This work describes the use of multivariate latent variable modeling to enhance fundamental understanding of the common cause variability present in the operation of a pharmaceutical spray dryer; and provide a real time process monitoring and fault detection tool for this process. A latent variable model (LVM) was built and tested using commercially available software in a pilot-scale facility at Bend Research Pharmaceutical Process Development Inc. (BRPPD) in Bend, Oregon.

As part of this study, the pilot-scale spray dryer was modified to closely replicate control structure of the commercial-scale spray dryer since the system showed to also capture and monitor the performance of the controllers installed in the unit.

An LVM process monitoring system based on this model successfully identified seven process faults (faults were artificially introduced into the system) in real time during pilot-scale spray-drying trials at BRPPD. The performance of the system exceeded the expectations since it was not only able to detect those planned faults, but also unplanned events during the test.

A key advantage of the latent variable approach is that the variability that drives the process is easily understood in a fundamental way by interpreting the model parameters in the light of fundamental engineering knowledge (transport phenomena, thermodynamics, etc). The understanding of the common cause variability is greatly enhanced and enables the better understanding of the differences across scales for this unit. In monitoring the process, the faults are not only detected in a “statistical” way, but are also understood in a fundamental way by using the model to track down the driving forces that were involved in detecting such fault (e.g. a change in dryer thermodynamics, an abnormal behavior of the gas momentum across the unit).

The LVM process monitoring system can be used to detect spray drying process upsets in real time, potentially avoiding production of off-specification product and reducing process downtime. The system alerted the operator in much less time than would be possible using conventional univariate monitoring techniques, enabling the operators to address the upset before much product was lost. This system enables the operators to keep the process under tight operation, assuring the continuous quality of the end product.