A deep-learning segmentation model was proposed to predict steady-state thermal fields. The U-Net architecture was used in a reversed order, with the body segments serving as inputs to the model, and the temperature and velocity fields as outputs. Inherent relationships between the segment shapes and fluid domains were extracted via convolution-sampling stages, incorporating skip connections between high- and low-resolution data. An attention gate was used to enhance focus on the body regions. CFD was used to generate the ground truth data. To inform the model about body temperature, a Rayleigh-scale range was considered for body segments. Geometry representation was simplified by utilizing only the boundary nodes, thereby eliminating the need for complex distance functions. The model accuracy was examined for free convection over elliptical cylinders of various shapes and circular cylinders at different positions within an enclosure. Attention U-Net (AUN) achieved the highest R-squared scores for Nusselt numbers at the enclosure wall, with values of 0.9933 and 0.9822 for elliptical and circular cylinder cases, respectively. It could improve the accuracy by 80.6%, with a 14.5% increase in complexity compared to the conventional auto-encoder. GPU accelerated training by 45 times and testing by 10 times compared to a single CPU core.

The present study proposes a deep learning regression approach for parametric modeling of thermal fluid flows based on a combination of convolutional auto-encoders (CAE) and neural networks (NN). Steady-state mixed convection inside concentric annulus with rotating outer wall is selected as the test case. Computational fluid dynamics (CFD) is used to generate the training dataset. The suggested regression model comprises of three steps. In the first step, the thermo-fluid field is dimensionally reduced by the convolutional encoder. Afterward, a multi-layer NN is connected to the encoded layer to generalize the reduced space with respect to the given Rayleigh (Ra) and Reynolds (Re) numbers. In the last step, the reconstruction process is completed by transforming back the generalized encoded data to the original dimension by the convolutional decoder. Such an approach can be interpreted as an estimation model Z which predicts the steady-state thermal field F as a function of input parameters, Ra and Re numbers, i.e. F=Z(Ra,Re). The model shows promising potential in predicting details of temperature and vorticity fields for a range of 103≤Ra≤106 and 0≤Re≤600. In average, there is 0.085 root-mean-squared error between the simulation (ground truth) and the predicted wall Nusselt numbers.

Adsorption thermal energy storage has received considerable attention as it can overcome the mismatch between supply and demand of renewables, providing high energy storage per volume. In the packed bed adsorption thermal energy storage, pressure drop is of key concern since higher pressure drop leads into lower energy storage efficiency. In this paper, an experimental and modeling investigation on the pressure drop inside the adsorption packed beds is performed. An accurate semi-analytical closed-form relationship is proposed to calculate the pressure drop inside a column of adsorbent materials, taking into account the Laplacian friction, as well as the inertial effects. The model covers a wide range of porosity, between low-permeability medium, a dense packed bed of spherical particles, and high-permeability media, a pure viscous fluid. A modified permeability is defined to consider the inertial effect for a moderate range of the particle Reynolds number (0 < Re-p < 300). An experimental apparatus is designed for measuring the pressure drop for different bed sizes and inlet air velocities. The proposed model shows good agreement with the experimental data with the relative difference of 7.6% at 0.73 m/s for silica gel and 15.3% at 0.84 m/s for zeolite 4A packed beds. The experiment reveals that the effect of water uptake on the pressure drop of packed bed with wet adsorbent is negligible in the tested particle Reynolds number range, with a relative difference of less than 1.0% compared to dry adsorbent for 18-30 cm long columns. The proposed formula for pressure drop, consequently, can be applicable for wet adsorbents regardless of the water uptake amount, with a good level of accuracy. Moreover, the analytical model shows up to +/- 2% change in pressure drop due to heat of adsorption of the tested adsorber columns.

To better quantify measurement uncertainty associated with solid precipitation gauges, time averaged and transient turbulent flow over gauges is studied using a Computational Fluid Dynamic (CFD) approach. By numerically solving the Reynolds-Averaged Navier-Stokes (RANS) equations, vortex formation in the bulk structure of the flow is examined. To study transient effects, Large-Eddy Simulation (LES) quantified the turbulence and eddy dynamics. The simulated gauge is a Geonor T200B gauge shielded with Alter slats for a wind velocity ranging from 1 to 10 m/s. A multi-layer mesh generation technique allows the capture of a wide range of turbulent length scales. Gauge exposure effects on the wind characteristics are investigated for unshielded and shielded cases. Updrafting results in an effective collection area of only 30% of the orifice for capturing precipitation. The RANS results show wind momentum reduction above the gauge orifice from shielding with vertical slats assists catch efficiency of the gauge. However, tilting the slats in high winds degenerates shielding benefits. LES results show considerable production of turbulence kinetic energy near the inlet orifice area. Transient analysis with LES demonstrates that the shields intensify the turbulence near the gauge. With the addition of wind shields, the flow momentum above the gauge is reduced.

The response of precipitation to turbulent fluctuations near gauges is studied using time-averaged (RANS) and unsteady (LES) turbulence modeling. Updrafting effects on catch performance are analyzed for unshielded and shielded gauges. The effective precipitation catchment area of the gauge for both wind-induced effects and snowflake characteristics is found to reduce significantly for small particles in high winds but can be partly recovered by shielding. The variation in the amount of precipitation caught is quantified for different free-stream wind speeds using LES and RANS. The fluctuations, captured with LES are analyzed to determine the local structure of eddies near the orifice plane. Wind-induced drag on precipitates are modeled for a wide range of particle Reynolds numbers from low speed Stokes flow condition to high speed flows with inertial effects. Results show noticeable effect of drag–force model on catch performance calculation of precipitation gauges with uncertainties of up to 40% in high winds and large snowflake sizes. Finally, particle–wall collision on the catch performance is studied for different restitution conditions. These simulations have differences of up to 5% in catch performance for large particle sizes in high winds, dependent on whether the particles undergo elastic or plastic collisions. Comparing RANS and LES results, turbulence fluctuations show a considerable influence on shielding performance degeneration at high winds. Double shielding the gauge can improve efficiency by maintaining a lower fluctuation-to-mean catch ratio as wind speed increases.

Sorption processes in packed beds are inherently transient and the maximum vapor removal occurs at the beginning of the adsorption cycle. Consequently, temperature swing adsorption (TSA), consisting of charging (adsorption) and regeneration (desorption), is required in order to frequently refresh the packed bed and provide sustainable dehumidification. In this study, the effectiveness of thin desiccant columns in controlling air humidity under periodic charge and regeneration cycles is analyzed. An experimental apparatus was designed to measure the rate of adsorption and desorption under different operating conditions. The TSA strategy is imposed by low and high temperature heat sources. Moreover, a numerical one-dimensional transient model is provided and validated by the experiments under different ambient conditions. Using this model, the effect of design factors such as cycle time, airflow, and regeneration temperature are investigated in order to optimize the performance of the desiccant dehumidifier.

An optimized implementation of a block tridiagonal solver based on the block cyclic reduction (BCR) algorithm is introduced and its portability to graphics processing units (GPUs) is explored. The computations are performed on the NVIDIA GTX480 GPU. The results are compared with those obtained on a single core of Intel Core i7-920 (2.67 GHz) in terms of calculation runtime. The BCR linear solver achieves the maximum speedup of 5.84x with block size of 32 over the CPU Thomas algorithm in double precision. The proposed BCR solver is applied to discontinuous Galerkin (DG) simulations on structured grids via alternating direction implicit (ADI) scheme. The GPU performance of the entire computational fluid dynamics (CFD) code is studied for different compressible inviscid flow test cases. For a general mesh with quadrilateral elements, the ADI-DG solver achieves the maximum total speedup of 7.45x for the piecewise quadratic solution over the CPU platform in double precision.

Finite-element based CFD solvers like the family of Discontinuous Galerkin (DG) solvers suffer severely from inaccurate boundary reconstruction.In this matter, developing an accurate and flexible strategy is highly demanded to provide high-order curved boundary representation in DG simulations.In this paper, a general framework is introduced to design the curved elements in discontinuous Galerkin finite-element (DGFEM) simulations.The aim is to connect the boundary to the surrounding mesh by defining an appropriate set of basis functions which deliver the curvature information inside the mesh region adjacent to the boundary.This information is then used in flux integral calculations.The proposed framework is applied in Lagragian and Hermitian boundary representations.The efficiency of the method is analyzed for compressible inviscid flow test cases using the discontinuous Galerkin scheme.It is illustrated that using the curved-side elements in the present approach, is adequate to reduce the artificial entropy generation near the boundaries.This leads to the simulations with the desired order of accuracy.The results show a well consistency in h/p-refinement which advocates the use of the proposed approach in high-order CFD simulations.

In this paper, the performance of the Cyclic Reduction (CR) algorithm for solving tridiagonal systems is improved with the aid of efficient global memory transactions on Graphics Processing Units (GPU). To achieve maximum memory throughput with a lower computational runtime, two different Sort algorithms are introduced for reordering the initial system of equations: direct and step-by-step. It is shown that the latter method is well-fitted to modern GPUs and achieves speedup of up to 3.47x in single precision and 2.1 x in double precision compared to the CPU Thomas algorithm. By benefiting from the new global memory implementation, the CR solver could run 2 x -100 x faster compared to previous works on parallel tridiagonal solvers. The CR solver is also applied to 2D & 3D compressible viscous flow simulations using the high-order compact finite-difference scheme. In this matter, the procedure of filtering, primitive variables, and flux derivative calculations are carried out by using the parallel tridiagonal solver on the GPU device. The GPU-accelerated calculations achieve speedups between 1.9 x -15.2 x in 2D and 6.4x-20.3x in 3D simulations for different grid sizes compared to CPU computations. The computations are performed on the NVIDIA GTX480 GPU. The obtained results are compared to those achieved on a single core of Intel Core 2 Duo (2.7 GHz, 2 MB cache) in terms of calculation runtime. (C) 2013 Elsevier Ltd. All rights reserved.

This paper introduces a new super-parametric element for representing solid-side curved boundary. This type of element is used for simulating fluid flow with discontinuous Galerkin method. Hermitian type piecewise cubic spline is employed as the high-order polynomial for modeling the curved boundary. The two-dimensional geometrical shape functions are extended for quadrilateral elements. The effect of curved approximation on the numerical accuracy of fluid flow simulation is studied for smooth cases. The required local surface derivatives on element boundaries are computed both from the exact surface profile differentiations and by the use of finite-difference discretizations. In this case, different types of approximation methods are presented to evaluate the geometrical derivatives and the accuracy of the resulting solutions is compared. A significant accuracy improvement in the flow computation in vicinity of the curved boundaries has been achieved using the curvedside element with Hermitian surface representation.

In this paper, the non-Fourier heat conduction problem in a finite slab with volumetric heat generation is investigated numerically. A linear dependence on space for heat generation has been considered. MacCormack's explicit predictor-corrector scheme is used for solving the wave propagation problem. Results show that the space-dependent heat generation strongly affects the temperature distribution. A temperature peak in the slab has been observed due to superposition of thermal waves. It has been shown that the magnitude of the temperature peak increases with increasing the slope of dimensionless heat generation function. In order to validate the approach, the results have been compared with the analytical solution which shows an excellent agreement.

An experimental study was conducted to obtain a correlation for free convection heat transfer from isothermal elliptic tubes of minor to major axis ratios of ∈ = 0.53, 0.67, 0.8, and 1 confined between two adiabatic walls. Local and average Nusselt numbers were determined for several different tube axis ratios, Rayleigh numbers, and wall spacings by means of the Mach-Zehnder Interferometery (MZI) technique. For each tube axis ratio, the Rayleigh number varied from 1,000–2,750 and the wall spacing to tube minor axis ratio from 1.25–∞. For all the tube axis ratios, the periphery and length of the tubes were kept constant. Experimental data are presented with a correlation which gives the average Nusselt number as a function of the tube axis ratio, Rayleigh number, and wall spacing to tube minor axis ratio. As the tube axis ratio decreases the average heat transfer coefficient increases. Also, for a constant Rayleigh number and tube axis ratio there is an optimum wall spacing which maximizes the heat transferred from the tube.

The multigrid method has proved to be effective for a large class of numerical methods. In this study, a strategy based on Full Approximation Storage (FAS) scheme is implemented together with Full Multigrid Algorithm (FMG) to accelerate convergence of steady state solutions of the two-dimensional compressible Euler equations on Graphics Processing Unit (GPU). The Beam and Warming linearization scheme in curvilinear coordinates is used to discretize the governing equation. The second-order central and the fourth-order compact finite-difference schemes are applied for spatial discretization. A high-performance GPU-implemented block-tridiagonal solver based on Block Cyclic Reduction (BCR) algorithm is utilized. The proposed BCR solver is applied to finitedifference discretization on structured grids via Alternating Direction Implicit (ADI) scheme. Attention is directed towards the computational performance of the V-cycle and W-cycle multigrid strategies in two and three grid levels using the NVIDIA GTX480 graphics card. Speedups between 2x–6.2x are achieved in comparison to the Intel Core i7-920 2.67GHz CPU for different grid sizes.