(674f) Towards Continuous Manufacturing By Enhancing Real-Time Slug-Flow Crystallization Analysis Using Noninvasive Inline Imaging and Computer Vision

The evolution of the pharmaceutical and fine chemical industries towards continuous manufacturing signifies a transformative shift from traditional batch processing, driven by the promise of improved efficiency, consistency, and product quality [1,2]. Within this paradigm, Continuous Slug-Flow Tubular (CSFT) crystallizers emerge as pivotal components, offering distinct advantages in the uniformity of product characteristics. However, the adoption of CSFT crystallizers poses significant challenges, primarily in the real-time monitoring of the crystallization process, which is essential for maintaining the desired product quality and process efficiency. Traditional monitoring approaches, optimized for batch processes, are ill-suited for the dynamic and intricate nature of continuous operations, often either disrupting flow or failing to capture essential real-time data comprehensively.

Addressing this critical gap, our study pioneers a novel approach for real-time monitoring in CSFT crystallizers through the integration of advanced computer vision techniques within noninvasive inline imaging systems. This methodological innovation enables the automated, real-time analysis of the crystallization process, facilitating the acquisition of crucial quality variables such as dynamic bulk crystal density and slug volumes and lengths without intruding upon the process flow. By leveraging a combination of single, binocular, and trinocular stereo-vision imaging systems, our system not only extends the capability of current monitoring technologies, but estimate critical quality variables, offering a robust solution to the industry’s pressing need for accurate, real-time process data.

Central to our methodology is the application of state-of-the-art machine learning models, including ResNet18 for image detection and Mask R-CNN [3] for the segmentation of solution slugs and bulk crystal regions. Demonstrating exceptional performance, these models achieve an average precision score of 96.4%, indicating a high level of accuracy in real-time data estimation. This precision is not only indicative of the models’ efficacy in identifying and segmenting crucial image components but also their reliability in accurately estimating quality variables that are pivotal for the control and optimization of the crystallization process.

The implications of our work extend beyond the technical advancement of monitoring technologies. By enabling the real-time estimation of key quality variables, our study facilitates a deeper understanding of the hydrodynamics within CSFT crystallizers. This understanding, in turn, empowers process engineers to engineer narrow particle size distributions—a critical factor in ensuring product quality and consistency. The capability to infer hydrodynamic conditions from the estimated quality variables offers a direct pathway to optimizing the crystallization process, enabling precise control over the formation and growth of crystals within the CSFT crystallizers.

In conclusion, our work represents a significant leap forward in the monitoring and control of continuous manufacturing processes, particularly within the context of CSFT crystallization. By integrating advanced computer vision techniques into noninvasive inline imaging systems, we provide a novel, effective solution for real-time monitoring that not only enhances process understanding and control but also significantly contributes to the continuous manufacturing paradigm. The implications of our research are profound, paving the way for further innovation in process monitoring and optimization, and ultimately supporting the pharmaceutical and fine chemical industries’ transition towards more efficient, reliable, and quality-focused manufacturing processes.

References

[1] A. Majumder, Z.K. Nagy, X.W. Ni, Recent advances in continuous crystallization, Chemical Engineering Research and Design 186 (2022). https://doi.org/10.1016/j.cherd.2022.08.028.

[2] M. Jiang, R.D. Braatz, Designs of continuous-flow pharmaceutical crystallizers: Developments and practice, CrystEngComm 21 (2019). https://doi.org/10.1039/c8ce00042e.

[3] K. He, G. Gkioxari, P. Dollár, R. Girshick, Mask R-CNN, IEEE Trans Pattern Anal Mach Intell 42 (2020) 386–397. https://doi.org/10.1109/TPAMI.2018.2844175.