CG1204-CTtalk-lab
CG1204-CTtalk-lab
CG1204-CTtalk-lab
CG1204-CTtalk-lab
CG1204-CTtalk-lab

New Paradigms for Lab Control Systems

March 29, 2012
What a Lab Control Systems Can and Should Do for Commercialization of Biopharmaceuticals
About the Authors
Greg McMillan and Stan Weiner bring their wits and more than 80 years of process control experience to bear on your questions, comments and problems. Write to them at [email protected].Stan: My first job was with an Engineering Consulting firm in downtown Philadelphia. We designed and built a synthetic rubber plant for Firestone in Orange, Texas. They hired me as a chemical process engineer and switched me to an instrument engineer on Day 1 because they had no instrument group. After the design was completed, I went to the field for construction supervision. I left because I was planning to get married and was concerned about long term security. I went to Rohm & Haas Chemical Company in their instrument research lab for a couple of years. Then they transferred me to the Engineering Department because they didn't have an instrument department. So I have the whole picture of the commercialization of a process from the lab, through design, and ending up in construction.

Greg: Scott Broadley, president of Broadley-James Corp., broke wide open what our view of what a lab control systems can and should do for commercialization of biopharmaceuticals. I met Scott as a result of a shared interest in pH measurement and bioreactor modeling and control. Scott supported the exploration of a virtual plant as described in the BioProcess International Journal article "PAT Tools for Accelerated Process Development" March 2008 Supplement Series. Scott also provided a 100-liter bioreactor for testing wireless pH measurement and control. In the bioreactor tests we found that the wireless pH measurement had a resolution or threshold sensitivity of 0.001 pH and ignored spikes from electromagnetic interference (EMI) seen in wired measurements. The enhanced PID could achieve setpoint changes with an overshoot of less than a 0.002 pH. The long term accuracy including junction effects was 0.01 pH. These phenomenal results compared to what we are accustomed to in chemical process pH measurement and control were described in the Control May 2009 article "Is Wireless Process Control Ready for Prime Time."

Stan: Scott, how did you get into providing bioreactor systems for research and development?

Scott: I always had a fascination with biotechnology. After our development of the steam sterilizible gel filled pH electrode in the 1980s, I started a dialog with biopharmaceutical companies. I always answered the phone, gave first hand info, and got involved with what the customer needed. We gave seminars on the "Top Ten Practices for DO and pH Measurement." This led us to the idea of a better solution to the process development (PD) lab. The original concept was to start with a DCS scaled down to bench top applications with sophisticated control capability to deal with complex processes.

Greg: What was the biggest surprise?

Scott: We didn't anticipate the networking implications in Process Development. The PD Lab vessels acquire large amounts of data in historians including PID outputs that create the ability to adjust the PID based on batch phases using gain scheduling, steps, and ramps. The DCS system also has advanced control tools, such as adaptive control, model predictive control, and data analytics. We brought the best of automation technology developed over decades in commercial processes to the PD lab. These BioNet lab optimized DCS systems are little powerhouses of data which is what a PD lab is all about.

Stan: Why is the demand for data so great?

Scott: Besides the design of experiments (DOE) to determine optimums and operating condition limits in the definition of the process for the Food and Drug Administration (FDA), statistical analysis requires a large number of vessel runs. Each vessel run takes 2 weeks and with project time always being critical, many more bioreactor runs are required to run in parallel for the same experiment. In our most recent installation we had a total of 64 one liter bioreactors at two different sites for a particularly visionary and astute biopharmaceutical PD lab. The data was networked revealing essentially the same results independent of site and operator. Furthermore, the automation of the labs at both facilities enabled twice as many runs to be completed with half as many operators. The data obtained had minimal variance, was reproducible, and was explainable within the design space. The data variance was actually cut in half. With all biopharmaceutical budgets being squeezed this 4x improvement in productivity is drawing a lot of attention.

Stan: The 1 liter bench top bioreactor requires incredibly precise dosing and extremely small samples for at-line analyzers. Why so small?

Lab may do 50 to 200 small vessel runs in their development of scale-up data. A 500-liter pilot plant scale run costs $100K to $200K whereas the 1-liter run costs $1K to $K. However, precise control is not lost at this small vessel size. The lab system uses mass flow controllers (MFC) instead of solenoid valves for dosing. The MFC has a thermal flow meter, PID, and an internal flow element that gets a remote setpoint from either pH or DO controllers in the DCS. The cascade loop provides tight control and a flow measurement that provides considerable knowledge for data analytics and first principle modeling and diagnostics.

Greg: The job of the process controller is to transfer variability from the process variable (PV) to the manipulated variables (MV), inputs to the process, which in this case is air, oxygen, and carbon dioxide flows. If the process controller does a great job, changes in the process are not seen in the PV but are seen in the MV. When a process controller output goes directly to a final control element (no cascade control and no flow measurement), the relationship between the MV and the PV has too much nonlinearity and uncertainty principally due to pressure changes to be used for statistical or first principle modeling. The story is in the controller output but the picture is much clearer if there is a flow measurement.

Scott: The data historian is such a great tool. The trend in PID outputs tells us so much more, which has led to significant improvements in product quality. Each large scale production run takes 6 weeks and 30 people in a manufacturing plant, not counting delays for trying to solve and document continuing problems. Consequently, the time besides the cost benefits of preventing rejected batches are huge.

Stan: What do contract manufacturing organizations (CMO) think of the results of these lab optimized automation systems?

Scott: The relationship between the biopharmaceutical company and the CMO, which can be contentious at times, has greatly improved when these systems are used. The BioNet/DeltaV equipped PD Lab can develop large amounts of well organized and analyzed data. The basic control strategy can be worked out in the PD lab well ahead of the first meeting with the CMO. Repetitive PD bench top runs give their data statistical significance and this greatly reduces the risks involved with the scale up process. There is much less risk of a "surprise" at the 6000 L scale. Everyone wins if risk is reduced. The technology transfer to the CMO is accomplished with less cost and much improved time-to–project completion. This makes both the Biopharma client and their CMO much happier about the project.

Greg: Besides knowing the flows of nutrients and gases, the real time measurement of cells, key components in the culture, and product concentration provide the knowledge that could be used not only for online optimization but for dynamic modeling. The resulting virtual plant can be run faster than 500 times real time to develop and test CMO automation systems.

Scott: An essential part of our lab optimized system is an integrated, multiplexed, at-line Nova BioProfile Flex analyzer with an automated sample system that can measure 14 parameters in a sample including cell count, size, viability, and osmolality, substrate components glucose and glutamine, an inhibiting components such as ammonia and lactate. Most recently, the technology has been developed to automatically fill and freeze 5 cc vials for measurement of product concentration and quality. This product analysis is done offline with liquid chromatography due to analysis complexity.. The sample volume taken per day is about the same as the substrate and nutrients added so there is no appreciable loss of bioreactor culture, which is critical for the 1 liter size.

Stan: What do the scientists think of these sophisticated control and automation systems?

Scott: They love them. The automation frees them up to do real science rather than mundane experimental procedures. Plus, with ease of the remote connectivity, they can monitor experiments and tweak something using their home laptop or even their iPhone so they don't have to come into the lab during the night or on weekends to tweak a parameter to get the most out of a run. There is less stress. The scientists can also be more innovative with the many automation features that are pre-configured and hence selectable and adjustable via graphics optimized for their PD Lab.

Greg: I expect the benefits in terms of batch repeatability are similar to what we see whenever we automate manual operations in industrial plants due to limitations in human predictability and monitoring and adjustment scope and timing. The automation in the PD lab may translate to even greater benefits due to the complexity of the decisions and interactions and the incredibly long time horizon before the consequences are seen. We have a two way street and a synergy. The best of automation technologies and techniques from industrial production can be moved into the PD lab that can be translated into more advanced control systems to be transferred to the CMO. Furthermore the capability of a CMO can be evaluated for CMO selection and improvement by the use of a virtual plant developed from the prolific PD lab data. We bring it home with a list of reasons to use an optimized DCS for your BBQ.

Top Ten Reasons to Use an Optimized DCS for Your BBQ

(10) Automated recipes
(9) Predicted BBQ times
(8) Five-course meal no problem
(7) Don't have to watch cooking shows
(6) Feed-forward control
(5) Process control comes home
(4) Children want to become automation engineers
(3) Spouse finally appreciates your expertise
(2) Griller not grilled
(1) More time to drink beer

About the Author

Greg McMillan | Columnist

Greg K. McMillan captures the wisdom of talented leaders in process control and adds his perspective based on more than 50 years of experience, cartoons by Ted Williams and Top 10 lists.