M.V. Le Lann

Adaptive control of a multipurpose and flexible semi-batch pilot plant reactor

Abstract Semi-batch operation is currently used in the pharmaceutical or fine chemical industry. Control of such an operation is difficult to ensure by conventional PID as operating conditions and operations change a lot even for the same apparatus. So it is very important in the fine chemical industry to maintain the flexible and multipurpose characters of batch reactors during their automation. This study deals with adaptive thermal control of different kinds of operation (heating, cooling, exothermic reactions carried out in a semi-batch jacketed pilot reactor. Experimental results show the robustness of the algorithm in respect to strong disturbances, its ability of tracking high nonlinear setpoint trajectories and its good control qualities during various chemical reactions.

Use of neural networks for liquid-liquid extraction column modelling: an experimental study

Abstract This paper presents a new application of neural networks to the modelling of a chemical pilot plant: a pulsed liquid–liquid extraction column. This separation process presents a highly non-linear behaviour and time-varying dynamics. Usually, physical simulation models of chemical plants describing some aspects of hydrodynamics and mass transfer are static or very complex and need excessive computer time. It is proposed that improved predictions can be obtained using a multilayer artificial neural network instead of the physical model of the process. The results obtained illustrate the successful application of such a neural network modelling approach.

Adaptive Model Predictive Control

This paper gives an overview of our studies on adaptive control performed in our laboratory for more than ten years enlightened by results of practical applications on different pilot plants. Applications have been made in two fields: continuous processes such as liquid-liquid extraction columns and batch processes, typically batch or semibatch reactors. Formerly a classical adaptive controller based on a “black-box” model: the Generalized Predictive Controller with Model Reference (GPCMR) has been used. Results were very good in the case of continuous processes but showed the limits of this type of algorithm when applied to batch processes with large changes of dynamics and long time delay. So, a special part is devoted to focus on the important advantages and improvements brought by the use of an Adaptive Model-based Controller in comparison with the classical GPCMR.

Realistic model-based predictive and adaptive control of batch reactor

Abstract This paper presents an original approach which makes use of a realistic thermal model for adaptive predictive control and supervision of a semibatch jacketed reactor. The first part is devoted to the development of an adaptive controller based on a reaUstic model established by thermal balances on the reactor and its jacket. As a first interest, an on-line deterministic estimator of the heat generated during the reaction is developed. In a second part a realistic model is used for model supported supervision to prevent temperature overshoot, especially in the case of industrial reactors, the temperature control being ensured by the adaptive controller.

Optimisation and scale-up of batch chemical reactors: Impact of safety constraints

This work presents a methodology in order to determine the operating conditions simultaneously optimising the chemical yield and considering the safety aspect. The aim is to determine time evolutions of the reacting mixture temperature and the feed rate of reactants which not only define a maximal efficiency but also involve an adequate generated heat which can be removed by the cooling system. The problem is converted into a non-linear programming problem which is solved by standard Non Linear Programming. The methodology is validated on an industrial synthesis of a pharmaceutical product which presents high exothermicity. Results show the influence of the safety and economical constraints on the determination of optimal operating conditions. Especially, when no constraints are considered, the optimal solution is to operate batch-wise. When the generated heat is limited, the reactant has to be introduced according to a specified feed-rate profile. Experiments carried-out in a 1 litre pilot plant reactor confirm the optimisation results. This reactor is equipped with a control methodology based on temperature control through pressure which allows to remove a large amount of heat through the condenser. Then, scale-up to a 8 m3 glass-line jacketed reactor shows the relevance of the constrained optimisation approach.

Optimization technique based on learning automata

Optimization techniques are finding increasingly numerous applications in process design, in parallel to the increase of computer sophistication. The process synthesis problem can be stated as a largescale constrained optimization problem involving numerous local optima and presenting a nonlinear and nonconvex character. To solve this kind of problem, the classical optimization methods can lead to analytical and numerical difficulties. This paper describes the feasibility of an optimization technique based on learning systems which can take into consideration all the prior information concerning the process to be optimized and improve their behavior with time. This information generally occurs in a very complex analytical, empirical, or know-how form. Computer simulations related to chemical engineering problems (benzene chlorination, distillation sequence) and numerical examples are presented. The results illustrate both the performance and the implementation simplicity of this method.

Constrained optimization for fine chemical productions in batch reactors

Abstract Batch and semibatch reactors are the main device of batch processes which are still widely used to produce pharmaceuticals, polymers, biotechnologicals etc. Fine chemical transformations are characterized by quite complex reaction systems which can produce undesirable end products. As the aim of the fine chemical industry is to produce high quality and purity products, it is essential to optimize batch operating conditions, i.e. temperature profile, feed flow rate, amount of reactant and final batch time, taking into account constraints on the experimental feasibility (heating and cooling rates) or the specificities of fine chemistry productions (purity constraints, …). In this work, the optimal control problem is converted into a non-linear programming problem solved by the generalized reduced gradient procedure coupled with the golden search method, for the search of the total batch time. The efficiency of the methodology is shown by its application to different formulations of the problem for different chemical reaction schemes and with stress laid on the influence of the constraints on the limitation of temperature variations and byproduct formation.

Elaboration of a neural network system for semi-batch reactor temperature control: an experimental study

Abstract A neural controller is developed and used to regulate the temperature in a semi-batch pilot-plant reactor. The experimental unit employs a monofluid heating-cooling system to raise the reactor to the operating temperature, and then remove the heat generated by an exothermic reaction. A methodology for installation of the neural controller is presented. Preliminary experiments form the neural-network learning database. These involve regulation of the reactor by either an advanced control algorithm or by operator (manual control). The results suggest that such neural controllers can provide excellent setpoint-tracking and disturbance rejection.

GENERALIZED PREDICTIVE CONTROL OF A PULSED LIQUID-LIQUID EXTRACTION COLUMN

This paper deals with the application of a general predictive controller to a pulsed liquid-liquid extraction column. The control purpose is to maintain the column in its optimal behaviour zone in spite or flowrates and physical properties of solvent and solute fluctuations. The complex dynamics of the column is modeled by a low order linear discrete model with time varying parameters which are recursively identified. Based on these estimates, the control policy is adapted on line. The obtained results illustrate the successful application of such an adaptive algorithm.

Design of a neural controller by inverse modelling

Abstract This paper deals with the development a neural controller (a control system using a neural network) and its application for the temperature control of an experimental semi-batch pilot-plant reactor equipped with a monofluid heating-cooling system. The neural controller design methodology is based on the process inverse dynamics modelling : the learning data base is generated in an open-loop structure and the learning of the neural network is carried out by considering the future process outputs as the reference set-point. The first results presented deal with an ideal simulated system modelled by a first order system. They demonstrate the importance to take the time-delay of the plant into account. The second part is concerned with the real time application of such a technique to the temperature control of a semi-batch pilot- plant reactor and shows the real capability of the neutral networks in process control.