(646f) Detection of Morphology and Modality Changes during Crystallization Using in Situ Chord Length Monitoring and Inference System Model.

A new framework has been recently introduced to correlate in situ chord length distribution (CLD) measurements and offline particle size distribution (PSD) data (Irizarry et al, 2017) and applied to different particle morphologies (Schoell et al, 2019). In this work we are expanding this modelling scheme to consider multimodal distribution measurements (Irizarry et al, 2020). This work also applies the proposed approach to predict if the particle morphology or distribution of different APIs can be predicted from a CLD signal. To accomplish this a data-driven inference system and a more robust CLD2PSD model was built for the studied pharmaceutical compounds since the differences in optical properties and particle morphology require a unique model for each individual substance. The inference system is trained on PSDs annotated by modality and morphology, and the individual models have been combined with online solute concentration data to allow for the estimation of crystal growth, secondary nucleation and agglomeration kinetics, depending on the supersaturation level encountered during the process. The validity of this approach is studied in terms of classification and prediction error rates on test sets. The main advantage of the proposed approach is that all data is generated online and in situ, thus avoiding large sampling errors. Also, it can operate in areas of high concentrations and on small particle sizes where PVM pictures are more difficult to resolve.

References

• Irizarry, R. et al., 2017. Data-driven model and model paradigm to predict 1D and 2D particle size distribution from measured chord-length distribution. Chemical Engineering Science, 164, pp. 202-218.
• Schoell, J. et al., 2019. Determining particle‐size distributions from chord length measurements for different particle morphologies. AIChE Journal, 65, e16560
• Irizarry, R. et al., 2020. CLD-to-PSD model to predict bimodal distributions and changes in modality and particle morphology. Submitted to Chemical Engineering Science.