The uniformity of the wet-particle cloudy zone in liquid-sprayed fluidized beds directly governs coating quality, yet the chaotic nature of gas–solid multiphase flow makes its active control extremely challenging. This work first establishes a closed-loop control framework coupling computational fluid dynamics with the discrete element method (CFD-DEM) and deep reinforcement learning (DRL), in which the multiphase flow simulator serves as the environment and a Soft Actor-Critic (SAC) agent acts as the controller to adjust three independent inlet fluidization velocities in real time. To overcome the computational bottleneck of multiphase flow simulation, dream training is introduced for the first time into active control of multiphase flow reactors: an ensemble surrogate dynamics model is constructed from CFD-DEM data, enabling the agent to explore control strategies entirely within the surrogate and improving the training efficiency by approximately four orders of magnitude. Leveraging this capability, multiple reward function designs are systematically investigated, revealing that the agent autonomously discovers a differentiated multi-inlet pulsed fluidization strategy. We also introduce the concept of meso-intelligence, which embeds mesoscale structural awareness into the reward design, and further propose a macro-meso dual layer reward function structure to enhance the spatial uniformity of the cloudy zone. This work demonstrates that replacing multiphase flow simulation with a surrogate model for policy training enables systematic reward exploration and policy convergence at negligible computational cost, offering a new paradigm for active flow control in multiphase reactors.

Multiphase flow dynamics critically governs mass transfer, reaction kinetics, and overall system efficiency in alkaline water electrolyzers (AWEs). To address the inherent flow maldistribution in large-scale cells, this work proposes a gradient-distributed concave-convex plate, featuring elements densely packed at the center and sparsely distributed at the periphery. A pilot-scale transparent electrolyzer (380 mm diameter) is developed to directly observe gas-liquid flow patterns, enabling comparative analysis between the proposed gradient structure and a conventional uniform design. The experiments demonstrate that the novel design significantly improves electrolysis performance, enhancing efficiency by 3.0% at 2 L/min and 4.4% at 3 L/min under a current density of 4000 A/m2. The gradient structure also achieves superior gas holdup control, with a 26% reduction at 5 A and a 15% average reduction across the investigated range compared to conventional plates. As electricity dominates AWE operating costs, this gain offers a low-cost path to reducing hydrogen production costs.

Scale-up effects on the hydrodynamics in bubble columns are particularly critical for reactor design and optimization. However, unlike empty bubble columns, there remains a lack of comprehensive understanding of scale-up effects in the presence of vertical internals. To address the gap, this study investigated the effects of column scale on gas holdup, flow field and turbulence properties in the bubble columns with tube bundle internals at superficial gas velocity 0.12 m/s through CFD simulation. The predictions indicated that, the radial distribution of gas holdup becomes more uniform with scale-up. Moreover, as the column scale increases, the flow pattern in the empty columns consistently exhibits typical gulf-stream pattern. However, for the bubble columns with internals, the flow pattern shifts from the gulf-stream to a dual-circulation pattern with the flow reversal in the center. The effects of column scale on the turbulence properties are also completely different for the empty column and the column with internals. Mechanism analysis demonstrated that the gas holdup distribution governs the large-scale liquid circulation pattern, while turbulence viscosity plays a pivotal role in regulating circulation intensity. We believe that these findings could provide more insight for the design and scale-up of bubble column reactors with internals.

Droplet size and its distribution are critical determinants of polystyrene (PS) properties in suspension polymerization, yet the specific roles of inorganic dispersants and surfactants remain insufficiently understood. In this study, a non-reactive styrene-water system was employed to study the coupling effects between tricalcium phosphate (TCP) and sodium dodecyl benzene sulfonate (SDBS). Interfacial tension measurements, combined with Szyszkowski-Langmuir analysis, provided quantitative insights into surfactant adsorption, while a systematic investigation of the effects of dispersant concentration, surfactant concentration, agitation speed, and oil-to-water ratio on droplet size revealed the roles of these factors in controlling droplet size and emulsion stability. Incorporating the relevant mechanisms of TCP and SDBS into the classical empirical correlation effectively captures their interaction and accurately predicts the Sauter mean diameter. This result enhances our understanding of the complex interactions among inorganic particles, surfactants, and hydrodynamic conditions in suspension polymerization, and provide quantitative guidance for droplet size control and the optimization of industrial polystyrene synthesis.

Achieving uniform liquid film distribution among multiple particles during droplet coating is critical yet challenging due to the nonlinear and stochastic nature of multiphase flows. This study proposes a closed-loop framework that synergizes Volume-of-Fluid (VOF) simulations with deep reinforcement learning (DRL) for intelligent coating control. Within this CFD-DRL framework, a Distributional Soft Actor-Critic (DSAC) agent perceives real-time flow fields via CFD probe arrays and continuously adjusts injection velocity and droplet diameter. The agent autonomously uncovers mesoscale games governing uniform droplet coating, discovering a synergistic-antagonistic coordination between velocity and diameter that stabilizes the Weber number within an optimal range. This significantly enhances inter-particle coating uniformity while maintaining high singleparticle coverage. The algorithm exhibits strong adaptability to geometric constraints, spontaneously switching strategies between dispersed and compact particle arrangements. Crucially, analysis of these mesoscale games reveals a form of emergent mesoscale intelligence: the agent achieves global statistical uniformity through temporal-domain alternating bias, demonstrating that local non-uniformity at the mesoscale is the necessary route to global uniformity. This work establishes a new paradigm for intelligent control in complex multiphase processes.

Bubble-induced turbulence (BIT) plays a crucial role in the computational fluid dynamics (CFD) simulation of bubbly flows, yet its precise effects remain an open question. This study investigates the influence of BIT modeling on the predicted hydrodynamics in two bubble columns of different diameters (14 cm and 44 cm). Our results indicate that while incorporating BIT significantly improves predictions for the larger column, it offers minimal advantage for the smaller one. Specifically, BIT modeling markedly accelerates the flow development in the 44 cm column, whereas in the 14 cm column, the flow already develops rapidly even without BIT, rendering its effect comparatively weak. To elucidate the underlying mechanism, we analyze the radial distribution of the BIT source term Sk. In the large-diameter column, Sk remains relatively flat across the core region before declining sharply near the wall. In contrast, for the small-diameter column, Sk begins to decrease rapidly from the center, suggesting a fundamentally limited influence of BIT. This study highlights the scale-dependent nature of BIT and offers practical insights for model selection in bubble column simulations.

Quantifying surfactant effects in bubble columns remains challenging. The difficulty lies in linking local interfacial phenomena, such as bubble deformation and liquid-film drainage, to macro-scale hydrodynamics. In our previous study (Liu et al., AIChE J. 2025; 71 (10):e18902), the Marrucci number (Ma)-the ratio of resistance induced by surface tension gradients to the driving force during liquid film drainage-was identified as a key parameter for quantifying surfactant effects, and a critical velocity correlation was developed based on bubble coalescence experiments. In this work, the critical velocity model was incorporated into the coalescence kernel within the PBM framework, and the CFD-PBM simulations reproduced the Sauter mean diameter () observed in experiments for various ethanol concentrations. Additional simulations in propanol solutions further demonstrate the model's applicability by capturing the distinct interfacial effects of different surfactants. These findings show the model's potential to reflect surfactant effects in macro-scale fluid dynamics.

Major iron and steel companies have identified the H2-based shaft furnace (SF) as a key decarbonization technology for future development. Its continued advancement aims to achieve full H2 operation and broaden the applicability of ores with different grades. However, published studies report contradictory trends in SF reduction performance with increasing H2 content, leading to significant confusion about the role of H2. The underlying causes remain unclear, and the influence of ore properties on SF performance has received limited investigation. In this work, using our recently developed CFD model for industrial SFs, the influence of ore properties on SF reduction behavior is investigated under various H2 contents. Differences in ore properties focus on thermodynamic equilibrium differences, as documented in two well-known databases, NIST and FactSage. The results show that thermodynamic equilibrium differences can modify the contributions of high- and low-temperature reduction, leading to inconsistent trends in metallization as H2 content increases. The gas flow rate is also a contributing factor, as it alters the H2:CO range in which the thermodynamic disadvantage of H2 reduction hinders reduction. These findings provide insight into how ore properties and the reducing gas flow rate modulate in-furnace H2 reduction and highlight the need for reliable ore characterization to properly assess H2-based SF performance.

As a promising low-carbon ironmaking technology, shaft furnace (SF) direct reduction using pure H2 is attracting significant attention from the steel industry. However, how SF performance evolves during the transition from H2-rich to pure H2 operation remains insufficiently understood, and how metallization under high H2 conditions can be improved remains unclear. A recently developed process model, further developed and validated to include both high- and low-temperature reduction steps, is applied to address these gaps. The comparative analysis highlights the importance of low-temperature reduction in the overall reduction process under high-H2 conditions. It also provides insight into the decline in reduction performance at a higher H2 content. Subsequently, systematic studies of gas-injection strategies are conducted to cover the effects of reducing gas temperature and flow rate, as well as N2 addition methods. The results identify feasible operational routes under pure H2 conditions, significantly increasing metallization while maintaining appreciable H2 utilization.

The shaft furnace is a promising low-carbon ironmaking technology, but its complex internal processes pose challenges for efficient operation and control. Numerical simulation provides an effective method for revealing the internal states, whereas many existing models incorporate simplifications that limit prediction accuracy. This study presents a CFD model for industrial MIDREX shaft furnaces, incorporating major reactions, including CH4-related side reactions. An improved unreacted shrinking core model (USCM) is introduced for reduction based on the thermodynamic equilibrium diagram. It allows for the decoupling of reaction and diffusion layers, reducing the reaction layers and enabling variable diffusion structures. Consequently, the reaction model provides a more physically realistic and thermodynamically consistent representation of the multi-step reduction process, is valid for H-2-CO mixtures, and increases computational efficiency by 65-75 %. The model is validated using four sets of laboratory and industrial data, covering inner states and overall performance parameters. It also successfully captures the influence of CH4, an inevitable component arising from reforming. A high CH4 content causes substantial heat loss near the gas inlet and reduces metallization, primarily due to the combined effects of endothermic reduction and side reactions. Based on simulation results, an optimal CH4 level is identified to balance metallization, gas utilization, and energy efficiency. The model can serve as a valuable tool for analyzing shaft furnace performance and has the potential to support the evaluation of advanced decarbonization strategies such as high H-2 content operation.

Abstract The spreading and evaporation of liquid droplets on hot particle surfaces are critical to liquid‐sprayed fluidized bed design. While coarse‐grid approaches (CFD‐DEM, Euler–Euler) are commonly used in reactor‐scale simulations, they lack resolution of microscale interfacial phenomena at the single particle level. This work performs high‐fidelity Volume‐of‐Fluid (VOF) simulations coupled with an interface evaporation model to investigate droplet impact and evaporation on heated spherical particles over wall temperature of 250°C–400°C and Weber number of 8–84. Results show that total evaporation time is insensitive to particle temperature, while droplet morphology and evaporated mass strongly depend on wall temperature and Weber number. Moreover, an effective evaporation area coefficient derived from VOF simulations is introduced as a correction factor for macroscopic models, enabling coarse‐grid simulations to retain essential microscale physics without explicitly resolving interfaces. It is expected to improve predictive accuracy for simulation of liquid‐sprayed fluidized bed with coarse‐grid approaches.

Rotating disk contactors (RDCs) have been commonly used for extraction due to their simple design and high throughout. However, accurate prediction of phase holdup remains difficult in view of the complex multiphase flow characteristics in RDCs. Drag models plays a crucial role as they directly determine both global and local predictions of phase holdup in RDCs. In this study, we present a new drag model developed based on the EnergyMinimization Multi-Scale (EMMS) approach. The proposed EMMS-based drag model represents a significant advancement as it eliminates the need for adjustable parameters, thereby enhancing the accuracy of phase holdup prediction compared to existing models that often rely on specified droplet diameter. Furthermore, phase inversion phenomenon was analyzed based on the predicted flow pattern, indicating that up to 50% of oil can transition into a continuous phase under high rotational speed. The new model offers a more comprehensive understanding of the complex multiphase flow behaviors in RDCs, and valuable guidance for the design and optimization of liquid-liquid extraction processes.

In liquid‐injected fluidized bed processes, such as ethylene polymerization, it is crucial to quantitatively identify the cloudy zone consisting of gas bubbles, droplets, and wet particles. While various experimental methods exist for measuring relevant parameters, a comprehensive understanding of the characteristics of the cloudy zone remains challenging. This study introduces a particle‐based method to identify the cloudy zone using CFD–DEM simulations, focusing on heat and mass transfer during liquid evaporation and particle‐droplet collisions. Image analysis of simulation results reveals a horseshoe‐shaped cloudy zone and elucidates the life cycle of wet particles, transitioning from dry to wet and back to dry. The investigation identifies five distinct stages characterized by changes in evaporation rate, temperature, and liquid mass during the core‐annular flow of wet particles. Importantly, the temperature gap between dry and wet particles diminishes as thermal energy transfer during droplet‐particle collisions weakens at higher liquid injection rates.

Liquid-liquid dispersion is often performed in stirred tanks, which are valued for their ease of operation, high droplet generation rate and effective droplet dispersion. Many relevant simulations use the Eulerian-Eulerian method, combining population balance equations with statistical models to forecast droplet breakage. Conversely, the Eulerian-Lagrangian (E-L) method provides precise tracking of individual droplets, which is crucial for simulating dispersion processes. However, E-L simulation faces challenges in integrating droplet breakage effectively. To address this issue, our research introduces a probabilistic approach for droplet breakages. It assumes that a longer time increases the likelihood of breakup; a droplet breaks if the calculated probability exceeds a random value from 0 to 1. Consequently, the simulated breakage frequency becomes independent of the Lagrangian time step. The Sauter mean diameter and droplet size distribution can be accurately predicted by this probabilistic approach. By closely monitoring droplet motion, we reveal the complexity of droplet trajectories and the detailed patterns of circulation in stirred tanks. These insights contribute to a deeper understanding of liquid-liquid dispersion dynamics. (c) 2025 The Chemical Industry and Engineering Society of China, and Chemical Industry Press Co., Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Electrolysis for hydrogen production is a key driver for the future of green energy. The pressure-filtered alkaline electrolyzer is widely used in industrial hydrogen production. A 3D Euler-Euler CFD model, coupled with electric field and electro-chemical model, was established to simulate the two-phase flow and the performance of a pilotscale mastoid plate electrolyzer. The fully-coupled model predicts higher overall gas holdup than the model without considering multi-physics. This discrepancy arises from the non-uniform distribution of current density and gas production rate. Moreover, the gas distribution pattern within the electrolyzer changes from a layered distribution to a tree-like distribution due to the increase in liquid flowrate, which, to our knowledge, has not been reported previously. Additionally, the effects of operating voltage and liquid flowrate on the components of overpotential were analyzed. The findings are beneficial for process intensification and optimization for industrial electrolyzers.

This study systematically investigates the influence of liquid level height (Vt/VU) on hydrodynamic performance and mixing efficiency in loop bioreactors, a promising technology for energy-efficient single-cell protein production. Through controlled tracer experiments across varied liquid levels (Vt/VU=1/8 to 3) and stirring speeds (1200-1800 rpm), we demonstrate that reduced liquid levels (Vt/VU <= 1/4) induce gas entrainment at high speeds (>= 1600 rpm), increasing flow resistance and reducing mean fluid velocity. Conversely, elevated liquid levels (Vt/VU >= 1) enhance diffusion-driven mixing, shortening mixing time by 2/3 at Vt/VU=3 compared to 1/8. A critical transition from convection-dominated to diffusion-dominated regimes is identified at Vt/VU=1/2, with mixing efficiency becoming less sensitive to stirring speed at Vt/VU >= 2. These findings provide feasible guidelines for optimizing loop scalability and advancing reactor optimization strategies for continuous bioprocessing.

The determination of the critical collision velocity for bubble coalescence plays a crucial role in predicting bubble size distribution within bubble column reactors through population balance modeling. In this study, experimental measurements in surfactant–laden systems demonstrate that even trace amounts of surfactant significantly reduce the critical velocity. We propose the Marrucci number—a dimensionless parameter derived from the Marrucci film drainage model—as a fundamental metric for quantifying surfactant effects. This parameter is formally defined as the ratio between the film drainage resistance caused by the Marangoni effect and the driving force resulting from capillary pressure. Furthermore, we developed a novel correlation for critical velocity by establishing a quantitative relationship between the critical Weber number and the Marrucci number. This correlation successfully predicts critical velocities in ethanol solutions and shows potential applicability to other systems, such as propanol solutions.

Integrating Bayesian Optimization with Volume of Fluid (VOF) simulations, this work aims to optimize the operational conditions and geometric parameters of T-junction microchannels for target droplet sizes. Bayesian Optimization utilizes Gaussian Process (GP) as its core model and employs an adaptive search strategy to efficiently explore and identify optimal combinations of operational parameters within a limited parameter space, thereby enabling rapid optimization of the required parameters to achieve the target droplet size. Traditional methods typically rely on manually selecting a series of operational parameters and conducting multiple simulations to gradually approach the target droplet size. This process is time-consuming and prone to getting trapped in local optima. In contrast, Bayesian Optimization adaptively adjusts its search strategy, significantly reducing computational costs and effectively exploring global optima, thus greatly improving optimization efficiency. Additionally, the study investigates the impact of rectangular rib structures within the T-junction microchannel on droplet generation, revealing how the channel geometry influences droplet formation and size. After determining the target droplet size, we further applied Bayesian Optimization to refine the rib geometry. The integration of Bayesian Optimization with computational fluid dynamics (CFD) offers a promising tool and provides new insights into the optimal design of microfluidic devices.

The anode porous transport layer (PTL) plays a crucial role in improving the performance of proton exchange membrane (PEM) electrolyzer. However, oxygen accumulation may cover the interface between the anode catalytic layer and PTL, decreasing mass transfer and electrolysis efficiency. In this study, we employed a comprehensive simulation approach, combining lattice Boltzmann simulation with the Immersed Boundary Method (IBM) and Phase Field Model (PFM), to investigate oxygen removal from the anode PTL. The simulation can accurately capture bubble morphology and gas saturation, demonstrating the four stages of gas removal: invasion, exhaust, breakup and retraction. Notably, the invasion stage contributes to approximately 3/4 of the total gas removal time, whereas the invasion time is significantly influenced by the critical throat in PTL. Removing the critical throat may lead to a 20% reduction in the invasion time. To determine the critical throat, we further simulate the breakthrough of a single pore by a bubble. At lower current densities (1.4 A/cm2), reducing interfacial tension facilitates bubble penetration through smaller pores but weakens the longitudinal permeability, leading to a longer invasion stage. However, at higher current densities (14 A/cm2), the breakthrough is predominantly governed by the critical throat. This suggests that optimizing the PTL structure is more effective in mitigating bubble coverage than adjusting interfacial tension under high current density conditions. These findings may provide some guidelines for enhancing the performance of PEM electrolyzer under high current density conditions.