Machine-Learning-Based Reduced Order Modeling for Operational Analysis of Industrial Glass Melting Furnaces Using CFD Solutions

Sensor and CFD data fusion for airflow field estimation

A rapid interpolation method of finding vascular CFD solutions with spectral collocation methods

Modeling of evaporation processes in glass melting furnaces

Spoilage of agrifood produce is a major issue in the industry. Cooling is an effective technique for extending the shelf life of fresh agrifood produce to minimize spoilage. Due to the practical inability of directly solving the wide spatial and temporal scales in large industrial agrifood cooling systems, the porous medium approach is mostly used. However, improvements of current porous medium models and modeling across much wider scales are needed to better understand the multiscale cooling process and system problems. Recently, as a result of increased computational capacity, multiscale computational fluid dynamics (CFD) modeling approaches have been developed to tackle some of these challenges. The associated problems and applications of CFD in the design and process optimization of cooling processes and systems at different scales are considered. CFD solution and scale bridging techniques relevant for handling multiscale cooling processes and systems problems are discussed. Innovative applications of various CFD modeling techniques at different scales in cooling processes and systems are reviewed. CFD modeling techniques can be used to handle multiscale cooling process and system problems. Lattice Boltzmann method (LBM) is a potentially viable discrete modeling technique for complimentary usages alongside current continuum techniques in future multiscale CFD modeling. The multiscale CFD modeling paradigm can overcome the computational resource limitations associated with the direct modeling approach and enhance model extension across wider spatial and temporal scales. Information from multiscale CFD could be used to improve the accuracy of current porous medium models, and thus the design of more efficient cooling systems.

In a wide range of applied problems involving fluid flows, Computational Fluid Dynamics (CFD) provides detailed quantitative information on the flow field, at variable level of fidelity and computational cost. However, CFD alone cannot predict high-level functional properties that are not easily obtained from the equations of fluid motion. In this work, we present a data-driven framework to extract these additional information, such as medical diagnostic output, from CFD solutions. This is a challenging task because of the huge data dimensionality of CFD, and the limited training data that can be typically gathered due to the large computational cost of CFD. By pursuing a traditional Machine Learning (ML) pipeline of pre-processing, feature extraction, and model training, we demonstrate that informative features can be extracted from CFD data. Two experiments, pertaining to different application domains, support our claim that the convective properties implicit into a CFD solution can be leveraged to retrieve functional information that does not admit an analytical definition. Despite the preliminary nature of our study and the relative simplicity of both the geometrical and CFD models, for the first time we demonstrate that the combination of ML and CFD can diagnose a complex system in terms of high-level functional properties.

In a wide range of applied problems involving fluid flows, Computational Fluid Dynamics (CFD) provides detailed quantitative information on the flow field, at variable level of fidelity and computational cost. However, CFD alone cannot predict high-level functional properties that are not easily obtained from the equations of fluid motion. In this work, we present a data-driven framework to extract these additional information, such as medical diagnostic output, from CFD solutions. This is a challenging task because of the huge data dimensionality of CFD, and the limited training data that can be typically gathered due to the large computational cost of CFD. By pursuing a traditional Machine Learning (ML) pipeline of pre-processing, feature extraction, and model training, we demonstrate that informative features can be extracted from CFD data. Two experiments, pertaining to different application domains, support our claim that the convective properties implicit into a CFD solution can be leveraged to retrieve functional information that does not admit an analytical definition. Despite the preliminary nature of our study and the relative simplicity of both the geometrical and CFD models, for the first time we demonstrate that the combination of ML and CFD can diagnose a complex system in terms of high-level functional properties.

<div class="section abstract"><div class="htmlview paragraph">In the realm of electric vehicles (EVs), effective battery thermal management is critical to avert thermal runaway, overheating, and extend the operational lifespan of batteries. The process of designing thermal management systems can be substantially expedited through the utilization of modeling and simulation techniques. However, the high-fidelity 3D computational fluid dynamics (CFD) simulations often demand significant computational resources to provide comprehensive results under varying conditions. In this paper, we develop a reduced order model (ROM) to capture the battery thermal dynamics employing a sub-space method. To construct this ROM, we use high-fidelity CFD simulations to generate step responses of battery temperature with respect to the heat generation and cooling power. These step responses are subsequently used as training data for the ROM. To minimize computational expenses while preserving accuracy, we determine the minimal dimensionality of the ROM through the analysis of the singular values of the oblique projection matrix. To assess the accuracy and reliability of the developed ROM, a comprehensive comparison is conducted between ROM results and both CFD solutions and experimental data, specifically in a quick charge scenario. The ROM exhibits good agreement with both CFD and experimental results. Furthermore, a novel predictive control strategy is developed to enhance battery thermal management by leveraging the ROM-derived predictive information for real-time adjustments to the cooling setpoint. The predictive control approach leads to a reduction in total charging time, achieving an improvement of up to 16.2% compared to a baseline case with a constant cooling setpoint. Furthermore, the developed predictive control strategy outperforms traditional feedback control systems that rely solely on current state information.</div></div>

Engineering simulations of coal gasifiers are typically performed using computational fluid dynamics (CFD) software, where a 3-D representation of the gasifier equipment is used to model the fluid flow in the gasifier and source terms from the coal gasification process are captured using discrete-phase model source terms. Simulations using this approach can be very time consuming, making it difficult to imbed such models into overall system simulations for plant design and optimization. For such system-level designs, process flowsheet software is typically used, such as Aspen Plus® [1], where each component where each component is modeled using a reduced-order model. For advanced power-generation systems, such as integrated gasifier/gas-turbine combined-cycle systems (IGCC), the critical components determining overall process efficiency and emissions are usually the gasifier and combustor. Providing more accurate and more computationally efficient reduced-order models for these components, then, enables much more effective plant-level design optimization and design for control. Based on the CHEMKIN-PRO and ENERGICO software, we have developed an automated methodology for generating an advanced form of reduced-order model for gasifiers and combustors. The reducedorder model offers representation of key unit operations in flowsheet simulations, while allowing simulation that is fast enough to be used in iterative flowsheet calculations.more » Using high-fidelity fluiddynamics models as input, Reaction Design’s ENERGICO® [2] software can automatically extract equivalent reactor networks (ERNs) from a CFD solution. For the advanced reduced-order concept, we introduce into the ERN a much more detailed kinetics model than can be included practically in the CFD simulation. The state-of-the-art chemistry solver technology within CHEMKIN-PRO allows that to be accomplished while still maintaining a very fast model turn-around time. In this way, the ERN becomes the basis for high-fidelity kinetics simulation, while maintaining the spatial information derived from the geometrically faithful CFD model. The reduced-order models are generated in such a way that they can be easily imported into a process flowsheet simulator, using the CAPE-OPEN architecture for unit operations. The ENERGICO/CHEMKIN-PRO software produces an ERN-definition file that is read by a dynamically linked library (DLL) that can be easily linked to any CAPE-OPEN compliant software. The plug-in unitoperation module has been successfully demonstrated for complex ERNs of coal gasifiers, using both Aspen Plus and COFE process flowsheet simulators through this published CAPE-OPEN interface.« less

Automatic extraction of reduced-order models from CFD simulations for building energy modelling

The Complex Cepstrum is shown to remove spurious reflections from artificial boundaries in computational fluid dynamic (CFD) solutions. First, the Complex Cepstrum theory is presented. A model time sequence consisting of a direct signal and reflections is analyzed theoretically with the Complex Cepstrum, and it is shown that the direct signal uncontaminated by reflections may be recovered in the time domain. Next, the Complex Cepstrum is applied to one- and three-dimensional CFD solutions, and spurious reflections from the boundary conditions are removed. By eliminating spurious reflections introduced by artificial boundary conditions, the applicability of CFD methods to aeroacoustic problems is greatly enhanced.

It is well-known that distributed parameter computational fluid dynamics (CFD) models provide more accurate results than conventional, lumped-parameter unit operation models used in process simulation. Consequently, the use of CFD models in process/equipment co-simulation offers the potential to optimize overall plant performance with respect to complex thermal and fluid flow phenomena. Because solving CFD models is time-consuming compared to the overall process simulation, we consider the development of fast reduced order models (ROMs) based on CFD results to closely approximate the high-fidelity equipment models in the co-simulation. By considering process equipment items with complicated geometries and detailed thermodynamic property models, this study proposes a strategy to develop ROMs based on principal component analysis (PCA). Taking advantage of commercial process simulation and CFD software (for example, Aspen Plus and FLUENT), we are able to develop systematic CFD-based ROMs for equipment models in an efficient manner. In particular, we show that the validity of the ROM is more robust within well-sampled input domain and the CPU time is significantly reduced. Typically, it takes at most several CPU seconds to evaluate the ROM compared to several CPU hours or more to solve the CFD model. Two case studies, involving two power plant equipment examples, are described and demonstrate the benefits of using our proposed ROM methodology for process simulation and optimization.

The Salt Disposition Integration (SDI) portfolio of projects provides the infrastructure within existing Liquid Waste facilities to support the startup and long term operation of the Salt Waste Processing Facility (SWPF). Within SDI, the Blend and Feed Project will equip existing waste tanks in the Tank Farms to serve as Blend Tanks where 300,000-800,000 gallons of salt solution will be blended in 1.3 million gallon tanks and qualified for use as feedstock for SWPF. Blending requires the miscible salt solutions from potentially multiple source tanks per batch to be well mixed without disturbing settled sludge solids that may be present in a Blend Tank. Disturbing solids may be problematic both from a feed quality perspective as well as from a process safety perspective where hydrogen release from the sludge is a potential flammability concern. To develop the necessary technical basis for the design and operation of blending equipment, Savannah River National Laboratory (SRNL) completed scaled blending and transfer pump tests and computational fluid dynamics (CFD) modeling. A 94 inch diameter pilot-scale blending tank, including tank internals such as the blending pump, transfer pump, removable cooling coils, and center column, were used in this research. The test tank represents a 1/10.85 scaled version of an 85 foot diameter, Type IIIA, nuclear waste tank that may be typical of Blend Tanks used in SDI. Specifically, Tank 50 was selected as the tank to be modeled per the SRR, Project Engineering Manager. SRNL blending tests investigated various fixed position, non-rotating, dual nozzle pump designs, including a blending pump model provided by the blend pump vendor, Curtiss Wright (CW). Primary research goals were to assess blending times and to evaluate incipient sludge disturbance for waste tanks. Incipient sludge disturbance was defined by SRR and SRNL as minor blending of settled sludge from the tank bottom into suspension due to blending pump operation, where the sludge level was shown to remain constant. To experimentally model the sludge layer, a very thin, pourable, sludge simulant was conservatively used for all testing. To experimentally model the liquid, supernate layer above the sludge in waste tanks, two salt solution simulants were used, which provided a bounding range of supernate properties. One solution was water (H{sub 2}O + NaOH), and the other was an inhibited, more viscous salt solution. The research performed and data obtained significantly advances the understanding of fluid mechanics, mixing theory and CFD modeling for nuclear waste tanks by benchmarking CFD results to actual experimental data. This research significantly bridges the gap between previous CFD models and actual field experiences in real waste tanks. A finding of the 2009, DOE, Slurry Retrieval, Pipeline Transport and Plugging, and Mixing Workshop was that CFD models were inadequate to assess blending processes in nuclear waste tanks. One recommendation from that Workshop was that a validation, or bench marking program be performed for CFD modeling versus experiment. This research provided experimental data to validate and correct CFD models as they apply to mixing and blending in nuclear waste tanks. Extensive SDI research was a significant step toward bench marking and applying CFD modeling. This research showed that CFD models not only agreed with experiment, but demonstrated that the large variance in actual experimental data accounts for misunderstood discrepancies between CFD models and experiments. Having documented this finding, SRNL was able to provide correction factors to be used with CFD models to statistically bound full scale CFD results. Through the use of pilot scale tests performed for both types of pumps and available engineering literature, SRNL demonstrated how to effectively apply CFD results to salt batch mixing in full scale waste tanks. In other words, CFD models were in error prior to development of experimental correction factors determined during this research, which provided a technique to use CFD models for salt batch mixing and transfer pump operations. This major scientific advance in mixing technology resulted in multi-million dollar cost savings to SRR. New techniques were developed for both experiment and analysis to complete this research. Supporting this success, research findings are summarized in the Conclusions section of this report, and technical recommendations for design and operation are included in this section of the report.

Development and application of a combined 3D computational fluid dynamics (CFD) and 3D bristle bending model for brush seals is described. The CFD model is created using commercial CFD mesh generation and solver software. A small gap is assumed between all bristles in the CFD model so as to avoid meshing problems at contact points and allow for imperfections in bristle geometry. The mechanical model is based on linear beam bending theory and allows large numbers of bristles to be modelled with arbitrary bristle-to-bristle contact and bristle to backing ring and shaft contact. Aerodynamic forces on the bristles are imported from the CFD solution. Deformed geometries may be exported directly to the mesh generation software, allowing iterative solution of the coupled aerodynamic/mechanical problem. Results from the model are presented for representative brush seal geometry. It is shown that even with initially circumferentially symmetric bristle packing and aerodynamic forces, bristle deflections tend to lead to asymmetric packing. This currently limits application of the fully coupled model, but use of the combined (but not fully coupled) model is nevertheless considered to give a powerful analysis tool. Shaft torque predictions are found to be in good agreement with measurements, and the effect of swirl in the inlet flow has been examined.

This paper presents a parametric reduced-order model (ROM) based on manifold learning (ML) for use in steady transonic aerodynamic applications. The main objective of this work is to derive an efficient ROM that exploits the low-dimensional nonlinear solution manifold to ensure an improved treatment of the nonlinearities involved in varying the inflow conditions to obtain an accurate prediction of shocks. The reduced-order representation of the data is derived using the Isomap ML method, which is applied to a set of sampled computational fluid dynamics (CFD) data. In order to develop a ROM that has the ability to predict approximate CFD solutions at untried parameter combinations, Isomap is coupled with an interpolation method to capture the variations in parameters like the angle of attack or the Mach number. Furthermore, an approximate local inverse mapping from the reduced-order representation to the full CFD solution space is introduced. The proposed ROM, called Isomap+I, is applied to the two-dimensional NACA 64A010 airfoil and to the 3D LANN wing. The results are compared to those obtained by proper orthogonal decomposition plus interpolation (POD+I) and to the full-order CFD model.

The breakdown of the liquid film at the wall in annular gas-liquid flow may lead to the formation of a stable dry patch. For the case of heat transfer surfaces this causes a hot spot, The dry patch is a partial area on the solid surface that is non-wetted due to a local disturbance of the flow and is sustained by surface tension. Dry patch stability is dependent on a balance of body and surface forces. In the present study the interfacial shear force drives the film and the gravity force is negligible. A new computational fluid dynamics (CFD) solution of the flow field in the film around the dry patch has been obtained. The CFD results confirm Murgatroyd's shear force model (1965), although the details are more complex. Furthermore, there is agreement between the CFD solution and the experimental value of the characteristic length scale, L, for the shear force. In addition new experimental data have been taken for adiabatic upward annular air-water and air-ethylene glycol flows at room temperature in a 9.5 mm diameter tube. They provide validation of Murgatroyd's model over a wider range of the film's Reynolds number than previous data.