Applying Supervisory Process Control Practices to Continuous Pharmaceutical Manufacturing

Successful supervisory process control implementations in the pharmaceutical industry have been hampered due to the batchwise nature of most process operations. In fact, batch processes present some specific challenges because the main steps operate discontinuously. Consequently, temperatures, concentrations and other properties vary in time which requires complex non-linear and/or time-varying models and control strategies. In contrast, continuous manufacturing is one step closer to the “holy grail” for process control engineers, i.e. a linear time invariant (LTI) system. The availability of common practices and numerous examples for such a LTI system representation significantly reduce the complexity of process identification and control implementation. Hence, dynamic process operation with feedback and automatic manipulation of quality attributes is proposed as the way forward for continuous manufacturing [1]. A technical framework for this approach is already available in continuous oil refining, and in the chemical and food industries.

In order for the pharmaceutical industry to embrace fully continuous manufacturing with supervisory process control, numerous challenges still exist. This said, the current work tries to face them and apply existing classical and modern continuous control laws to manipulate quality attributes of a commercial continuous wet granulation unit within a continuous from-powder-to-tablet installation, being the ConsiGma^TM-25. The idea is to have stable automatic control of species concentration in the granules formed during twin-screw wet granulation. In this case the water content is considered, as it is directly related to both chemical stability and physical characteristics of the granules.

A first encounter in this effort was the inline estimation of moisture concentration from the particulate stream using the possibilities offered by high-speed NIR measurements. This includes process-instrument interfacing, industrial data communication, time-alignment of instrumentation and process data, latent variable modeling and signal filtering. These steps are typically highly customised to the situation, but some general comments will be presented. In addition, control of a process requires some form of model, and here a black-box mathematical model, of the type auto-regressive with moving average and exogenous inputs (ARMAX), was identified to predict the fast dynamic behaviour from system actuator to the actual measurement. This allowed us to tune, simulate and validate empirical (proportional-integral) as well as model-based (extended prediction self-adaptive) controllers for the granule moisture concentration loop under study. Another key point that will be addressed is that real-time calculations, here for analysis and control, require computationally inexpensive methods to keep up with the dynamics of the system under study.

[1] Myerson, A. S., Krumme, M., Nasr, M., Thomas, H. and Braatz, R. D. (2015), Control Systems Engineering in Continuous Pharmaceutical Manufacturing. May 20–21, 2014 Continuous Manufacturing Symposium. J. Pharm. Sci., 104: 832–839. doi: 10.1002/jps.24311