Christos Georgakis

Process systems engineering tools in the pharmaceutical industry

Abstract The purpose of this paper is to provide a summary of the current state of the application of process systems engineering tools in the pharmaceutical industry. In this paper, we present the compiled results of an industrial questionnaire submitted to pharmaceutical industry professionals. The topics covered in the questionnaire include process analytics, process monitoring, plant-wide information systems, unit operation modeling, quality control, and process optimization. A futuristic view of what process systems engineering tools will enable the pharmaceutical industry will be also be presented. While the industry is regularly using the traditional Design of Experiments approach to identify key parameters and to define control spaces, these approaches result in passive control strategies that do not attempt to compensate for disturbances. Special new approaches are needed for batch processes due to their essential dependence on time-varying conditions. Lastly, we briefly describe a novel data driven modeling approach, called Design of Dynamic Experiments that enables the optimization of batch processes with respect to time-varying conditions through an example of a simulated chemical reaction process. Many more approaches of this type are needed for the calculation of the design and control spaces of the process, and the effective design of feedback systems.

A Model-Free Methodology for the Optimization of Batch Processes: Design of Dynamic Experiments

The new methodology presented provides a way to optimize the operation of a variety of batch processes (chemical, pharmaceutical, food processing, etc.) especially when at least one time-varying operating decision function needs to be selected. This methodology calculates the optimal operation without the use of an a priori model that describes in some accuracy the internal process characteristics. The approach generalizes the classical and widely used Design of Experiments (DoE), which is limited in its consideration of decision variables that are constant with time. The new approach, called the Design of Dynamic Experiments (DoDE), systematically designs experiments that explore a considerable number of dynamic signatures in the time variation of the unknown decision function(s). Constrained optimization of the interpolated response surface model, calculated from the performance of the experiments, leads to the selection of the optimal operating conditions. Two examples demonstrate the powerful utility of the method. The first examines a simple reversible reaction in a batch reactor, where the time-dependant reactor temperature is the decision function. The second example examines the optimization of a penicillin fermentation process, where the feeding profile of the substrate is the decision variable. In both cases, a finite number of experiments (4 or 16, respectively) lead to the very quick and efficient optimization of the process.

OPERABILITY OF MULTIVARIABLE NON-SQUARE SYSTEMS

Abstract Non-square process control systems, with fewer inputs than the controlled outputs, are quite common in chemical processes. In these systems, it is impossible to control all measured variables at specific set points and many of the outputs are controlled within an interval. The objective of this paper is to introduce a multivariable Operability methodology for such non-square systems to be used in the design of nonsquare constrained controllers. In order to motivate the new concepts, we examine some simple non-square systems obtained from the control system of a Steam Methane Reformer process.

Identification of reduced order average linear models from nonlinear dynamic simulations

Presents a general methodology for the identification of average linear low order models (ALLOM) from data collected from detailed nonlinear models. While there are many methods available in the literature for identifying linear models, these methods tend to produce inaccurate and ill-conditioned models when used on nonlinear data sets. The method in this paper differs from traditional linearization methods in that it better approximates the dynamic characteristics over a wider area around the reference steady state.

Dynamic Optimization of a Batch Pharmaceutical Reaction using the Design of Dynamic Experiments (DoDE): the Case of an Asymmetric Catalytic Hydrogenation Reaction

Abstract The present research work aims to demonstrate the effectiveness of the new methodology of Design of Dynamic Experiments (DoDE) in optimizing an important pharmaceutical reaction. An easily developed response surface model (RSM) is used instead of a hard to develop knowledge-driven process model. The DoDE approach allows the experimenter to introduce dynamic factors in the design, which during the RSM optimization are treated as all the other factors, simplifying the analysis significantly, leading to the rapid optimization of batch processes with respect to time-varying decision variables. The DoDE approach enables the discovery of optimal time-variant operating conditions that are better than the optimal time-invariant conditions discovered by the classical Design of Experiments (DoE) approach. In the present case of the asymmetric catalytic hydrogenation, 24 experiments are conducted for the DoDE approach and the best run results in a 45% improvement comparing to the best run of 17 runs of the DoE approach. This is achieved by applying a decreasing temperature profile during the batch reaction. Optimization of the economic performance index of the process through the respective response surface models defines an optimum operation. The DoDE optimum operation is better than the respective one through the DoE. The DoDE advantage increases as the required quality level for the final product is higher. For the medium quality, the DoDE approach results in an improvement of 30% over the DoE one.

MODEL PREDICTIVE CONTROL AND DYNAMIC OPERABILITY STUDIES IN A STIRRED TANK: RAPID TEMPERATURE CYCLING FOR CRYSTALLIZATION

A stirred-tank reactor was built with the objective of rapid and accurate temperature control in the reaction vessel. A first-principles heat transfer model was developed for the jacketed batch system, with the jacket inlet temperature used to control the vessel temperature. A model predictive controller was implemented to follow a rapidly changing temperature profile that cycled between steep heating and cooling motifs, and it was tested experimentally at progressively shorter temperature cycles. For a water-solvent-water-jacket system, a cycle consisting of increasing and decreasing the temperature by 15°C over a period of 20 min was achieved in practice. The performance of the MPC controller was explained by calculating the dynamic operability characteristics of the process.

A methodology for optimal sensor selection in chemical processes

A methodology for the optimal selection of measured variables for chemical processes is presented. This methodology determines the Pareto optimal trade-off between the process information that can be obtained and the sensor cost for the selected process measurements. The approach is demonstrated using a simulated CSTR process.

Design of Dynamic Experiments Versus Model-Based Optimization of Batch Crystallization Processes

Abstract A new data-driven optimization methodology is applied to a batch cooling crystallization simulation to evaluate how it compares with a model-based optimization technique. The method, Design of Dynamic Experiments [Georgakis, 2009], is an extension of the classical Design of Experiments approach and can be applied to any process where time-variant profiles are important for optimizing key objectives of the process. As a data-driven approach with no first-principles model required for process optimization, this methodology may be particularly useful for complex processes for which no knowledge-driven model exists or where the objective function cannot be modeled.

Optimizing Batch Crystallization Cooling Profiles: The Design of Dynamic Experiments Approach

Abstract A new data-driven methodology for optimizing time-variant profiles in batch processes without the need for a first-principles model is applied to a batch cooling crystallization to find the optimum cooling trajectory that minimizes the total amount of nucleation during the crystallization. The method, Design of Dynamic Experiments (Georgakis, 2009), is an extension of the classical Design of Experiments approach and can be applied to any process where time-variant profiles, typically batch and semi-batch operations, are important for optimizing key aspects of the process. As a data-driven approach with no first-principles model required for process optimization, this methodology may be particularly useful for complex processes for which no knowledge-driven model exists.

Analysis of the constraint characteristics of a Sheet Forming Control Problem using interval operability concepts

An Interval Operability-based approach [1, 2] is applied to calculate operable output constraints for the Sheet Forming Control Problem (SFCP) from DuPont. The SFCP attempts to control the sheet thickness at 15 different points, which represent 15 output variables, using 9 manipulated variables in the presence of 3 disturbances. Thus, this problem represents a computationally complex, high-dimensional non-square system with more outputs than inputs. The SFCP is addressed here under two study cases: 1) a non-square, where all the 15 outputs are controlled independently of each other; 2) a square, where 6 outputs are combined in pairs, or zone variables, and controlled within their corresponding zone. Results show that significant reduction of the constrained region of process operation can be achieved for different output targets specified. Specifically, the hyper-volume ratio of the initial to the designed constrained regions range between 103–105. The calculated constraints are validated by running DMCplusTM (Aspen Tech) simulations for the extreme values of the disturbances. These constraints are intended for use online in model-based controllers (e.g., Model Predictive Control) to calculate the tightness with which each of the outputs can be controlled.