Fundamental models and process control

Thomas F. Edgar, Department of Chemical Engineering, University of Texas at Austin, tfedgar@austin.utexas.edu -- Control Engineering, 11/1/2006

Fundamental dynamic models play a central role in process dynamics and control. Such models can be used to:

• Improve understanding of the process. Dynamic models and computer simulation allow transient process behavior to be investigated without having to disturb the actual process;
• Train plant operating personnel. Process simulators play a critical role in training plant operators to run complex units and to deal with emergency situations;
• Develop a control strategy for the process. A dynamic model of the process can help identify the process variables that should be controlled and those that should be manipulated; and,
• Optimize process operating conditions. To maximize profit or minimize cost, a steady-state process model and economic information can be used to determine the most profitable operating conditions.

The fundamental modeling of chemical and biological processes using first principles usually leads to mathematical models that are systems of differential and algebraic equations, typically material and energy balances plus the physical, chemical, and thermodynamic relationships of the system. Although commercial software packages contain the required physical property libraries and computational tools to simulate the equations, dynamic modeling packages are not used widely in industry, except as “point” solutions.

Dynamic modeling tools do not have the same level of acceptance as steady-state simulation packages. One long-term deficiency of dynamic model software in the chemical and petrochemical industries is that the steady state package is not derived from the dynamic package by setting time derivatives to zero. So normally users must learn how to use two different packages to perform steady-state and dynamic simulation.

The quandary of an engineer who must develop a dynamic physicochemical or biological model to use in process control is that a very large range of possible models can be used, from simple to complex. Every model incorporates assumptions that must be made by the modeler, who usually does not know a priori the impact of the assumptions on model accuracy or control quality. If the model is too complex, then the computation time for control calculations may be prohibitive, in the range of several hours, for a process that responds with time constants on the order of several minutes. The relatively slow simulation speed can also be a problem for operating training. Using rigorous models for design, control, data reconciliation, and optimization has been trending upwards in the process industries, but the shortage of experts in process modeling in most companies is an impediment.

In pharmaceutical manufacturing, the use of fundamental models is rare, due to a lack of confidence in the accuracy and/or lack of maintainability of such models. In an effort to increase the safety, efficiency, and affordability of medicines, the FDA has recently proposed a new framework for the regulation of pharmaceutical development, manufacturing, and quality assurance. PAT (Process Analytical Technology) has become an acronym in the pharma industry for designing, analyzing, and controlling manufacturing processes to reduce variability. Process variations that could contribute to patient risk are determined through modeling and timely measurements of critical quality attributes, which are then addressed by process control.

An emerging opportunity is the use of dynamic first principle models with accurate mass and energy balances and kinetics that includes the activities of the nutrients and cells as well as the relationship between intercellular metabolic pathways and extracellular parameters. These models can provide real-time inferential measurements of bioreactor composition without the noise and delay associated with analytical results obtained from an on-site laboratory.

The models can be run faster than real-time for rapid testing and the design of experiments and can be used for batch profile optimization, batch performance monitoring, and model identification. It is expected that PAT will begin to have an impact during the next several years.