Transferring Monitoring Models Between Different Scales Through Multivariate Statistical Techniques
- Type: Conference Presentation
- Skill Level:
The challenge of transferring a process monitoring model across different scales is a different problem than the one addressed in previous studies that have focused on transferring operating procedures looking to produce the same grade of product [1, 2, 3, 4].
In the development of a new pharmaceutical product, it is of paramount importance to shorten the launch time of the product, and for new compounds it may be that this is the first time the product is manufactured at pilot scale. At the pilot scale, an extended experimental campaign can be carried out to gain process understanding and disclose potential pitfalls in the process equipment, operating conditions or control configuration. This experimental campaign eventually leads to the definition of a process equipment layout as well as a set of proper (“normal”) process operating conditions that can guarantee the required product quality with acceptable variability. If the experimental work at the pilot scale is extended to produce a data set where common cause variability can be observed, then a multivariate statistical process monitoring model can be designed and used to detect (and possibly diagnose) process faults and malfunctions at the pilot-plant scale in real time, thus gaining further process understanding.
When the product is transferred to the commercial manufacturing scale, it is highly desired to have an effective process monitoring framework available as quickly as possible. Carrying out an extended experimental campaign to produce data (where the common cause variability can be observed) is not a viable option at this scale, and therefore the issue arises on whether the monitoring model developed at the pilot-plant scale can be transferred up to the production scale to monitor the manufacturing process until a sufficient amount of data have been collected to design a full production-scale monitoring model.
This work addresses the above issue. Some approaches will be presented that use multivariate statistical techniques [1, 5, 6] and enable transferring a process monitoring model between different scales. The approaches are adaptive, in that initially they rely heavily on the wealth of information embedded in the pilot-scale data, but the relevance of these data is progressively discounted as long as more production-scale data become available. The proposed approaches are tested and critically evaluated using real-world data coming from a spray-drying process , and prove to be very effective at monitoring normal conditions and quickly detecting process faults at the production scale using data from the pilot scale. The presented methods will accelerate the implementation of continuous quality assurance programs in pharmaceutical manufacturing by transferring a greater portion of the knowledge generated at the pilot scale.
 Garcìa-Muñoz, S., J. F. MacGregor, and T. Kourti (2005). Product transfer between sites using Joint-Y PLS. Chemom. Intell. Lab. Sys., 79, 101-114.
 Jaeckle, C. M., and J. F. MacGregor (2000). product transfer between plants using historical process data. AIChE J., 46, 1989-1997.
 Jaeckle, C. M., and J. F. MacGregor (2000). Industrial application of product design through the inversion of latent variable models. Chemom. Intell. Lab. Sys., 50, 199-210
 Jaeckle, C. M., and J. F. MacGregor (1998). Product design through multivariate statistical analysis of process data. AIChE J., 44, 1105-1118.
 Jackson, J. E. (1991). A user’s guide to principal components. John Wiley & Sons Inc., New York (U.S.A.).
 Geladi, P. and R. Kowalski (1986). Partial least squares regression: a tutorial. Anal. Chim. Acta, 185, 1-17.
 Dobry E.D., D.M. Settell, J.M. Baumann, R.J. Ray, L.J. Graham and R.A. Beyerinck (2009). A model-based methodology for spray-drying process development. J. Pharm. Innov., 4, 133-142.