(14d) Multivariate Latent Variable Modeling Across the Lifecycle of a Product

Muteki, K., Pfizer Worldwide R&D

This talk focuses on the use of multivariate methods (such as PCA ,PLS  are other advanced derivations of these) across the lifecycle of a new product.  For the cases presented here, the lifecycle of the new product begins at the design phase with the selection of raw materials. This selection can be simultaneously done with the design of the process train, along with its processing conditions. Another component of the design phase is the development of analytical techniques to detect the concentration of a component, or to characterize the end product. These methods (ranging from the spectroscopy based to more advanced digital-image based or multi-signal based techniques) are critical in the development of monitoring, real time release and control schemes. As the product evolves, multivariate tools are applied in the transfer of the product from laboratory and pilot scale to commercial manufacturing to ensure the quality and properties of the end product are unchanged from what was produced at smaller scales. Once the product enters full scale manufacturing, multivariate latent variable methods are applied in monitoring of the state of the process and in control algorithms that take advantage of the degrees of freedom in the process to compensate for disturbances entering the system.