**2. Reactive distillation dynamics and control**

Non linearity of reactive distillation was first reported by Ciric and Gu et al. [1] using simulation of reactive distillation processes involving simultaneous solution of material and energy balances and stoichiometric relationship which corresponds to the solution of a considerable large set of non-linear modeling equations of reactive distillation column. They have used a mixed integer nonlinear programming (MINLP) approach was used to synthesize an optimum reactive distillation column. The MINLP minimizes the total annual cost subject to a MESH model. The solution of this MINLP yields the optimal number of trays, the tray holdups, the feed tray locations, and their feed distribution. Amte et al. [2] presented a MINLP optimization technique that would assist to identify a suitable configuration for selectivity maximization at conceptual design level. Results obtained through MINLP gives a good agreement with those obtained by performing independent simulation using ASPEN PLUS. In this work, authors have considered feed and catalyst tray location, reflux as the variables for the maximization of selectivity. Thus, MINLP optimization process proves conceptual design for the selectivity engineering with reactive distillation. Doherty et al. [3] have given pioneering contributions to the analysis and design of the reactive distillation and developed thermodynamically based approach for analyzing equilibrium limited, thermally neutral reactive distillation systems. This work employed a novel composition coordinate system to transform the problem into a form completely analogous to nonreactive distillation.

Subawalla and Fair et al. [4] worked on dynamic study of TAME synthesis in reactive distillation column by considering different design parameters such as number of trays, feed flow rate etc., resulting into nonlinear interaction of these input parameters. He has also provided some intuitive guidelines about coupling of these design parameters order. Schenk et al. [5] have presented equilibrium and non-equilibrium models for predicting the steady state and dynamic behavior of RDC based on a rate-based model in which mass transfer rates between liquid and vapor phase are considered explicitly, based on the Maxwell-Stefan equations. A switching policy makes it possible to switch from one model to the other, based on the knowledge gained, by following the Gibbs free energy as a function of time. Cardoso et al. [6] have proposed a new simulation/optimization model for the Mixed Integer Nonlinear Programming (MINLP) formulation of reactive distillation columns as used by Ciric and Gu [1], where the simulation algorithm is based on conventional distillation and optimization is performed by stochastic algorithm. Estimate of the initial composition profile in the column was obtained by relaxation method and material balance equations are solved by Newton–Raphson method, to compute the composition profile. Vora and Daoutidis et al. [7] represented work on dynamics of reactive distillation by proposing a different feed configuration for two feed and two product system. They have also presented control of an ethyl acetate reactive distillation system to achieve higher conversion than the previously proposed configuration which involves single feed reactive distillation column. Taylor and Krishna et al. [8] have presented work on modeling of homogeneous and heterogeneous reactive distillation processes by considering equilibrium stage models, non-equilibrium stage modeling, the conventional NEQ model, NEQ modeling, hydrodynamics, and mass transfer. Dynamic simulation for NEQ model without back mixing was also presented by Kataria et al. [9] in which model equation for reactive distillation without back mixing was developed for open loop system. For the NEQ model with partial mixing, bifurcation behavior with stability analysis and simulation were carried out which confirmed the existence of oscillation in the NEQ

#### *A Review on AI Control of Reactive Distillation for Various Applications DOI: http://dx.doi.org/10.5772/intechopen.94023*

model as well. Once the nonlinear nature of reactive distillation is confirmed, various control strategies were reported by the authors to make the system linear. Linearization was carried out to make system to oscillate around the set points of desired parameters which makes the system easy to operate without much affecting the performance over the time. Mehran et al. [10] have employed Takagi-Sugeno fuzzy model to express the local dynamics of each fuzzy implication by a linear system model to plot the resultant error of surface with time and reported that although system's non linearity can be linearize to some extent but still many control problems involve uncertainties in the model. In view of this, Gruner et al. [11] have taken an industrial case of reactive distillation and proposed a nonlinear control scheme for the system operated by Bayer AG.

Interaction of various parameters in a reactive distillation leads to higher order nonlinearity and dynamics like interaction between number of trays, reflux ratio, reboiler duty, etc. This interaction will lead to multiple steady states and multiplicity in the system. The type of multiplicity will limit the values which should be broken by application of suitable controllers. Heath et al. [12] studied the interactions of design variables and applied control schemes for an ethyl glycol reactive distillation system and assumed that the process variables are fixed i.e. no revamping is done. Author has taken process economics as a major issue which ultimately depends on the number of variables interacting in a system and its control was achieved as per the desired objective. For the same assumptions, Schenk et al., [13] and Georgiadis et al. [14] have worked on interaction of design variables in a reactive distillation column and proposed advanced optimization technique for the control of a reactive distillation system to smoothen the operability and increase the system efficiency.

**Figure 1** shows general control scheme of reactive distillation column. The common controls used are Flow Control (FC), Level Control and Composition Control (CC). Level control is given in condenser and reboiler while composition control is provided to condenser, reboiler and one of the feed flow rates to get tighter composition analysis. Flow control is provided to feed flow rate at the top of the reactive section.

**Figure 1.** *Reactive distillation control.*

## **2.1 Control using conventional techniques**

Simplified models clarify process dynamics but cannot represent the process under wide range of operating condition. Thus, various mathematically driven controllers like proportional, integral, derivative or a combination of theseareusedas hardware sensor-controller system which are mathematically functioned or scripted. Furthermore, the RD process contains a large degree of uncertainties, which cannot be well described using single mathematical expression. Therefore, techniques without using exact process models are more attractive for RD control. Monroy-Loperena et al. [15] studied the control problem of reactive distillation system for ethyl glycol by proposing a robust PI control configuration. They have also revealed the existence of input multiplicity in the system and proposed a first order output feedback control system to regulate the product composition. The control design involves interpretation of error signal whose dynamic is again constructed based on available data. Sneesby et al. [16] moved in process control by highlighting an integrated control scheme by taking ETBE as the case column. He proposed to change control objective to reflect changing economic variables like starting from optimum purity, minimum number of trays, optimum reflux ratio, etc. In view of this author has also presented a rigorous MESH based modeling to represent the main chemical reaction. Al-Arfaj & Luyben et al. [17] applied different conventional PI, PID and other conventional controller scheme for reactive distillation as well as for simple reactor. They compared the control of both these systems to produce methyl acetate. Various control strategy was proposed in this paper, the first one was for a composition and temperature control while second scheme was based on tow temperature controllers. A comparison between these schemes shows that different scheme corresponds to over design or under design system hence proper balancing of degree of freedom of a system is equally important.

Various tuning methods are proposed in the literature to calculate the ultimate period or ultimate gain. Chandra et al. [18] have calculated ultimate gain and ultimate period using Ziegler Nicolas tuning rule for an ARX model structure. The objective was to control the desired product purity in distillate stream. In this work, an ARX model structure relates the plant output with present and past plant input output to formulate a predictive control. Recursive least square estimator was used which provides updated parameters to ARX model. Goyal et al. [19] have presented support vector regression to tune a PID controller. Model gain scheduling was included in one of the control strategies for reactive distillation. Temperature control was given priority because to balance the stoichiometry, temperature of feed trays can be used to adjust the fresh feed streams. For this the gain of controller was define as the change in temperature with respect to the feed flow rate. Nizami et al. [20] have constructed one or two loop composition PID controller, however, the conventional controllers applied for the control of reactive distillation was not capable of actually control the simultaneous interacting parameters because of occurrence of reaction and separation in single column. Lei et al. [21] have described the design and optimization of reactive distillation column for the synthesis of Tert-Amyl Ethyl Ether (TAEE), the temperature–composition cascade strategy was proposed to control the Reactive Distillation (RD) process for the synthesis of TAEE. In those optimized conditions, the proposed control strategy was introduced to manage the RD process by changing the sensitive variables. Dimian et al. [22] have carried out thermodynamic analysis in residue curve map and simulation of reactive distillation column. Process dynamic and control was considered in detail to design the column. The feasibility of fatty acid esterification

#### *A Review on AI Control of Reactive Distillation for Various Applications DOI: http://dx.doi.org/10.5772/intechopen.94023*

with individual alcohol was studied by means of residue curve maps. By sensitivity analysis it was found that the reflux of heavy alcohol is the key manipulating variable for controlling the water content in the reactive liquid phase.

The controllability of the system could be studied using tools from control theory. Feedback control of inventory is measured, and a feedback control loop is implemented with the fresh feed as manipulating variable. Konakom et al. [23] have proposed to control distillate rate subject to a given product purity constraints. A conventional batch reactive distillation model described by a system of differential algebraic equations is formulated and solved using an optimal control algorithm. In open loop simulation of production of industrial grade ethyl acetate of 90% purity, dynamic optimization programming was implemented which increases the purity as per the product specification.
