“…The selection of n i and n k is not a trivial task since these two values determine the size and number of data samples available for the model to be trained with (refer to section Data Reconditioning: Augmentation through Windowing), which naturally has a direct effect on model performance and uncertainty. Considering the differing lengths of the time-series data sets for each simulation case, acknowledging that they have been truncated for each mixing system (τ f = 385 and τ f = 98 for the stirred and static mixer, respectively), the values for n i and n k were fixed to (50,50) and (40,30) for the stirred and static mixer, respectively. These values were subjected to an early sensitivity test, but a full-scale tuning process would be required to discover the optimal configuration for each mixing case study.…”
Section: ■ Methodologymentioning
confidence: 99%
“…Each mixer operates in a completely different flow regime, with the former handling transitional/turbulent flows (Re = ρND r 2 /μ ≈ [9000, 18,000]) and the latter operating under laminar conditions (Re = ρ U r D r /μ = 1.63). For brevity, specifics on the problem formulation of each system will not be described herein, but readers are encouraged to refer to previous publications detailing the geometrical and operational specifications, fluid properties, numerical considerations (e.g., grid refinement), and validation. − The extracted data sets comprise multidimensional time-series data encompassing three key metrics integral to the dispersion performance: interfacial area growth (IA), drop count (ND), and DSD, calculated as the approximate volume of cells resolving a fully detached structure or “drop”. The choice of these parameters capitalizes on the explicit and robust nature of the interface-tracking scheme [level-contour reconstruction method (LCRM)] embedded in the CFD code used, which furnishes a more accurate and well-resolved representation of the intricate interfacial dynamics compared to other traditional schemes (e.g., level-set methods) …”
Section: Methodsmentioning
confidence: 99%
“…Inspired by the work reviewed before, this study seeks to develop an inexpensive time-series model using RNNs by capitalizing on a comprehensive set of high-fidelity three-dimensional CFD simulations, some of which have been exploited in recent works − to unravel the fundamental governing mechanisms underlying extensively utilized mixing systems handling L–L dispersions across a range of industrially relevant scenarios. These simulations have been conducted with a state-of-the-art DNS code, which comprises a hybrid front-tracking/level-set interface-tracking algorithm, embedded along a well-validated multiphase solver for surfactant transport at the interface and in the bulk phase .…”
We demonstrate the application of a recurrent neural network (RNN) to perform multistep and multivariate timeseries performance predictions for stirred and static mixers as exemplars of complex multiphase systems. We employ two network architectures in this study, fitted with either long short-term memory and gated recurrent unit cells, which are trained on highfidelity, three-dimensional, computational fluid dynamics simulations of the mixer performance, in the presence and absence of surfactants, in terms of drop size distributions and interfacial areas as a function of system parameters; these include physicochemical properties, mixer geometry, and operating conditions. Our results demonstrate that while it is possible to train RNNs with a single fully connected layer more efficiently than with an encoder−decoder structure, the latter is shown to be more capable of learning long-term dynamics underlying dispersion metrics. Details of the methodology are presented, which include data preprocessing, RNN model exploration, and methods for model performance visualization; an ensemble-based procedure is also introduced to provide a measure of the model uncertainty. The workflow is designed to be generic and can be deployed to make predictions in other industrial applications with similar time-series data.
“…The selection of n i and n k is not a trivial task since these two values determine the size and number of data samples available for the model to be trained with (refer to section Data Reconditioning: Augmentation through Windowing), which naturally has a direct effect on model performance and uncertainty. Considering the differing lengths of the time-series data sets for each simulation case, acknowledging that they have been truncated for each mixing system (τ f = 385 and τ f = 98 for the stirred and static mixer, respectively), the values for n i and n k were fixed to (50,50) and (40,30) for the stirred and static mixer, respectively. These values were subjected to an early sensitivity test, but a full-scale tuning process would be required to discover the optimal configuration for each mixing case study.…”
Section: ■ Methodologymentioning
confidence: 99%
“…Each mixer operates in a completely different flow regime, with the former handling transitional/turbulent flows (Re = ρND r 2 /μ ≈ [9000, 18,000]) and the latter operating under laminar conditions (Re = ρ U r D r /μ = 1.63). For brevity, specifics on the problem formulation of each system will not be described herein, but readers are encouraged to refer to previous publications detailing the geometrical and operational specifications, fluid properties, numerical considerations (e.g., grid refinement), and validation. − The extracted data sets comprise multidimensional time-series data encompassing three key metrics integral to the dispersion performance: interfacial area growth (IA), drop count (ND), and DSD, calculated as the approximate volume of cells resolving a fully detached structure or “drop”. The choice of these parameters capitalizes on the explicit and robust nature of the interface-tracking scheme [level-contour reconstruction method (LCRM)] embedded in the CFD code used, which furnishes a more accurate and well-resolved representation of the intricate interfacial dynamics compared to other traditional schemes (e.g., level-set methods) …”
Section: Methodsmentioning
confidence: 99%
“…Inspired by the work reviewed before, this study seeks to develop an inexpensive time-series model using RNNs by capitalizing on a comprehensive set of high-fidelity three-dimensional CFD simulations, some of which have been exploited in recent works − to unravel the fundamental governing mechanisms underlying extensively utilized mixing systems handling L–L dispersions across a range of industrially relevant scenarios. These simulations have been conducted with a state-of-the-art DNS code, which comprises a hybrid front-tracking/level-set interface-tracking algorithm, embedded along a well-validated multiphase solver for surfactant transport at the interface and in the bulk phase .…”
We demonstrate the application of a recurrent neural network (RNN) to perform multistep and multivariate timeseries performance predictions for stirred and static mixers as exemplars of complex multiphase systems. We employ two network architectures in this study, fitted with either long short-term memory and gated recurrent unit cells, which are trained on highfidelity, three-dimensional, computational fluid dynamics simulations of the mixer performance, in the presence and absence of surfactants, in terms of drop size distributions and interfacial areas as a function of system parameters; these include physicochemical properties, mixer geometry, and operating conditions. Our results demonstrate that while it is possible to train RNNs with a single fully connected layer more efficiently than with an encoder−decoder structure, the latter is shown to be more capable of learning long-term dynamics underlying dispersion metrics. Details of the methodology are presented, which include data preprocessing, RNN model exploration, and methods for model performance visualization; an ensemble-based procedure is also introduced to provide a measure of the model uncertainty. The workflow is designed to be generic and can be deployed to make predictions in other industrial applications with similar time-series data.
“…This process increases the contact area between the oil droplets and the air, thus enhancing the combustion efficacy of heavy oil and minimizing carbon emissions. 3,4 These traditional methods, such as mechanical agitation emulsification, [5][6][7] ultrasonic emulsification, 8,9 and static mixer emulsification, [10][11][12][13] involve high energy consumption and usually result in an uneven droplet size distribution, which causes agglomeration of the emulsion and limits the practical application of the produced heavy oil emulsions.…”
Continuous ceramic membrane emulsification is a promising and scalable technique to prepare water‐in‐heavy oil (W/O) emulsions. The droplet size of W/O emulsions is comprehensively influenced by phase parameters, operational parameters, and membrane parameters, which collectively impact the forces acting on water droplets. In this work, a droplet size prediction model involving multiple factors is established. The forces are analyzed by considering the influence of transmembrane pressure and the viscosity ratio between the dispersed and continuous phases, which are not well considered by current researchers. Additionally, the effects of pore size, crossflow velocity, temperature, and transmembrane pressure were experimentally verified. The experimental results show a high degree of agreement with the predictions. Also, based on the relaxation time difference in oil and water, magnetic resonance imaging was used for the first time to assess the stability of W/O emulsions which was found to be stable for 4 months.
“…The CFD results revealed that it achieved the same mixing effect as the conventional design with fewer mixing elements and a significantly reduced pressure drop by introducing gaps between the mixing elements. Valdes et al 29 used a direct numerical simulation method to study the liquid−liquid dispersion performance in SMX mixers under different dispersed phase morphologies at the inlet. Their results provide new insights into droplet deformation and breakup in industrial SMX mixers.…”
Mixing is an important unit of operation in the process industry. A static mixer is a device commonly used for liquid mixing in a pipe or channel. This study proposes a new curved-sheet blade-folded (CBF) static mixer. The CBF mixing element is characterized by folded blades arranged alternatively on both sides of a curved sheet. The features of the flow and mixing of two miscible liquids in the new CBF static mixer, e.g., velocity magnitude distribution, vortices, coefficient of variation, and extensional efficiency, are investigated by using computational fluid dynamics. It has shown that in terms of distributive mixing and dispersive mixing, the CBF performs better than the Kenics, Komax, and LPD static mixers. Compared to the other static mixers, the multicomponent system reaches full homogeneity within a short distance in the CBF static mixer. The radial motion of liquid induced by the vortices is attributed to the intensified mixing performance of the CBF static mixer. Furthermore, this study employs proper orthogonal decomposition to reconstruct the velocity and vorticity fields. It has been shown that the fundamental features of the mixing process can be characterized by the velocity and vorticity fields reconstructed by the first seven modes.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.