(169c) Combining PAT with Data Visualization and Machine Learning to Understand and Control Batch Cooling Crystallization

Authors: 
Grover, M. A., Georgia Institute of Technology
Griffin, D. J., Georgia Institute of Technology
Kawajiri, Y., Georgia Institute of Technology
Rousseau, R. W., Georgia Institute of Technology
With the rapid development of PAT, online crystallization monitoring tools have become commonplace. Spectroscopic monitoring tools, such as ATR-FTIR, and particle-detecting tools, such as FBRM, are especially popular in that regard [1].

Using these tools, large data sets can be collected, which presumably contain detailed information on crystallization dynamics. But this is where the promise of big data often falls short. Simply collecting raw data does not automatically lead to understanding. Instead, the data most be processed, analyzed, and presented in a coherent way.

In this talk, we will describe how data from ATR-FTIR, FBRM, and temperature measurements can be brought together to visualize batch cooling crystallization as movement through a two-dimensional space [2]. The data visualization, we will show, fosters an intuitive understanding of the crystallization dynamics. It also sets the stage for a low-dimensional mathematical model of crystallization and dissolution dynamics that can be learned directly from data.

Combined, the data visualization and application of machine learning creates a framework for establishing control. In particular, we demonstrateâ??on two experimental case systemsâ??the use of the framework to accurately control the mean size of crystals produced by unseeded batch cooling crystallization [2, 3].

[1] L. L. Simon, H. Pataki, G. Marosi, F. Meemken, K. Hungerbühler, A. Baiker, et al. 2015. Assessment of Recent Process Analytical Technology (PAT) Trends: A Multiauthor Review. Organic Process Research & Development 19, 3-62.

[2] Griffin, D. J., Grover, M. A., Kawajiri, Y. and Rousseau, R. W. 2015. Mass-Count Plots for Crystal Size Control. Chemical Engineering Science 137, 338-351.

[3] Griffin, D. J., Grover, M. A., Kawajiri, Y. and Rousseau, R. W. 2016. Data-Driven Modeling and Dynamic Programming Applied to Batch Cooling Crystallization. Industrial & Engineering Chemistry Research 55, 1361-1372.