Application of multivariate latent variable modeling to pilot-scale spray drying monitoring and fault detection: Monitoring with fundamental knowledge

This work describes the use of multivariate latent variable modeling (LVM) to enhance fundamental understanding of the operational space, the scale differences and the common-cause variability present in the operation of a pharmaceutical spray-dryer. LVM provided a real-time process monitoring and fault detection tool for continuous quality assurance. A latent variable model was built and tested using commercially available software in a pilot-scale facility at Bend Research Pharmaceutical Process Development Inc. (BRPPD) in Bend, OR. The key learning from the exercise at the pilot-scale helped identify and understand the normal variability of the commercial scale equipment.A key advantage of the LVM approach is that the variability that drives the process is easily understood in a fundamental way by interpreting the model parameters in light of fundamental engineering knowledge (e.g., transport phenomena, thermodynamics). The understanding of the common-cause variability 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 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., an abnormal behavior of the gas momentum across the unit).