(656d) A Semi-Empirical CLD-to-Psd Model for Crystallization Process Monitoring

Controlling the particle size distribution of the final API is often one of the main objectives of a pharmaceutical crystallization process since API particle properties impact the subsequent formulation process and the ability to produce a drug product with the desired quality attributes. Thus, API crystallization process monitoring and control have become more and more important as part of the increasing QbD efforts in the pharmaceutical industry.

 

Among the possibilities to characterize particle populations in situ and in real-time, the Focused Beam Reflectance Measurement (FBRM) has become broadly accepted mainly due to its ease of process implementation as well as due to its performance robustness. However, one major limitation of the FBRM technique is that its measurement signal, the Chord Length Distribution (CLD), cannot be easily transferred into the particle size distribution (PSD) despite the application of several different resolution approaches over the last 15 years. Thus, the CLD data is often used qualitatively rather than quantitatively and generally provides limited value for advanced crystallization process modeling purposes.

 

In this work a semi-empirical method is introduced that translates CLD to PSD data based on a training set of several particle sieve cuts at increasing particle concentrations. Unlike matrix-based methods, this approach is not limited to compact particle morphologies but holds as long as changes to the particles shape distribution do not exceed certain ranges. This optimization-based method consists of two steps. First, the â??spectral decompositionâ? of the experimental CLD is generated which yields an idealized particle size distribution. In the subsequent optimization step, fuzzy-trained sets are used to correlate the idealized PSD with the actual experimentally measured PSD. This work highlights the application of this model to several relevant compounds and discusses the possibility of using in situ FBRM data for crystallization process modeling.