(89e) Advanced Characterization of Dynamically Structured Fluidized Beds Via Machine Learning Based Image Segmentation and Thermal Imaging

Oscillating the gas flow in bubbling gas-solid fluidized beds presents a compelling approach to structure these complex systems, drawing inspiration from self-organized patterns observed in nature under periodic perturbations, e.g., sand patterns on beaches and dunes [1][2]. This approach leads to the formation of dynamically structured fluidized beds, which exhibit self-organized spatial and temporal patterns in the gas–solid flow. The dynamic structuring of flow enables better control over bubbling behavior, solid mixing, and transport phenomena, offering promising solutions for intensifying fluidized bed operations [2]. However, the inherently dynamic and multi-scale nature of these systems introduces significant challenges in experimental characterization, especially under varying illumination conditions, particle properties, and complex bubble morphologies.

In this contribution, we present investigations of structured oscillating gas-solid fluidized beds, combining advanced image analysis with new insights into heat transfer mechanisms. We introduce a machine learning (ML)-assisted image segmentation methodology for the robust, automatic identification of bubbles in quasi-2D fluidized beds. The ML model, trained with minimal data and achieving an accuracy of 98.75%, is coupled with an in-house Lagrangian bubble tracking algorithm to capture the evolution of individual bubbles, including their position, velocity, and morphological features over time. This approach is validated across a wide range of conditions, including different particle sizes and oscillation parameters, and demonstrates excellent reliability in tracking key hydrodynamic events, such as bubble coalescence and splitting [3].

By applying this method to oscillating beds, where bubbles nucleate, propagate, and rupture in a highly repeatable cycle, we uncover new characteristic features of the dynamic bubbling process. These include the dependence of bubble shape, aspect ratio, and velocity distributions on changing oscillation frequency and particle size, as well as their relevance to the stability of the structured pattern. The segmentation approach also proves valuable in standardizing image analysis and addressing the long-standing challenge of reproducibility in hydrodynamic studies, offering potential applicability across a broad spectrum of multiphase systems.

We also report new findings on the heat transfer performance of oscillating quasi-2D beds. Using infrared thermography (IRT), enhanced with digital detail improvement and convolutional neural network (CNN)-based postprocessing, we investigated the local and global heat transfer in beds operating near minimum fluidization. Compared to constant flow, oscillatory flow enhances mixing and results in up to 19% higher average heat transfer coefficients, despite operating at similar or lower gas throughput. Analysis of the local temperature fields reveals compartmentalized flow structures reminiscent of Rayleigh-Bénard convection cells, directly influenced by the presence and periodicity of the bubble patterns. While oscillatory flow can generate larger bubbles at the same average gas flowrate, it promotes finer, more uniform mixing and effective heat transfer without requiring increased gas flowrate [4].

These insights not only advance the fundamental understanding of dynamic gas-solid systems but also lay the groundwork for future scale-up and application of nature-inspired, dynamically structured fluidized beds in drying, coating, and reaction engineering.

References

[1]. Wu, K.; de Martín, L.; Coppens, M.-O. Pattern formation in pulsed gas-solid fluidized beds–the role of granular solid mechanics. Chem. Eng. J 2017, 329, 4–14.
[2]. Francia, V.; Wu, K.; Coppens, M.-O. Dynamically structured fluidization: oscillating the gas flow and other opportunities to intensify gas-solid fluidized bed operation. Chem. Eng. Proc. Proc. Int. 2021, 159, 108143.
[3]. Jiang S.; Wu K.; Francia V.; Ouyang Y.; Coppens M.-O. Machine learning assisted experimental characterization of bubble dynamics in gas-solid fluidized beds. Ind. Eng. Chem. Res. 2024, 63 (19), 8819-8832.
[4]. Jiang S.; Xu L., Wu K.; Francia V.; Coppens M.-O. Heat transfer in pulsed fluidised beds via infrared thermography. Chem. Eng. J. 2025, 161548.