Milling process is one of the complex unit operations in the pharmaceutical domain. Its complexity arises due to the various breakage mechanisms that happen in a variety of mills. Traditionally population balance modeling (PBM) approach has been used to model the milling process. But certain important decisions such as selection of the mill type based on mill and material properties, scale up factors etc. have been heuristic in nature and are not obtained directly from PBM. Knowledge based tools and data driven models can be used to overcome these drawbacks. But the data driven (DD) model development depends on the knowledge of the expert who is building the model. The know-how of the product and DD model development are largely hidden in the expert's mind and in the development environment's syntax. Thus, this knowledge is not available for sharing across different platforms and for computer interpretation.

Our present work focuses on extracting the product and model development knowledge from the expert and making it computer interpretable. This knowledge includes the domain knowledge (e.g. milling) as well as data-driven modeling (e.g. statistical, neural nets) knowledge. This results in a transparent and standardized model development environment along the lines of the framework to manage mechanistic modeling knowledge proposed by Suresh et al [1]. Our ontology centric framework includes milling ontology and guidelines for mill selection, statistical ontology, guidelines for the statistical methods and a Java engine to integrate these with the Matlab solver. Product development for any pharmaceutical unit operation can be developed using our generic framework. We demonstrate our framework by developing a hybrid neural network model for the milling process.

References

1. Informatics Based Approach for Mathematical Knowledge Modeling in Process Operations, Pradeep Suresh, Chunhua Zhao, Girish Joglekar, Venkat Venkatasubramanian, 60f, AICHE Annual Meeting – 2007