Integrating PAT and process control
The next step after integrating PAT into the continuous process as a monitoring tool is to incorporate the data from PAT into
process control. In a feed-forward/feed-backward or closed-loop control system, critical process parameters (CPPs) are dynamically
adapted and fed to the controller to keep the process within specification. In today’s batch processes, unit operations are
operated as “islands of automation,” but the goal of continuous processing is to control multiple, single units as one unit
by using an overarching process-control system.
Closed-loop control has been used successfully for many years in other process industries. “The frightening part for the pharmaceutical
industry is moving from manual control with people making decisions to advanced process control, in which process data and
modeling software automatically control the process,” comments Jonathon Thompson, senior manager of Compliance Services Consulting
at Invensys. “PAT can give you a lot of data, which you need to turn into information about the process and whether it is
within the CQA parameters,” he explains. “Process-modeling software compares real-time data to an ideal or ‘golden’ batch
profile, identifies what parameters need to be changed to meet the ideal, and feeds this back into the control system.”
Researchers have been addressing several issues while integrating process control. One issue is that although some process
equipment (e.g., newer feeders) is typically already instrumented for control, other equipment may require retrofitting. The
tubular blender used at C-SOPS, for example, had a simple motor with a tachometer to control the speed, and C-SOPS engineers
added instrumentation to allow more sophisticated control of the blender speed.
Integration of instrument software with plant equipment is a challenge. Some installations that require simple endpoint or
on/off control could use simple, analog 4–20 mA connections, says Hammond, but the most valuable applications generally require
more sophisticated control that involves developing either direct communication with the manufacturing equipment’s control
systems or communication with plant supervisory control and data acquisition (SCADA) systems.
“Ten years ago these communication links were custom developed for each piece of equipment. In recent years the development
of “open architecture” software has simplified this aspect of instrument integration,” explains Hammond. Open-architecture
software (e.g., OPC) uses open standards that enable connectivity. This software meets the need for communication protocols
to enable analytical instruments to communicate with control systems, says Hammond, who notes that, after working with instrument
vendors and control-system companies over the past five years, most of Pfizer’s vendors’ instruments now communicate via OPC.
Another challenge has been ensuring that data from PAT flows into the process-control system quickly enough to enable meaningful
control of process fluctuations to keep CQAs within specification. “If the measurement takes longer than the residence time
of the material—for example, 30–45 seconds of residence time in the blender—then the measurement can not be used for process
control,” explains Hausner. He says that available NIR instruments are adequate for some applications, but faster analysis
may be needed for other applications, such as smaller doses or lower percentages of API, which would necessitate more scans
to obtain a measurement. C-SOPS continues to investigate solutions for NIR measurement in its production-scale Continuous
Pharmaceutical Advanced Manufacturing Laboratory (CpAML). Analysis costs are part of the equation, notes Hausner. For example,
a current CpAML project is comparing the use of one, fast (i.e., expensive) spectrometer with an array of slower but more
economical spectrometers. An alternative method is inferential sensing, in which offline laboratory data is used with known
correlations to give the control system the data it needs, adds Thompson.
Tracking material flow through the system is a crucial aspect of process control. A SCADA system tracks the location of a
“product plug” as it moves through the process and adds this context to the PAT data in the process-control modeling software,
explains Ivo Backx, manager of business and project development for the pharmaceutical industry at Siemens Industry Automation
Division. “In tablet production, for example, multiple variables control the CQAs and you need to know what these attributes
are and the correlation between them at specific points in the process,” says Backx. “For example, you need to know that the
material in the tablet press now had a certain content uniformity at the blender.” Material tracking through the system is
linked to requirements for traceability, notes Backx.
Traceability is simple in a batch system because it is assumed that all the product from one batch is the same. “For a continuous
operation, traceability becomes much more important because not all the product is submitted to the same process at the same
time. Using a first-in/first-out (FIFO) principle is, therefore, very important,” says Schoeters, who notes that GEA has adapted
the unit operations of mixing, granulation, drying, compression, and coating in such a way that the product is traceable throughout
the production line and back-mixing is limited as much as possible. The ConsiGma system, such as the line shown in Figure 2, continuously monitors CPPs to keep them in control and maintain product quality. This includes measurements using PAT (e.g.,
particle size, moisture, uniformity), but primarily involves monitoring of and control-feedback loops on machine parameters.
Torque of the granulator screws, for example, is continuously monitored, and a deviation from the set value triggers actions
Figure 2: A continuous, high-shear granulation and drying system
in operation at the GEA test center (ConsiGma, GEA Pharma Systems).