on principal aspects of RD and the DWCs. Fields of application and insights into the key factors for an energy-efficient operation were systematically derived. A semi-shortcut method was proposed to determine the minimum vapor demand of the R-DWC, and its energy-saving mechanism was explained. Already during the process synthesis, one can evaluate whether R-DWCs are a promising option: The possible energy savings can be quickly quantified by this approach. In a further paper, Schröder and Fieg (2016) systematically investigated the influence of the separation properties of the reaction system on the energy saving potential. The saving mechanisms were also analyzed. Sander et al. (2007) presented experimental results of methyl acetate hydrolysis in a DWC obtained from a miniplant and a column of industrial scale. Ehlers et al. (2017) for the first time compared experimental results found in DWC and simulation results. A complex reaction system showing nonnegligible side reactions was used. The data obtained by the DWC were also compared with the operation of a conventional, nonreactive DWC with a similar chemical system analyzed for the same experimental setup. The differences between experimental and simulation results were small. Therefore, that paper gives a sound basis for the design of R-DWCs. Further details on integrated chemical processes may be found in an edited book by Sundmacher et al. (2005).

2.3.3 Reactive absorption, reactive extraction, and reactive crystallization

Reactive absorption (RA) combines absorption of gases with chemical reactions in the liquid phase, whereby the reactions improve the mass transfer and enhance solubility. It largely depends on the stoichiometry of the reactions, concentration of the reactants, and the mass transfer rates. In most cases, RA is carried out in plate or packed columns. In plate columns, the phases flow countercurrently, whereas in packed columns, liquid flows along the packing surface, and the gas occupies the rest of the void volume. Guidelines on how to select the most appropriate internals for specific RA applications are presented by Kohl and Nielsen (1997). RA may also be performed in spray columns, Venturi scrubbers, bubble columns, or thin film contactors (Yildirim et al. 2012). The main advantages of RA are the reduction in investment and operating costs, improved product purity, and lower energy demands.

As RA occurs in multiphase and multicomponent fluid systems with complicated heat and mass transfer phenomena, combined with one or several reactions, and the presence of neutral and sometimes ionic species, the modeling of RA is quite complicated. Phase equilibria, chemical equilibria, physical properties, like diffusion coefficients and viscosities, hydrodynamic and mass transport properties, and reaction kinetics have to be taken into consideration. At present, the most suitable and reliable approach for the description of staged absorption units is the rate-based approach (Kenig et al. 2003, Kenig 2008, Yildirim et al. 2012). As the conditions for RA processes vary considerably, an effective combination of different modeling approaches was suggested by Kenig (2008). The rate-based approach takes the rates of multicomponent heat and mass transfer and the chemical reaction directly into account. The mass transfer between the gas/liquid phases can be described by various models, like the two-film model (Wang and Langemann 1994) or the penetration/surface renewal model (Chung et al. 1971). The multicomponent diffusion in the films can be simulated by the Maxwell-Stefan equations (Taylor and Krishna 1993). The chemical reactions are included using a source term in the balance equations for each bulk phase and the mass transport equation for the film. Many two-film parameters can be found in the literature. Typical applications of RA are CO₂/H₂S removal from acid gas by ethanolamines, CO₂ removal from power plant flue gas, NO_x and SO₂ removal from nitric acid plant and power plant off gases, respectively, and sulfuric acid production (Yildirim et al. 2012, Ramaswamy et al. 2013, Sharma et al. 2013, Smit et al. 2014). Olefin/paraffin separations by RA were reviewed by Safarik and Eldridge (1998). A review of advances in liquid adsorbents for CO₂ capture was presented by Rosli et al. (2017) and by Yu et al. (2012).

Reactive extraction (RE) is a combination of physical extraction and chemical reaction in a single processing unit, whereby the reactions enhance the capacity of the solvent (Bart 2001). Esterification is mainly chosen as case study in the literature. For the simulation of countercurrent columns, one has to describe the reactive equilibria involved, based on Gibbs excess modeling, mass transfer, and diffusive resistances, as well as complex formation involving aqueous electrolytes and organic species. The column hydrodynamics is an additional item to be taken into account (Bart 2003). RE can be categorized into liquid-liquid extraction and liquid-solid extraction. In liquid-liquid RE, a second liquid phase or solvent is added into the reaction system. In general, this exhibits high miscibility and selectivity with the intermediates or products of the reaction, so that they can be continuously extracted from the reaction as soon as they have been generated. It must be chemically inert and highly immiscible toward the reactants in order to avoid undesired side reactions. Removal of products prevents further reactions and, therefore, a reduction in the yield of desired products. Solid-liquid RE involves the extraction of materials from the solid into the liquid phase, where they react and create the products wanted. For example, solid biomass often contains several useful raw materials such as lipids, proteins, fatty acids, and hydrocarbons, which can be processed to valuable products after being extracted. The kinetics of solid-liquid RE can be increased through the addition of an appropriate catalyst. Datta et al. (2015) reviewed the applications of RE in organic chemistry and biochemistry, Bart (2005) in hydrometallurgy, and Boodhoo and Harvey (2013a,b) in green chemistry. Mizzi et al. (2017) developed a general design methodology for the reactive liquid-liquid extraction, which is composed of three steps: feasibility analysis, predesign determination, and simulation validation. Their paper focused on the first two steps. The authors adapted a calculation procedure by Samant and Ng (1998) for the design of a reactive liquid-liquid column. The feasibility analysis allows verifying that the separation specifications are thermodynamically feasible. The reactive liquid-liquid diagram and the reactive extractive curve map were employed as tools for the feasibility analysis. In the second predesign step, a column configuration is determined that allows to achieve the separation goals. Additionally, the structural parameters of the column are determined. The validation steps allow validating the previously determined configuration of the column. This procedure was applied to different reactive systems with different strategies and numbers of compounds with succinct acid and leads to the design of an RE column. Thakre et al. (2016) studied RE of citric acid from dilute aqueous solutions using three different extractants. The linear solvation energy relationship model was successfully applied in that paper to predict the distribution coefficients. The process was optimized by a differential evaluation procedure, which is based on a genetic algorithm (Kumar et al. 2011a). Wasewar et al. (2004) reviewed the fermentation of glucose to lactic acid coupled with RE.

Reactive crystallization (RC) or precipitation is an important production step for many chemical and pharmaceutical industries to produce solid particles with desirable characteristics, such as high purity, large crystal size, narrow crystal size distribution (CSD), enantiomeric purity (Berry and Ng 1997, Yu et al. 2007b, Beckmann 2013, Lewis et al. 2015). RC combines the typical phenomena in crystallization, namely nucleation and crystal growth with chemical reaction. Zimmermann et al. (2015) investigated sodium chloride (NaCl) nucleation from supersaturated brines using seeded atomistic simulations, polymorph-specific order parameters, and elements of classical nucleation theory. The authors found that NaCl nucleates via the common rock salt structure. Ion desolvation was identified as the limiting resistance to attachment. The simulations were performed at realistic supersaturations to enable the first direct comparison to experimental nucleation rates for this system. Berry and Ng (1997) presented a systematic method to synthesize RC processes. It showed how to selectively crystallize a desired solid product after a reaction step and how to use compound information to effect separation of a mixture. The method is based on the generation of phase diagrams with liquid-phase reactions. Using transformed coordinates, systems with three or fewer degrees of freedom can be conveniently analyzed, regardless of the number of components and reactions. Features of the solid-liquid phase diagram that are relevant to process synthesis were identified. Li et al. (2012) analyzed the problem of particle size distribution of precipitation products resulting from inhomogeneous micromixing during RC. Wang and Ward (2015) studied conceptual methods for the rapid development of operating recipes for seeded batch RD processes using two case studies: production of barium sulfate and production of l-glutamic acid (LGA), which are model systems for RC that are widely studied in the literature. Seed loading analysis predicted that nucleation can be effectively suppressed by seeding for the LGA process, but not for the barium sulfate process. These results were supported by rigorous numerical simulations. Sarkar et al. (2007) demonstrated the potential for multiobjective optimization for a semibatch RC process. A genetic algorithm was used to obtain the Pareto-optimal solutions for constrained multiobjective optimization problems that are related to the quality of product CSD in reactive crystallizers. Insights and guidelines were obtained regarding the optimal operation of the crystallizer to obtain crystals of desired characteristics. Su et al. (2014) derived a mathematical model for semibatch pH-shift RC of LGA that takes into account the effects of protonation and deprotonation in the species balance of glutamic acid, CSD, polymorphic crystallization, and nonideal solution properties. The kinetic parameters were estimated by Bayesian inference from batch experimental data collected from the literature. Probability distributions of the estimated parameters, in addition to their point estimates, were obtained by Monte Carlo simulations. The first-principles model was observed to be in good agreement with experimental data. This model may be used for robust control strategies. Alatalo et al. (2010) developed a feedback control process approach for semibatch precipitation that allows control of both the form of the polymorphs and the CSD. The effectiveness of the use of an ultrasound probe for start-up of a crystallization process producing ammonium sulfate crystals from aqueous solution in a 75-l crystallizer was investigated by Lakerveld et al. (2011). Quantification of the number of produced crystals with ultrasound revealed the importance of ultrasonic power input, insonation time, and the geometry of the insonated volume. Devices for RC PI, like the spinning disk reactor, will be discussed below.

2.3.4 Reactive distillation

RD, sometimes called catalytic distillation, is a combination of reaction and distillation in a single column. The reactants are converted with simultaneous separation of the products and recycling of unused reactants. This implies that the boiling points of the reactants must be different from those of the reactants. By continuously removing the products, RD makes it possible to use only the stoichiometric reactants ratio and to pull the equilibrium to high conversions (see Figure 21) (Harmsen 2007).

Figure 21:

(A) Reactive distillation column (RDC); (B) RDC with double reactive sections.

It is one of the most important industrial applications of PI. The reactions in RD include heterogeneous catalysis reactions, homogeneous catalysis reactions, and thermal noncatalytic reactions. In almost all cases, reactions take place in the liquid phase. Although already known in the 1920s, industrial application of RD did not take place before the 1980s. Eastman Chemical Co.’s methyl acetate RD process and the production of fuel ethers were the first large-scale RD processes. The Eastman process used 80% less energy at only 20% of the investment costs (Luyben and Yu 2008). Today, RD is utilized for processes like esterification, transesterification, hydrolysis, etherification, hydrogenation, dehydrogenation, alkylation, metathesis, carbonylation, polymer production, acyloxysilanes production, chlorination, nitration, amination, carbonates production, and chiral separations (Harmsen 2007, Kiss 2014a,b). All these reactions are limited by chemical equilibria. RD has several advantages (Taylor and Krishna 2000, Tuchlenski et al. 2001, Sundmacher and Kienle 2002, Noeres et al. 2003):

• Reduction in capital investment, because two process steps are carried out in the same device.

• Exothermic heat of reaction is used for vaporization of liquid.

• Avoidance of hot spots and runaways using liquid vaporization as thermal fly wheel.

• Conversion of reactants is improved due to overcoming of chemical and thermodynamical equilibrium limitations.

• Selectivity is increased through suppression of undesired consecutive reactions; therefore, by-product formation is reduced. Products are removed fast from the reaction zone.

• Azeotropes may be avoided by “reacting away” the azeotropes, but one should keep in mind that reaction can induce the formation of azeotropes that were not there to begin with (Song et al. 1997, Ravan 2013, Skiborowski et al. 2013).

• If the reaction zone in the RD column is placed above the feed point, poisoning of the catalyst can be avoided, which results in a longer catalyst lifetime compared to conventional systems.

RD technology is applicable only if the temperature window of the vapor-liquid equilibrium matches the reaction temperature. CDTECH and Sulzer are the major commercial RD process technology providers. Over 200 commercial-scale processes have been installed up to now. The RD process design starts with a feasibility study (Malone and Doherty 2000, Tuchlenski et al. 2001). This study investigates the feasible product compositions from an RD for a given feed, column pressure, and Damköhler number. The analysis must incorporate all of the features for ideal and azeotropic mixtures, as well as phenomena caused by the introduction of chemical reactions. Various process synthesis approaches are available to determine whether the desired process can be performed as RD (Subawalla and Fair 1999, Taylor and Krishna 2000, Sundmacher and Kienle 2002, Kenig and Gorák 2007, Ahmadi et al. 2010, Altman et al. 2010, Huang et al. 2010, Segovia-Hernández et al. 2015, among others). The design of an RD is quite complex due to the strong interaction of chemical reactions and heat and mass transfer. Segovia-Hernández et al. (2015) have provided a review on current applications of deterministic and stochastic optimization techniques for the design of RD. The optimal design of RD systems is a highly nonlinear and multivariable problem, with the presence of both continuous and discontinuous design variables. The objective function used as optimization criterion is generally nonconvex with several local optima and subject to several constraints. Therefore, global optimization algorithms are needed. Simulated annealing algorithms are in use for such purposes (Segovia-Hernández et al. 2015). Aneesh et al. (2016) highlight the advantages of simultaneous design and control of RD processes. They proposed a mixed-integer dynamic optimization (MIDO) approach (Flores-Tlacuahuac and Biegler 2007) with combined control structure selection and tuning parameters with process design. This method has the capability to deal with both linear and nonlinear systems. But before applying this method with nonlinear systems, prior investigations are required on parameter uncertainty and external disturbances. Applying MIDO leads to a stochastic and large-scale MINLP problem, which is difficult to handle numerically. Huang et al. (2010) established a fundamental principle for PI in RD columns through deliberate consideration of internal mass integration and internal energy integration between the reaction operation and the separation operation involved. For RD processes with highly thermal effect, PI can be achieved with an exclusive consideration of internal energy integration between the reaction operation and the separation operation involved. However, in the presence of a highly endothermic reaction, if it features an extremely low reaction rate and/or small chemical equilibrium constant, internal mass integration has also to be considered between the reactive section and the stripping section of the RD column. For RD columns involving reactions with negligibly or no thermal effect, PI can be performed with an exclusive consideration of internal mass integration between the reaction and the separation, while for moderately thermal effects, a careful trade-off between mass integration and internal energy integration has to be conducted. The authors applied their concept to several examples. Wang et al. (2010) explored the impact of operating pressure on PI for RD columns. The authors identified three specific situations. For reversible endothermic reactions in RD columns, PI can be reinforced between the reaction operation and the separation operation involved, because increasing operating pressure enhances reactant conversion and reaction heat load. In the case of equilibrium-limited exothermic reactions, PI can be reinforced between the reaction and separation, because decreasing operating pressure leads to increased reactant conversion and reaction heat load. For RD involving kinetically controlled exothermic reactions, PI can be profitably employed if the operating pressure is appropriately selected such that reactant conversion and reaction heat load are maximized. Keller et al. (2013) optimized RD for multiple-reaction systems by using an evolutionary algorithm. Thereto, a multiobjective optimization study was performed to determine the Pareto-optimal combinations of ethyl-methyl carbonate, selectivity, and dimethyl carbonate (DMC) conversion that can be obtained in the pilot-scale RD column. The results showed that diethyl carbonate (DEC) can be produced in the pilot-scale RD column with a selectivity of 90% with respect to DEC, while still having reactant conversions of ethanol (ETOH) and DMC of 77% and 86%, respectively. Subawalla and Fair (1999) discussed guidelines for solid-catalyzed RD systems. The guidelines were used to generate initial estimates for column pressure, reactive zone location, catalyst mass, reactant feed location, reactant ratio, reflux ratio, column diameter, number of equilibrium stages, and packed height. They formed a part of a methodical design procedure that made extensive use of both nonequilibrium (rate-based) and equilibrium-stage simulation models. The guidelines were tested for two etherification systems and experimentally validated for a hydration reaction. It was found that column diameter depended both on maximum vapor velocity and on packing catalyst density, reactant ratios were a function of conversion and azeotrope formation, and the operating pressure affected the relative volatility, chemical equilibrium, and reaction rate. The reflux ratio had an impact on separation and conversion.

A large part of the literature on RD has a focus on modeling RD columns. Taylor and Krishna (2000) presented an extensive review of modeling RD. Kenig and Gorák (2007), Noeres et al. (2003), and Ahmadi et al. (2010), among others, reviewed RD modeling papers. RD occurs in multicomponent and multiphase systems, with heat and mass transfer at interphases and inside phases. Additionally, one or several, mostly reversible, reactions are involved. Complex hydrodynamics and multicomponent diffusional processes occur. Therefore, RD models are inherently very complex and require elaborate numerical methods. In practice, RD models are split into several submodels, which describe a part of the problem. Subproblems are, for example, heat and mass transfer models, chemical reaction kinetics, hydrodynamics, thermophysical models of the multicomponent systems, and Stefan-Maxwell approach for the multicomponent diffusion. Mass transfer can be described on rate-based methods. For heterogeneously catalyzed systems, phenomena inside the particles have to be described, that means the intrinsic kinetics, multicomponent diffusion phenomena, and reaction heat transport. Wetting of the outer surfaces of the particles is also a modeling problem. The hydrodynamics are often simplified by dispersion models, liquid holdup, and pressure drop simulations. Equilibrium stage models, extended by chemical reaction equilibria for fast reactions and augmented by correlation parameters, like tray efficiencies of HETS, result in a simple RD model. The numerics are ordinary or partial differential equations and algebraic equations for the thermophysical relations. One obtains a differential-algebraic system of equations. However, diffusional interactions of the multicomponent diffusional interactions cause unpredictable behavior of the tray efficiency factors, which show a strong dependency on the component concentration. Mass transfer acceleration due to chemical reactions in the interfacial regions is often described by enhancement factors (Fogler 2016). A more sophisticated way to describe column stages is the rate-based approach (Seader 1989), which directly takes into account the actual rates of multicomponent mass and heat transfer and chemical reactions. Mass transfer at the vapor-liquid interface can be described using the film model or the penetration/surface renewal model (Westerterp et al. 1988), whereas the model parameters are estimated via experimental correlations. Multicomponent diffusion in the films can be rigorously described by the Stefan-Maxwell approach. Vapor-liquid equilibria are calculated by thermodynamic models, such as non-random two liquid model, universal quasichemical equation, or statistical associating fluid theory equation (Kontogeorgis and Folas 2010, Sandler 2011, Gmehling et al. 2012), whereas the heat transfer coefficients can be determined using the Chilton-Colburn analogy (Aris 1989). Finally, one obtains mass and heat balance equations with mass transfer and reaction coupling expressions. Fluid flow, which is often nonideal in catalytic packings, has to be modeled too. Details may be found in the papers mentioned above.

Yu et al. (2014) and Kaymak et al. (2017), among others, have investigated RD columns with double reactive sections. Yu et al. (2014) have proposed this column for the separations of two-stage consecutive reversible reactions. The arrangement of two reactive sections (see Figure 21B) provides more degrees of freedom for the coordination of the two reaction operations involved and the reinforcement of internal mass integration and/or internal energy integration between the reaction operations and separation operations. The columns with double reaction zones require less investment and operating cost than equivalent columns with one reacting zone. As an example, for a reaction scheme to be operated in a double reaction zone column, the trans-esterification of DMC with ETOH to produce DEC and MEOH was chosen. The reaction scheme is as follows:

DMC+ETOH⇔EMC+MEOH

EMC+ETOH⇔DEC+MEOH

EMC=ethyl−methyl carbonate intermediate.

The synthesis and design of the double reactive sections column in question has been given by Yu et al. (2014). Kaymak et al. (2017) have designed several alternative control structures for columns with double reaction sections and compared their robustness based on dynamic simulations. The results have shown that direct composition control is necessary. The authors could demonstrate that two temperature controllers and two composition controllers provide a robust control in terms of steady-state deviations and settling time of product purities.

Kookos (2011) developed a superstructure-based mixed-integer nonlinear model of an ideal RD column and employed it for steady-state optimization and optimal design. Additionally, the optimal point of operation was discovered quite easily and effectively, avoiding the exhaustive enumeration used in previous works. A systematic methodology was applied in order to develop an effective control structure that satisfies product purity constraints at steady state and minimizes the economic penalty associated with product overpurification.

A review of industrial applications of RD has been presented by Hiwale et al. (2004) (see also Kiss 2014a,b). An important problem is the internals of RD columns. A large European project, INTENT (Intelligent Column Internals for Reactive Separations), aimed at fundamental improvement of column internals. The results are presented in a special issue of Chemical Engineering and Processing (Volume 44, 2005). The key theoretical innovation has been the application of CFD software in combination with rigorous, rate-based process simulation. The project resulted in a software tool for internal preselection, ADVISER; a software tool for the fluid flow and reaction simulation, CFX-INTINT; and a rate-based flow simulator, PROFILER. All models were experimentally verified for five reactions in pilot plants. The details have been described in the special volume mentioned above. Discussions of scale-up problems for RD columns may be found in the book by Sundmacher and Kienle (2002).

Besides the standard design of RD columns, there are special designs, for example, horizontal RD for multicomponent chiral resolution. Au-Yeung et al. (2013) have proposed a novel horizontal RD apparatus and a new overall process scheme for continuous multicomponent chiral resolution via reversible enantioselective acylation of a chiral substrate by a chiral acyl donor. The process enables simultaneous production of up to four enantiomers with enhanced chiral purity. The horizontal RD vessel was employed to provide a longer liquid-phase residence time needed for adequate conversion. Low vapor-traffic pressure drop allows operation under vacuum at reduced temperatures for good enzyme stability and enantioselectivity. The general technology has potential as a means to producing a wide range of chiral synthons used in asymmetric syntheses of chiral pharmaceuticals and other biologically active products.

An industrial view of reactive separations has been presented by Stankiewicz (2003), Kiss (2014a,b), Harmsen (2007, 2010), and Tuchlenski et al. (2001), among others.

2.3.5 Membrane separations and hybrid separations

A membrane is a barrier that separates and/or contacts two different regions and controls the exchange of matter and energy between the regions. The membrane can be a selective or a contacting barrier. In the first case, it controls the exchange between the two regions adjacent to it in a very specific manner. In the second case, its function is mainly to contact the two regions between where the transport occurs (see Figure 22).

Figure 22:

Membrane separation principle.

A membrane may be solid or liquid, homogeneous or heterogeneous, isotropic or anisotropic in its structure. A membrane can be a fraction of a micrometer or several millimeters thick. Its electrical resistance can vary from millions of Ohm to a fraction of an Ohm. Membranes are permselective, which is determined by differences in the transport rates of various components in the membrane matrix. The permeability of a membrane is a measure of the rate at which a given component is transported through the membrane under specific conditions of concentration, temperature, pressure, and/or electric field. The transport rate of a component through a membrane is determined by the structure of the membrane, by the size of the permeating molecules, its geometrical structure, and the possible electrical charge of the molecule and/or of the membrane. The selectivity of the membranes is determined by their pore radii distribution for porous membranes (see Figure 23).

Figure 23:

Ranges of pore sizes in membranes and their respective processes.

For polymer membranes, the solute solubility is of importance. Ultrapure hydrogen may be produced by palladium membranes. From an economic point of view, the flux through the membranes is of importance. Selectivity and flux often conflict sharply with each other. Membrane technology has been compiled in many books, for example, Bungay et al. (1983), Drioli and Barbieri (2011), Drioli et al. (2011), Scott and Hughes (2013), Mulder (2012), Marcano and Tsotsis (2002), Strathmann (2011), Drioli and Curio (2007), and Basile (2013). They can be tailored and adjusted to specific separation tasks. Membrane processes meet the requirements of PI because they can, in many cases, replace conventional energy-intensive techniques to accomplish the selective and efficient transport of specific components and to improve the performance of reactive processes. Energy requirements, easy control and scale-up, large operational flexibility, low capital and operating costs, and good stability under operation conditions are further advantages of membrane processes. At present, membrane techniques are essential to a wide range of applications, including the production of potable water (Alzahrani and Mohammad 2014), remediation of textile effluents (Van der Bruggen et al. 2004, Dasgupta et al. 2015), lactic acid production (Pal et al. 2009), biotechnology (Rios et al. 2007, Sirkar et al. 2015), membrane distillation (MDi) (Alkihudhiri et al. 2012), MRs (Dautzenberg and Mukherjee 2001, Marcano and Tsotsis 2002, Drioli et al. 2012, Brunetti et al. 2017), membrane crystallizers (Pramanik et al. 2016), seawater desalination (Elimelech and Phillip 2011), gas separation (GS) (Bernardo et al. 2009), food industry (Mohammad et al. 2012), etc.

Membranes are made of various materials like polymers, ceramics, carbons, zeolites/MOFs, oxides (alumina, titania, and zirconia), and metals (palladium, silver, and alloys). Porous and nonporous polymer membranes are employed for GS, reverse osmosis (RO), nanofiltration, ultrafiltration (UF), and microfiltration (Ulbricht 2006). The ceramic membranes are chemically inert and stable at high temperatures. This stability makes ceramic microfiltration and UF membranes suitable for food, biotechnology, and pharmaceutical applications, in which membranes require repeated steam sterilization. Ceramic membranes are also in use in catalytic reactors. Carbon sieve membranes are employed for GS because of their molecular sieving effects. Zeolites are microporous crystalline alumina silicates with uniform pore size. Although they have molecular sieving effects, there is still no GS using zeolite membranes in industrial use (Rangnekar et al. 2015), but they are utilized for drying organic liquids such as ETOH, acetonitrile, glycerol, etc., by steam permeation/pervaporation using Linde Type A (LTA) zeolite membranes. Dry bio-ETOH is mostly produced by pressure swing adsorption (PSA) using 3A zeolites (K⁺ exchanged LTA). A Japanese institute (Bussan Nanotech Research Institute) developed a Na-LTA zeolite membrane for dehydration of bio-ETOH by a hybrid distillation/vapor permeation process (Sato et al. 2008). In 2012, the Dalian Institute of Chemical Physics installed at Jiangsu Xinhua Chemical Co. Ltd. an LTA zeolite membrane unit for i-propanol dewatering with a capacity of 50,000 tons/year (Rangnekar et al. 2015). Metal-organic frameworks (MOFs) are constructed from metal ions or metal ion clusters and bridging organic linkers. They exhibit, like zeolites, regular crystalline lattices with relatively well-defined pore structures. As yet, more than 20,000 different MOFs have been synthesized. On the laboratory scale, several MOF membranes have been prepared for separation processes, but Qin et al. (2014) state in their comprehensive review that there is still a long way to go in achieving the practical application of MOF membranes in separation processes. Different ways of membrane surface modifications are mentioned in the literature (Rios et al. 2007). Deposition of biopolymers at the membrane surface, covalent attachment of enzymes onto ceramic membranes through glutaraldehyde bonds, and low-temperature-plasma-induced graft polymerization are such surface modifications. Some of these modifications enhance filtration performance. In dense metal membranes, molecular transport occurs through a solution-diffusion mechanism. In particular, in palladium membranes, hydrogen atoms interact with palladium metal. Hydrogen permeation through the membrane is a complex process with stages such as dissociation of molecular hydrogen at the gas/metal interface, adsorption of the atomic hydrogen on membrane surface, dissolution of atomic hydrogen into the palladium matrix, diffusion of atomic hydrogen toward the opposite side, recombination of atomic hydrogen to form hydrogen molecules at the metal/gas interface, and desorption of hydrogen molecules. Palladium membranes allow the production of ultrapure hydrogen.

In general, the solution-diffusion model applies to nonporous membranes (dense metals, ceramics, and polymers) in which separation is a result of differences in solubility and diffusivity of permeates. The solution-diffusion model assumes thermodynamic equilibrium at the membrane interface and both solvent and solute transport through the membrane are driven by chemical potential gradients. No structural information of the membrane is included in the diffusion-solution model. Transport through porous membranes is dominated by size exclusion and is often described by the pore-flow model. This model is based on three assumptions: first, fluids on either side of the membrane are in equilibrium with the membrane at the interface, meaning that there is a continuous gradient of chemical potential across the membrane; second, the solute and solvent activity gradients across the membrane are zero and the chemical potential gradients across the membrane are zero, whereby the chemical potential gradient across the membrane can be expressed as a pressure gradient; third, straight cylindrical pores are assumed across the thickness of the membrane. No other pore structural characteristics (e.g. pore shape, pore radii distribution, and tortuosity) are accounted for. These models and their refined modifications are reviewed, for example, by Wang et al. (2014). More elaborate models are based on the extended Nernst-Planck equation improved by pore size distribution data (Tsuru et al. 1991, Bowen and Welfoot 2002a,b, Mohammad et al. 2015). These models incorporate more physical realism of the membrane process in order to better match measurable parameters. The development of rigorous physical descriptions, such as molecular dynamics simulations, has been limited due to the lack of detailed knowledge of the physical structure and electrical properties of real membranes and process streams. For the hydrogen-palladium system the situation is better. Li and Wahnström (1992) carried out molecular dynamics simulations for hydrogen diffusion in palladium. Only after coupling of hydrogen atoms to the low-lying electron-hole pair excitations among the conduction electrons, close agreement with experimental quasielastic-neutron-scattering data could be obtained (see also Jewell and Davis 2006).

Seawater desalination through membrane technology, and in particular through RO, represents one of the unique successes of membrane technology and is one of the clearest examples of PI (Fritzmann et al. 2007, Dreizin et al. 2008, Elimelech and Phillip 2011). In the coming decades, surging population growth, urban development, and industrialization will increase worldwide demand for fresh water. Early large-scale desalination plants, mostly in the Gulf countries, were based on thermal desalination, where the seawater is heated and the evaporated water is condensed to produce fresh water. Such plants consume substantial amounts of thermal and electrical energy. Therefore, over the last two decades, RO desalination plants have become the leaders in water production from seawater. In RO, seawater is pressurized against a semipermeable membrane that lets water pass through but retains salt. This technology is described in detail by, for example, Fritzmann et al. (2007). The energy consumption of RO plants could be reduced significantly over the last decades. Therefore, the widespread use of RO for desalination is due to its lower energy consumption, higher water recovery factor, and lower cost with respect to thermal plants. Additionally, the less CO₂ discharge from the energy consumed and less concentrated brine to be disposed of as waste are further advantages. The lower energy consumption is due to higher-permeability membranes, installation of energy recovery devices, and the use of more efficient pumps. An energy consumption rate of 1.8 kWh/m³ in a modern seawater RO plant could be demonstrated in a pilot-plant at 50% water recovery. The theoretical minimum energy of desalination, for example, for seawater at 35,000 ppm salt and at a typical recovery of 50% is 1.06 kWh/m³. As RO plants do not operate as a reversible thermodynamic process, the actual energy consumption, however, is larger. The energy demand for seawater desalination by state-of-the-art RO is about a factor of 2 of the theoretical minimum energy for desalination. Due to the need for extensive pretreatment and posttreatment steps, the overall energy consumption of new seawater RO plants is three to four times higher than the theoretical minimum energy demand. Therefore, further improvement of seawater RO plants should focus on the pretreatment and posttreatment (brine disposal) stages. Nevertheless, modern seawater RO plants are a real PI success compared to thermal plants. Hagedorn et al. (2017) have developed a seawater desalination plant using MDi. Permeate gap MDi (PGMD) and direct contact MDi (CDMD) spiral wound modules with different channel lengths have been used to evaluate the most suitable module configuration. DCMDi modules have shown a higher flux compared to PGMDi modules; therefore, this configuration has been chosen for further investigations. In order to develop large modules, an upscaling study was performed to reveal possible flow limitations. No limitations could be observed under sufficiently high flow velocities. A first pilot plant desalinating seawater was operated for 9 months to gain experience under real conditions. Fouling is one of the main problems in any membrane separation (van der Bruggen et al. 2008, Mohammad et al. 2015). Classical solutions to fouling are the optimization of pretreatment methods and cleaning of membranes. Suggested pretreatment methods often make use of other pressure-driven membrane separations such as UF and microfiltration. Flocculation, adsorption, sand bed filtration, backflush, forward flush, reverse flush, sonication, air sparging, chemical cleansing, and turbulence promotion by static mixers are further cleaning techniques. Of course, the membrane stability should not be reduced by these methods. Cleaning of groundwater, surface water, and waste water is also in the field of RO applications. Wastewater from textile production can also be cleaned by a combination of ultratiltration/microfiltration and nanofiltration or RO. A major contribution on the total amount of textile wastewater generated is made by the processing stages, which include scouring, dyeing, printing, finishing, and washing (Dasgupta et al. 2015). At present, most investigations on textile waste water were carried out in an academic environment. Fouling problems hampered large-scale industrial applications yet. The situation is different in the food industry (Mohammad et al. 2012). The dairy industry has been one of the pioneers in the development of UF equipment and techniques on the experience gained from its application in the production of cheese. Membrane technology is utilized in the beverage industry for processing a variety of fruit and vegetable juices. In juice clarification, UF can be used to separate juices into fibrous concentrated pulp (retentate) and a clarified fraction free of spoilage microorganisms (permeate). The pasteurized clarified fraction can then undergo nonthermal membrane concentration and, eventually, whole juice reconstitution by combination with pasteurized pulp, in order to obtain a product with improved organoleptic properties. In fish industry, UF is mainly used for fractionation and waste recovery processes. In food industry, membrane processes are advantageous because of their environmental friendliness, cost saving, and product improvement. A multitude of research activities have been carried out for production of lactic acid in fermentative process using cheap, renewable carbon sources (Pal et al. 2009). Lactic acid is the most widely occurring hydroxy-carboxylic acid used as food preservative and acidulent. In several membrane-based studies, successful separation of the components of fermentation by micro, ultra, nano, and RO and electrodialysis membranes has been demonstrated. A combination of microfiltration and nanofiltration might be able to reach commercial production of monomer grade lactic acid. As yet, there seem to be no large-scale membrane plants for lactic acid production.

MDi is a well-known membrane contactor unit (Khayet and Matsuura 2011, Alkihudhiri et al. 2012, Drioli et al. 2012). MDi is a thermally-driven separation process consisting of the combination of water evaporation and vapor condensation in an integrated process. Only vapor molecules are able to pass through a hydrophobic membrane. MDi separation is driven by the vapor pressure difference existing between the porous hydrophobic membrane surfaces. The operating temperatures in MDi are low compared to conventional distillation of evaporation processes as the solution is not necessarily heated up to the boiling point. Moreover, the hydrostatic pressure encountered in MDi is lower than that used in pressure-driven membrane processes like RO or UF. Therefore, the mechanical membrane properties are less demanding. The porous hydrophobic membrane, which is not wetted by the process liquids due to surface tension forces, does not alter the vapor/liquid equilibrium of the involved species. Condensation inside the pores does not occur. A 100% rejection of the nonvolatile solute can theoretically be achieved. Mass transfer in MDi occurs through the pores of the membrane, whereas the heat is transferred through both the membrane matrix and its pores. The heat transfer within the membrane is due to the latent heat accompanying vapor or gas flux and the heat transferred by conduction across both the membrane material and the gas-filled pores. There are boundary layers on the feed and permeate sides of the membrane, giving rise to the so-called temperature polarization and concentration polarization. The transport of gases and vapors through the pores can be described by the Stefan-Maxwell approach. Microporous MDi membranes are available in tubular, capillary of flat sheet forms. Spiral wound membranes are also available. In this type, flat sheet membrane and spacers are enveloped and rolled around a perforated central collection tube. The feed moves across the membrane surface in an axial direction, while the permeate flows radially to the center and exits through the collection tube. Common materials are polytetrafluoroethylene (PTFE), polypropylene, polyethylene, or polyvinylidene fluoride. The pore sizes are larger than in RO, which results in lower fouling problems. In terms of permeate collection and driving force generation, MDi can be classified into four categories:

• Direct contact MDi, where the hot and cold fluid are in direct contact with the membrane surface. This version of MDi has been widely studied.

• Vacuum MDi, where a pump is used to create a vacuum in the permeate membrane side. Condensation takes place outside the membrane module.

• Sweeping gas MDi, where inert gas is used to sweep the vapor at the permeate membrane side to condense outside the module.

• Air gap MDi, where the feed solution is in direct contact with the hot side of the membrane surface only. Stagnant air is introduced between the membrane and the condensation surface. The vapor crosses the air gap to condense over the cold surface inside the membrane unit. The advantage of this design is the reduced heat loss by conduction.

Although there are hundreds of papers on academic investigations, which show promising results in various applications, only a very few solar-assisted MDi demonstration systems for water production are in operation (Shirazi and Kargari 2015). A comprehensive overview of the MDi technology and its applications has been presented in the books by Khayet and Matsuura (2011) and Drioli and Giorno (2010).

Far better, with respect to industrial applications is the situation for GS in the chemical industry (Bernardo et al. 2009). A prominent example is hydrogen recovery as a first large-scale commercial application of membrane GS technology (Ockwig and Nenoff 2007). Currently, the most important membrane applications in the petroleum industry are employed in nitrogen production, hydrogen recovery, natural gas sweetening, nitrogen removal, enhanced oil recovery via CO₂, monomer recovery in polyolefin production, pervaporation processes, and organic solvent nanofiltration. The commercial success of membrane technology started in the mid-1970s of the Permea polysulfone hollow-fiber prism system for in-process recycling of hydrogen from ammonia purge gases. As the ammonia reactor operates at 130 bar, the necessary driving force for N₂/H₂ separation is available. This technology has been extended to H₂/CO or H₂/CH₄ separations. Hydrogen demand is increasing owing to new environmental regulations. Hydrotreating, hydrocracking, and hydrodesulfurization are typical hydrogen consumers. The investment costs for membrane processes are lower compared to PSA or cryogenic separation. Polyimide membranes are successfully applied for H₂ recovery in refineries due to their stability and high separation factors (H₂/N₂ ~100–200). Hydrogen production in an MR and integrated GS reduces the energy consumption and investment costs. An example mentioned in an edited book by Drioli and Barbieri (2011) is presented in Figure 24, where a conventional production plant and an intensified hydrogen production plant are compared.

Figure 24:

Hydrogen production from natural gas; (A) conventional plant; (B) integrated membrane plant (Drioli et al. 2012).

One of the largest membrane processes in use is the nitrogen production from air. Membranes are dominating applications of less than 50 tons/day and relatively low purity (0.5–5% O₂). Thousands of compact on-site membrane systems generating nitrogen gas are installed in offshore and petrochemical industry. Mostly, single-stage membrane operation is employed. Air is pressurized and fed into the membrane separators where the “faster” gases O₂, CO₂, and H₂O vapor permeate through the polymeric fibers. These gases are collected and vented to the atmosphere, while the slower, nonpermeate N₂ is available at the other end of the separator. Nitrogen is used to prevent fires and explosions in tanks and piping systems and to prevent equipment degradation during shutdown periods, in compressors, pipelines, and reactors.

Since CO₂ reduces the heating value of natural gas, is corrosive, and freezes at relatively high temperature, forming blocks of dry ice that can clog equipment lines and damage pumps, CO₂ removal from natural gas is mandatory. Membrane technology is possible for CO₂ and H₂S separations as the high wellhead gas pressure can be used as driving force for separation, and many membrane materials are very permeable to these species. For example, cellulose triacetate (Cynara-NATCO) is taken in hollow-fiber modules. PRISM (Air Products) is another commercial membrane for CO₂/H₂S removal. Membrane systems are typically installed for small size applications (less than 6000 Nm³/h) and remote locations. Further applications may be found in the paper by Bernardo et al. (2009).

A combination of distillation or reaction with pervaporation to form a hybrid separation process is an attractive PI process, as reduction of energy consumption, improvement of product quality, and less investment costs can be achieved. An example of a combination of distillation with pervaporation is presented in Figure 25 (Lipnizki et al. 1999, Suk and Matsuura 2006, Gorák and Stankiewicz 2011).

Figure 25:

Hybrid process combining distillation with pervaporation for a binary system with a low-boiling azeotrope.

Dehydration of solvents is the mostly used industrial hybrid separation process for drying organic solvents, such as alcohols. A combination of RD and membrane separation is also possible (Buchaly et al. 2007). Figure 26 (Bengtson et al. 2002) presents a hybrid process in which hydrogen and 4-chlorphenol (0.5% in H₂O) react in a poly(ether-b-amide) (PEBA) membrane, with palladium nanoclusters as catalytically active material inside the membrane, to form phenol.

Figure 26:

Catalytic pervaporation of 4-chlorophenol/Pd/Rh (Bengtson et al. 2002).

The elastomer PEBA is known as a polymer that effectively concentrates slightly polar chemicals such as phenols, higher alcohols, and the like from water. The reactive pervaporation enriched and simultaneously hydrated 4-chlorophenol compared to the feed. The enrichment factor has been 100. A careful selection of the membrane thickness, temperature (~30°C), and Pd particle size is essential.

MRs are so-called multifunctional reactors which are useful unit operation in PI, for example, in chemical and petrochemical industry, biotechnology, energy conversion, etc. Due to their multifunctionality in one unit operation, they fulfill PI requirements like lower investment costs, lower energy consumption, lower volume occupied, lower material consumption, higher conversion and/or selectivity, etc. MRs are capable of carrying out chemical reactions and separation in the same unit. The membrane selectively removes one or more products from the reaction mixture, or distributes and controls the addition of reactants in the reactor volume, or improves the contact among reactants and catalyst or nonmiscible phases. Combinations of these functions are also possible. The membranes can be selective or nonselective, with or without catalytic activity. There are many configurations of MRs to combine the membrane separation module and the reactor into a single unit. Marcano and Tsotsis (2002) have classified these configurations for catalytic MRs into six basic types:

• Catalytic MR

• Catalytic nonpermselective MR

• Packed-bed MR (PBMR)

• Packed-bed catalytic MR (PBCMR)

• Fluidized-bed MR

• Fluidized-bed catalytic MR

The most commonly utilized catalytic MR is the PBMR (see Figure 27A).

Figure 27:

Catalytic membrane reactors: (A) catalytic packed bed, inert membrane; (B) catalyst membrane.

The reaction is carried out in the packed bed of catalyst particles placed inside a membrane tube. One or more products, depending on the membrane properties, are separated by the membrane. For reversible reactions, the conversion can be increased owing to removal of products (circumventing equilibrium limitation). The membrane can be inert or also catalytically active to provide an additional catalytic function for further reactions (PBCMR). In biotechnology, reactive membrane extractors are in use, where the membrane removes inhibitory products. Dissolved biocatalysts may be flushed along the membrane or segregated within the membrane module. Immobilization (adsorption, ionic binding, covalent binding, or molecular recognition) of the biocatalyst onto the membrane is also possible. The MR may be operated as a sidestream membrane unit outside the bioreactor or as a submerged MR inside the bioreactor. The submerged membrane unit may be also put into a separate tank outside the bioreactor tank. This is the preferred version in waste water treatment because bioreactor tank and membrane tank can be optimized independently, with practical advantages for operation and maintenance.

PBMRs are often designed by one-dimensional pseudo-homogeneous models for the packed bed, whereby the mass-flow of products through the membrane is included in the mass balances of the components withdrawn by the membrane. The Ergun equation can be employed for pressure drop calculations along the tube. More elaborate CFD simulations are, of course, also possible. Heat balances for the reactions are coupled to the mass balances (see Tóta et al. 2007).

In catalytic MR, the membrane acts as a catalyst support and as a separation layer (see Figure 27B). Coating the surface of a dense membrane by catalytically active material is also possible. Both polymeric and inorganic catalytic MRs have been used. Bhat and Sudhukhan (2009) have presented a comprehensive overview of PI aspects for steam methane reforming using membranes. Further applications of MRs have been listed by Brunetti et al. (2017), Dautzenberg and Mukherjee (2001), Bengtson et al. (2002), Basile (2013), Basile and Charcosset (2016), Drioli et al. (2012), Marcano and Tsotsis (2002), and Basile and Gallucci (2011), among others. Diban et al. (2013) have reviewed investigations of MRs for in situ water removal during catalytic reactions in the food, pharmaceutical, cosmetics, and petrochemical sectors. The most comprehensive studies on water removal with MRs have been performed in esterification reactions.

Porous catalytic membranes may be also used as catalyst supports without separating function (see Figure 28).

Figure 28:

Convective flow in catalytic membranes (Fritsch et al. 2004).

The reactants are pressed through the cylindrical catalytic membrane having large pores. This results into rather large fluxes, which are controllable by setting the proper pressure drop over the membrane. The advantages of this mode of operation is the controllable catalyst contact time, which is useful for consecutive reactions, and a smaller amount of catalyst for a given conversion owing to the convective flux inside the pores instead of almost uncontrollable diffusional transport (Fritsch et al. 2004). Modeling of such reactors may be found in Garayhi et al. (1998).

In general, it is noticeable in PI by means of membranes that, except for the few applications mentioned above, academic investigations predominate work on membranes compared to industrial applications.

2.3.6 Continuous chromatography

A combination of continuous separation and reaction is the simulated moving bed reactor (SMBR). This technology allows the integration of reaction in the merely separative simulated moving bed (SMB) process. SMB provides continuous operation of normally discontinuous chromatography. The combination with reaction enables overcoming equilibrium reaction limitations or achieving thermal coupling (Lode et al. 2001). A comprehensive book by Rodrigues (2015) and Ludanes et al. (2013), and reviews by Gomes and Rodrigues (2012), Lode et al. (2001), and by Juza et al. (2000) demonstrate the features and applications of this technology. Modeling of SMBs has been described by Johannsen (2007). Broughton and Gerold (1961) introduced SMB by a first universal oil products (UOP, American company) patent. This technology was originally developed in the areas of petroleum refining and petrochemicals and become known as the Sorbex process (Parex/Molex Process) licensed by UOP. The idea behind SMB is depicted in Figure 29A.

Figure 29:

(A) A four-section simulated moving bed (SMB) unit rhs: conc. profiles of A and B in zones; (B) schematic representation of an SMBR unit with four sections and three columns per section considering a reaction of type A+B⇔C+D, in which C is the less-adsorbed product and D is the more adsorbed one (Rodrigues 2015).

A series of connected columns filled with fixed solid adsorbents is used. To mimic the countercurrent flow of solid and fluid, the positions of the two inlet and the two outlet ports are shifted periodically by one column after a certain shift time in the same direction as the fluid flow. In order to make the discrete movement closely simulate the continuous movement, which is implemented in a true moving bed (TMB), each section of a TMB unit is divided into several subsections. In Figure 29B, three subsections are shown. Owing to its periodic character, the SMB process never settles down into a steady state. However, after a certain number of shifts, a cyclic steady state is reached. This state is characterized by identical concentration profiles in each cycle. an example of internal concentration profiles is presented on the rhs of Figure 29A. These profiles are found in the middle of the shift period. A and B can be collected continuously with high purity at the raffinate and extract ports. Seidel-Morgenstern et al. (2008) summarized the features of classical SMB processes as follows:

• Constant operating parameters are applied (flow rates, feed concentrations, and column configurations).

• There are at least four columns in four different zones.

• Isocratic operation is used (i.e. the same elution strength throughout the entire unit).

• There are direct connections between the columns.

• High product purity is the goal.

• Permanent collection of the products at the raffinate and/or extract ports is realized.

• Two outlet (product) fractions can be controlled.

In the Sorbex SMB technology, a rotary valve is used to periodically change the position of the eluent, extract, feed, and raffinate lines along the adsorbent bed. At any particular moment, only four lines between the rotary valve and the adsorbent bed are active. Examples from the petrochemical industries are the Parex units for the separation of p-xylene from its isomers on zeolites, the Molex for the separation n-paraffins from branched and cyclic hydrocarbons, and the Olex process to separate olefins from paraffins. The sugar industry employs the Sarex process for the separation of fructose from glucose. SMB is now a well-established technology in the separation of fine chemicals with several SMB units installed in the pharmaceutical industries for the production of single-enantiomer drugs, for example, Bayer, Merck, Carbogen, GlaxoSmithKline, Novartis, Pfizer, and UCB Pharma. Papathanasiou et al. (2017) have developed advanced control strategies of the multicolumn countercurrent solvent gradient purification process, an industrial, semicontinuous, chromatographic process, used for the purification of several biomolecules. The authors have presented a novel control approach that manages to drive the process toward continuous, sustainable operation. The presented controllers have been designed within the Parametric Optimization and Control (PAROC) framework/software platform that enables the development of intelligent, model-based controllers through a step-by-step approach. Since the 1990s, new SMB modes of operation have been developed (Seidel-Morgenstern et al. 2008, Rodrigues 2015). Other companies developed alternative SMB schemes. For example, the Institut Francais du Pétrole (Rueil-Malmaison, France) has introduced a commercial SMB process in which the rotary valve was replaced by a combination of commonly used chromatography columns and commercial valves (Rodrigues 2015). Many new modes of SMB operation have been developed, such as Varicol process, MultiFeed, Outlet Streams Swing, M3C process, and JO process, among others (Rodrigues 2015). Besides separations of binary mixtures, also ternary separators have been investigated (see, for example, Hur and Wankat 2005).

An extension of the SMB is the SMBR, in which the continuous countercurrent chromatographic separation is combined with chemical reaction. It also comprises fixed-bed columns arranged in series and in a close circuit, but these columns are packed with an adsorbent loaded with a catalyst or a mixture of adsorbent and catalyst particles. In the SMBR, instead of feeding the components to be separated, the reactants are fed and react at the catalyst/adsorbent. The products are simultaneously separated and collected in the outlet streams (extract and raffinate) diluted in desorbent (usually one of the reactants is used as desorbent). For instance, if the feed comprised two reactants (A and B), in which A is used as desorbent, and A and B react reversibly to give two products C and D, the latter being more adsorbed than the former, then a mixture of D and A is obtained in the extract and a mixture of C and A in the raffinate (see Figure 29B). The inlet/outlet streams divide the unit into four different sections, each one with a specific role and having a specified number of columns (Rodrigues 2015):

• Section 1, located between desorbent and extract nodes, in which the solid is regenerated by desorption of the most-adsorbed product (D) using the desorbent (A);

• Section 2, located between the extract of feed nodes;

• Section 3, located between the feed and raffinate nodes, in which the reaction takes place and the products (C and D) are separated as they are formed; and

• Section 4, located between the raffinate and desorbent nodes, in which the desorbent (A), before being recycled to Section 1, is regenerated by adsorption of the less-returned product (C).

Mazzotti et al. (1996) have presented results about esterification of acetic acid with ETOH on a highly cross-linked sulphonic ion-exchange resin (Amberlyst 15) in a continuous SMBR laboratory unit. The resins acted simultaneously and effectively both as a catalyst as well as a selective sorbent. This allowed obtaining reaction products of high purity and, since the reaction is equilibrium limited, to achieve greater conversion with respect to that attainable without separation. The thermodynamic and kinetic descriptions of the system have been combined to develop a fully predictive mathematical model of the chromatographic reactor. Qamar et al. (2017) have derived analytical solutions and moments of a two-dimensional linear general rate model to simulate a single-solute transport in the chromatographic columns of cylindrical geometry. The derived analytical solutions were compared for verification with the numerical solutions of a high resolution flux limiting finite volume scheme. Typical case studies were analyzed (see also Qamar et al. 2016, 2014). Bashir et al. (2017) analyzed a linear general rate model of two-component liquid chromatography considering heterogeneous reactions of types A→B and A⇔B. The model equations incorporate axial dispersion, external and intra particle pore diffusion, interfacial mass transfer linear sorption kinetics, and first-order heterogeneous chemical reactions. Coupling of a flow reactor with SMB chromatography has been reported (O’Brian et al. 2012). The system can be operated continuously to provide the target compound in high yield.

Modeling of SMBs (Johannsen 2007) and SMBRs (Lode et al. 2001, Ströhlein et al. 2005, Rodrigues 2015) has been subject of intensive research over more than 50 years. Models of SMBs are based on the following assumptions:

• An axially dispersed plug flow model is used to describe the fluid flow.

• A plug flow model is used to represent the countercurrent solid flow in the TMB approach. Dispersion models are sometimes also in use.

• The adsorbent particles are considered homogeneous and mass transfer between fluid and solid is described by the linear driving force model.

• The model can handle any kind of adsorption equilibrium isotherm.

• Isothermal operation.

• No bed radial concentration gradients.

• Constant fluid velocity.

• Uniform particle size.

The model equations for the real SMB result from the mass balances over a volume element of the bed and at a particle level (Rodrigues 2015). A critical evaluation of assumptions made in SMB and SMBR models has been presented by Tondeur (1995). SMBRs require kinetic models for the chemical reactions. Constant operating temperature has to be checked. Rodrigues (2015) has described case studies of ethyl lactate and acetals production in SMBRs. The experiments and modeling activities needed are explained in detail.

2.3.7 Dynamic operations and multiphase reactors

The possibilities of PI due to instationary operation have been known for decades (Boreskov and Matros 1983, Silveston et al. 1995, Zhdanov 2004, among others). Periodic operation of catalytic reactors, multifunctional autothermal reactors operated in countercurrent or reverse-flow, periodically operated trickle-bed reactors, periodic temperature forcing of catalytic reactions, oscillatory flow continuous reactors, pulsed bubble columns, and cyclic distillation are typical examples. These modes of operation lead to higher yields, higher selectivities, and lower energy consumption.

In order to improve the reaction performance, e.g. to increase the reaction rate or selectivity with respect to desired products, large temporal perturbation of conventional (stable) kinetics via variation of reactant pressure, temperature, and feed composition of flow reversal has been studied in chemical engineering for decades (Matros and Bunimovich 1996, Matros 1996, Silveston and Hanika 2002, Emig and Liauw 2002, Zhdanov 2004). Bailey and Horn, Zolotarskii and Matros, Renken, Mordvintsev, and Si and Blackburn have published simple kinetic models that have illustrated the effect of periodic perturbation of reactant concentration on the reaction selectivity and yield (Zhdanov 2004). Zolotarskii and Matros have demonstrated that, for their proposed reaction scheme, the reaction selectivity may have a maximum at an optimal frequency comparable with the characteristic response time. Kinetic models to illustrate the benefits of periodic perturbation of reactant concentrations on the performance of catalytic reactions usually include only elementary steps forming a catalytic cycle, which runs often quite rapidly. Accordingly, one can hardly reach beneficial unsteady-state regimes especially if reactions are run in a large-scale reactor. In principle, one can employ slow side steps for optimization. Mordvintsev and Zhdanov (see Zhdanov’s review 2004) have introduced a model that uses side steps for investigations. Many mechanistic concepts presented in the literature do not describe in detail real reactions. The rate constants employed in such models are often considered to be free parameters and are chosen such that the effects predicted are large. Response to bistable reactions, such as CO oxidation on platinum, and spatiotemporal phenomena have been described by Zhdanov (2004), Larter and Showalter (2007), Ertl (2009), and Mikhailov and Ertl (2017) among others. Boreskov and Matros (1983) have outlined the main advantages of unsteady-state operation:

• Noticeable increases in selectivity and yield are possible.

• Catalyst provides not only its basic function as accelerator of chemical reactions but also that of a heat regenerator. This allows one to exclude inefficient recuperative heat-exchangers and significantly reduce the reactor cost.

• Industrial gases with low initial concentrations can be processed without heat expenditure.

• Owing to the dynamic properties of the catalyst bed in reactors operated in unsteady-state conditions, it is possible to create almost optimal conditions with respect to productivity and selectivity.

Silveston et al. (1995) have presented an overview and introduction into periodic operation of catalytic reactors (see also Silveston and Hudgins 2012). They explain the variables that arise in periodic operation, such as the period, split, amplitude, and phase lag. Several experimental setups and examples have been presented in that paper. Except for off-gas treatment, a wider application of the unsteady-state technologies has not been realized. One of the key problems is that the optimization of full-scale reactors cannot be performed explicitly in experiments. Extensive simulations, based on reliable kinetic and transport data under instationary conditions, are indispensible. A very reliable control of the reactor operation is mandatory. Additionally, one needs buffer tanks to put a damp on oscillations of reactor output.

The catalytic cleaning of petrol or diesel motor vehicle exhaust gas is based on the so-called three-way Pt-Rh catalysts, converting simultaneously the two reducing pollutants, CO, unburned hydrocarbons, and NO_x to nontoxic products, such as CO₂, H₂O, and N₂. Rhodium is used as a reduction catalyst, while platinum is employed as an oxidation catalyst. Ceria or ceria-zirconia oxides are added as oxygen storage promoters. The catalytic materials are suspended in a washcoat to disperse the active material over a large surface area. The core, on which the washcoat is distributed, is usually a ceramic monolith with a honeycomb structure. The coat must retain its surface area and prevent sintering of the catalytic metals even at high temperatures (~1000°C). The reduction of NO_x and oxidation of CO and hydrocarbons occur most efficiently when the catalytic converter receives exhaust from an engine running slightly above the stoichiometric point. Technologically, this condition is met by using an electronic engine control module, which receives a signal from an oxygen sensor for maintaining at the air/fuel ratio around the stoichiometric setpoint. Owing to the load change of the engine and the time delay in the air/fuel correction, the catalytic process operates inherently in an instationary mode (Güthenke et al. 2007, Chatterjee et al. 2011).

Kolios et al. (2000) explained in detail autothermal fixed-bed reactor concepts. The integration of reaction and heat exchange within one piece of equipment has advantages over the conventional unit operation design: heat losses can be minimized and the reactor is less sensitive to perturbations because of its inherent adaptivity. Autothermal reactors are widespread in industrial applications in the field of waste air purification. The design of autothermal reactors is a nontrivial task because the intrinsic feedback of heat gives rise to a complex parametric and dynamic behavior. A useful application of autothermal multifunctional reactors is the coupling of endothermic and exothermic reactions where the product of the endothermic reaction is the desired one (Haynes et al. 1992, Ercan and Gartside 1996, Kulkarni and Duduković 1996). Kolios et al. (2002) have developed a systematic procedure for analyzing coupling of endothermic and exothermic reactions in one multifunctional autothermal unit. A modified design employing concurrent heat exchange in the catalytic part combined with countercurrent end sections extends the limits of the optimal operating regime substantially. The most flexible design allows for the direct control of the heat release through an axially distributed fuel feed. A novel reactor design, the so-called folded plate reactor, has a separating wall between the exothermic and endothermic compartment. The separating wall consists of a single, folded sheet of metal that forms channels with rectangular cross-sections. The channel width can be made as small as approximately 1 mm, providing optimal conditions for an efficient heat exchange. Both channels are accessible from the side along the entire length of the reactor. The channels are filled with spacers that fulfill multiple functions: they support the channel walls and they act as firs for improving the heat transfer over the separating walls. Furthermore, they can be used as catalyst carriers and as static mixers in order to distribute the side stream uniformly within the main stream. The concept has been tested successfully for steam reforming reactions. Stolte et al. (2013) have introduced a novel form of dynamic operation named pulsed activation method. It can be viewed as a form of periodic operation in which very fast temperature pulsing is used to induce chemical reactions directly and locally as needed. The main goal in this method is to activate catalytic reactions at will and within a time scale such that physical transport related dynamics cannot follow. The CO oxidation over a Pt catalyst has been investigated as a test reaction. It has been observed that the higher the pulse energy, the higher the conversion of CO is. The reaction rate also has increased with increasing pulse frequency.

Beckmann and Keil (2003) could increase the yield of hydroformylation of ethylene to propionaldehyde utilizing a supported-liquid phase catalyst in a fixed-bed reactor by a factor of two. For this purpose, the hydrogen feed was varied in an optimized way whereby the ethylene and CO feed were kept constant. The effect of instationary hydrogen feed on the yield could be explained by means of the Wilkinson cycle. Harvey et al. (2001, 2003) have described an oscillatory flow reactor (OFR) consisting of tubes containing equally spaced orifice plate baffles (see Figure 30) (see also McGlone et al. 2015).

Figure 30:

Schematic of oscillatory flow reactor for biodiesel production (Harvey et al. 2003).

An oscillatory motion is superimposed upon the net flow of the process fluid, creating flow patterns conductive to efficient heat and mass transfer, while maintaining plug flow. Unlike conventional plug flow reactors, where a minimum Reynolds number must be maintained, the degree of mixing is independent of the net flow, allowing long residence times to be achieved in a reactor of greatly reduced length-to-diameter ratio. The OFR allows batch processes that need long residence times, like, for example, saponification reactions, to be converted to continuous, thereby intensifying the process. Conversion to continuous processing may improve the economics of suitable processes, as the improved mixing should generate a better product at lower residence time. The tanks-in-series model can be used to describe the OFR, whereby the number of tanks is fitted to experimental data. Harvey et al. (2003) demonstrated the OFR for biodiesel production from rapeseed oil in a pilot-scale plant (see also Qiu et al. 2010). The OFR combines a reduction of reactor volume and a better control of the reaction with enhanced safety and better selectivity and yield. Effective droplet break-up and controlled droplet size in liquid-liquid dispersion are also features of the OFR (Pereira and Ni 2001, Anxionnaz et al. 2008).

Employing a vibration exciter with the proper frequencies improves mass transfer in bubble columns considerably (Krishna et al. 2000, Ellenberger and Krishna 2002, Budzyński et al. 2017) (see Figure 31A).

Figure 31:

(A) Bubble column with vibration exciter; (B) Temporal development of air bubbles in water; amplitude of vibrator 1 mm; frequency 200 Hz; initial bubble diameter 5 mm (Kniazev and Keil 2013).

Vibrations of a membrane below the gas sparger with a suitable frequency or intermittently changing frequencies and amplitudes break up bubbles. Additionally, in the liquid boundary layer, the liquid is vigorously moved, which also improves mass transfer. The vibration frequencies preferably should hit the bubble eigenfrequencies to induce their vigorous oscillating movements. Depending on the liquid and gas properties, frequencies in the range between 50 and 500 Hz of the membrane are suitable for that purpose. Owing to Bjerknes forces, which is an additional force opposite to buoyancy, occur in a pulsed bubble column. These forces decrease the bubble velocity toward the top of the column. As the gas/liquid mixture is somewhat compressible, one does not have to accelerate the entire liquid mass inside the column. The energy input needed in a pulsed bubble column is less than in an equivalent gasified stirred tank. Bubble coalescence along the way to the top of the column is avoided (see Figure 31B) (Kniazev and Keil 2013). The larger bubbles break into smaller ones. This maintains the large gas/liquid interface. The pulsed bubble column is particularly useful for fast gas/liquid reactions, like hydrogenation of natural oil to margarine, where the size of the gas/liquid interface is rate determining. This has been demonstrated experimentally. Budzyński et al. (2017) have presented results of mass transfer and gas hold-up data and respective correlations over a wide range of operating parameters.

Forced periodic operation and induction of pulses in trickle-bed reactors can improve yield and avoid hot spots for exothermic reactions (Lange et al. 1994, 1999, 2004, Boelhouwer et al. 2001, Silveston and Hanika 2004, Nigam and Larachi 2005, among others). Trickle-bed reactors have found extensive application in the organic and petrochemical industries. Inducing continuous pulsing flow by fast periodic increases in liquid loading or in gas loading has been demonstrated experimentally. Boelhouwer et al. (2001) have made clear by experiments that by means of periodic operation of the trickle-bed reactor, it is possible to obtain pulsing flow at throughputs of liquid associated with trickle flow during steady-state operation. Enhanced mass and heat transfer rates are then considered to be due to the change in flow regime. This feed strategy to force pulse initiation is termed liquid-induced pulsing flow. The advantages associated with pulsing flow may then be utilized to improve reactor performance in terms of an increase in capacity and the elimination of hot spots, while interfacial contact times are comparable to trickle flow. An additional advantage of liquid-induced pulsing flow is the possibility to time the pulse frequency and, therefore, the time constant of the pulses. Further investigations of this group have been published in a review by Silveston and Hanika (2004). Lange et al. (1994) described the results of experiments of cyclohexene to cyclohexane and α-methylstyrene to cumene on palladium catalysts. Changes in the control variables such as feed composition, feed rate, or temperature were investigated. The aim of the study was to find out the conditions of periodic operation that would enable higher average inlet concentrations without evaporating the feed mixture and to improve time-averaged conversion. For the reaction of α-methylstyrene to cumene, Lange et al. (1999) could demonstrate that under liquid flow interruption, the time average conversion is higher than under steady-state operating conditions. This phenomenon could be explained by a mathematical model presented in their paper. Figure 32 gives an overview on different effects that influence the performance of a trickle-bed reactor (Lange et al. 2004).

Figure 32:

Overview of different effects on trickle-bed reactor performance (Lange et al. 2004).

A comparison of experimental results with simulations revealed, for the same reaction as mentioned above, that reliable chemical kinetics and mass transfer correlation under respective instationary conditions are absolutely necessary to obtain a good agreement of simulations and experiments. Nigam and Larachi (2005) explained the hysteresis behavior in pressure drop and liquid hold-up in trickle-bed reactors in terms of particle wetting properties. The limits of the present trickle-flow reactor models have been compiled by Lange (2007).

Cyclic distillation emerged as another important trend in PI (Maleta et al. 2011). Essentially, a cyclic distillation column has an operating cycle consisting of two key operations: a vapor flow period, when vapor flows upward through the column and liquid remains stationary on each plate, and second, a liquid flow period, when vapor flow is stopped, reflux and feed liquid are supplied, and liquid is dropped from each tray to the one below (Kiss 2013). As a result, the throughput of such a column using the controlled cycle mode of operation is typically more than two times higher than the conventional throughput. There are several advantages employing cyclic distillation (Kiss 2013):

• High tray efficiencies (140–300% Murphree efficiency).

• Reduced energy requirements (30–50% less).

• Increased quality of the products due to higher separation efficiency.

• Smaller pressure drops on trays compared with conventional steady-state operation.

• The cyclic distillation configuration and operation allows larger liquid holdups that can be beneficial for RD, such as catalytic distillation. Cyclic distillation is used in food industry on industrial scale for concentrating alcohol from 8 wt% to 27–45 wt% (Kiss 2013).

Silveston and Hudgins (2004) have investigated periodic temperature forcing of catalytic reactions. The experimental results seem to be rather disappointing. Dautzenberg and Mukherjee (2001) reviewed further multifunctional reactors.