Step-by-step iteration multi-objective optimization design method and system for heat exchanger
Through step-by-step iterative multi-objective optimization design method, combined with deep neural networks and genetic optimization algorithms, the problem of insufficient attention to compactness in PCHE design is solved, more efficient multi-objective optimization is achieved, and the comprehensive performance of PCHE is improved.
Patent Information
- Application Number
- CN202510577063.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-06
- Publication Date
- 2025-08-29
AI Technical Summary
When optimizing the design of printed circuit board heat exchangers (PCHEs), the prior art mainly focuses on flow heat transfer performance, while paying less attention to compactness and other aspects, resulting in insufficient design efficiency and effect.
The step-by-step iterative multi-objective optimization design method is adopted, combining experimental design, numerical simulation, deep neural network and genetic optimization algorithm, and by establishing alternative models and multi-objective genetic optimization algorithms, key optimization variables are screened, and the optimization scope is gradually narrowed, and a low-cost and high-precision multi-objective optimization design is achieved.
Effectively screening out key optimization variables improves the comprehensive performance of PCHE, especially in terms of heat exchange per unit volume, average Nussel number and average friction coefficient. Reducing the lateral spacing of the fins can improve the heat exchange per unit volume, and increasing the lateral spacing of the fins can reduce the friction coefficient, providing new design calculation methods and theoretical support.
Smart Images

Figure CN120562249A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of heat exchangers, and in particular relates to a step-by-step iterative multi-objective optimization design method and system for a heat exchanger. Background Art
[0002] Based on the physical properties of supercritical carbon dioxide (S-CO2), such as high density, low viscosity, and stable chemical properties, the S-CO2 power cycle has the advantages of high cycle thermal efficiency, wide adaptability to heat source temperature range, and high power density. It is one of the most promising next-generation energy conversion devices and has broad application prospects in the fields of nuclear energy, solar energy, fossil energy, and waste heat recovery.
[0003] In the S-CO2 Brayton Cycle (SCBC), the operating pressure is relatively high, exceeding the critical pressure of CO2 at 7.38 MPa, and the maximum pressure typically exceeds 15 MPa. Therefore, the heat exchanger must be resistant to high pressure. Traditional heat exchangers, due to limitations such as manufacturing processes, have low pressure limits and cannot meet the requirements of SCBC. The Printed Circuit Heat Exchanger (PCHE) is a new type of compact heat exchanger. Its heat exchange core is composed of a large number of metal heat exchange plates. Microchannels are machined on the plates through a photochemical etching process, and the plates are connected by diffusion welding. This makes the PCHE compact and capable of withstanding high temperatures and high pressures. Under the same thermal load conditions, the volume of a shell-and-tube heat exchanger is only 1 / 6, making it lightweight and easily modularizable. The core can withstand temperatures up to 900°C and pressures up to 40 MPa. The PCHE meets the SCBC requirements for heat exchange devices and has been widely used in the design and demonstration and verification platform construction of SCBC.
[0004] The flow and heat transfer characteristics of PCHE are mainly affected by the channel structure. According to the channel structure, PCHE can be divided into continuous channels and discrete fin channels. Continuous channels include straight channels, Z-shaped, wavy, trapezoidal, spiral and other forms; built-in discrete fin channels include S-shaped fins, C-shaped fins and wing-shaped fins. The heat transfer area in the continuous channel is larger, and the bending parts can be added to improve the heat transfer effect, but the bending parts will cause fluid separation and generate large pressure losses. Compared with the continuous channel, the discrete fin channel can improve the overall performance of the heat exchanger by selecting a low flow resistance fin shape and rationally designing the fin distribution position, thus outperforming the continuous channel. Among them, the application of wing-shaped fins can significantly reduce the flow resistance loss along the way, which is a current research hotspot.
[0005] The cross-sectional shape parameters of airfoils are key geometric parameters that influence the flow and heat transfer characteristics of PCHEs. Some researchers have conducted research based on the NACA series of airfoils and proposed improved designs. The existing art replaces the rear portion of the NACA 0020 airfoil with an inward concave edge tangent to the maximum inscribed circle. Compared to the original design, the PCHE with this new airfoil fin exhibits better overall performance and reduced buildup dissipation. The existing art investigated the effects of airfoil fin camber, maximum camber location, thickness, lateral spacing, and longitudinal spacing on the PCHE's thermal-hydraulic characteristics. The results showed that thickness has the greatest impact on the thermal-hydraulic characteristics. The existing art proposed a three-dimensional fin with a variable cross-section along the height direction. This 3D fin significantly reduces the impact area of the fluid, helping to reduce local flow resistance caused by large velocity gradients, thereby improving overall heat transfer performance. The existing art also investigated the effects of the maximum thickness, maximum thickness location, and leading edge radius of the airfoil fin on the channel's thermal-hydraulic performance. The mechanism by which geometric factors influence flow and heat transfer processes was analyzed from the perspectives of turbulent kinetic energy, boundary layer thickness, the synergy between velocity and temperature gradient fields, and local entropy production. A new slotted fusiform fin structure has been proposed. The friction coefficient of the fin channel in this new structure is reduced by 18.69%-20.01% compared to the original structure, while maintaining a nearly constant overall heat transfer rate.
[0006] The distribution parameters of airfoil fins determine the extent to which the front-row fin disturbance affects the flow and heat transfer process in the rear row, and thus significantly influence the flow and heat transfer characteristics of PCHEs. Prior research has studied the effects of non-uniform airfoil fin placement on flow and heat transfer performance. Results show that non-uniform fin placement can improve the uniformity of the channel temperature difference distribution, thereby enhancing heat transfer. A sparse front-end, dense back-end arrangement is optimal, improving overall performance by 23-29% compared to a uniform arrangement and by 2-7.6% compared to a dense front-end, sparse back-end arrangement. Prior research has also studied the effect of the airfoil fin angle of attack on the S-CO2 flow and heat transfer performance in PCHE channels. Results show that the optimal overall performance is achieved when the airfoil fin angle of attack is in the range of 15°-25°. Prior research has proposed a novel channel structure with a highly staggered distribution based on the NACA 0025 airfoil. Results show that while achieving the same overall performance, this novel channel structure can effectively reduce fin volume, thereby achieving a lightweight PCHE design. Prior research has proposed a novel airfoil fin PCHE with tip clearance. Results show that utilizing tip clearance can reduce flow resistance while enhancing heat transfer. Existing technology involves inserting vortex generators into airfoil-shaped fin channels and numerically studying the flow and heat transfer characteristics of supercritical liquefied natural gas within them. The results show that combining the two can improve overall heat transfer performance.
[0007] Current research has primarily analyzed the influence of parameters such as chord length, maximum thickness, maximum curvature, and location of maximum curvature on the thermal-hydraulic characteristics of the channel. However, most studies have been based on single-channel models, with the wall boundary conditions set to fixed values, which differs significantly from the actual working process of the regenerator.
[0008] Optimizing the cross-sectional shape and position distribution of the fins is an effective way to improve the comprehensive thermal-hydraulic performance of PCHEs. Due to the complex structure of PCHEs and the drastic variations in the physical properties of supercritical fluids, the thermal-hydraulic characteristics of PCHEs are influenced by numerous factors, resulting in complex and time-consuming solutions. Therefore, optimizing the structural parameter space and rapidly determining the thermal-hydraulic characteristic parameters for specific parameter combinations are key to optimizing the design.
[0009] Combining existing technologies with CFD (Computational Fluid Dynamics) simulation, machine learning, and optimization algorithms, researchers conducted optimization studies on four parameters of the PCHE (Physical Helical Energizer) (PCHE): maximum thickness, maximum thickness location, lateral spacing, and staggered spacing. Results showed that shifting the maximum thickness location backward improved the overall performance of the PCHE. After optimization, the coefficient of performance (COP) improved by approximately 6.2% compared to the basic structure at Re = 45,000.
[0010] The existing technology uses PCHE as a micro gas turbine regenerator, and takes heat transfer rate, pressure drop and compactness as the goals. Using the NSGA-II optimization algorithm, a multi-objective optimization design of the airfoil fin structure parameters and arrangement parameters is carried out, and the optimal solution is given according to the design requirements.
[0011] A multi-objective optimization design of an airfoil-fin printed circuit heat exchanger was conducted using an improved PSO-BP neural network and NSGA-II algorithm. The optimization objectives were to maximize the Nusselt number (Nu) and minimize the Fanning friction coefficient (f). Results showed that the best overall performance was achieved when the airfoil fins were arranged in a completely staggered pattern. The TOPSIS method was used to identify the optimal decision point for achieving high heat transfer performance and low flow resistance.
[0012] Existing technology uses PCHE as a hydrogen cooler and establishes a cost model using An economic approach was used to evaluate its performance. A multi-objective optimization approach was used to optimize the channel's geometric parameters, revealing that the channel radius had the most significant impact on construction and operating costs. Compared to coaxial tube evaporators, PCHE demonstrated superior performance in terms of volume and total cost, reducing them by 88.0% and 4.0%, respectively.
[0013] Existing technology for PCHE uses heat transfer rate as the objective function and the airfoil pressure drop as the constraint, conducting multi-layer topology optimization. Compared to the airfoil structure, the optimized structure reduces pump power by 17.0% and increases heat transfer by 10.8% at Re of 300. At Re of 20,000, pump power is reduced by 52.0% and heat transfer is increased by 1.2%.
[0014] The existing technology combines the response surface methodology and the NSGA-II genetic optimization algorithm to optimize the design of a straight-channel printed circuit board heat exchanger. Compared with the original design, the new structure's performance evaluation index (PEC) increased by 8.5% and its entropy generation rate decreased by 9.98%.
[0015] A multi-objective optimization design of the Z-shaped printed circuit heat exchanger structure was conducted. The results showed that the channel with a regular hexagonal cross-section has better thermal and hydraulic performance.
[0016] A novel approach to predicting flow nonuniformity within PCHEs, based on dimensionless number analysis and machine learning, yielded a neural network response surface consisting of 162 design points reflecting flow characteristics and velocity distribution. Optimizing the structure using the NSGA-II algorithm reduced flow nonuniformity by 93.39% while increasing pressure drop by only 32.67%.
[0017] Current methods for rapidly determining thermal-hydraulic characteristic parameters primarily rely on establishing surrogate models based on data from three-dimensional numerical simulations or experiments. These methods include correlation fitting, response surface modeling, artificial neural network prediction, and one-dimensional reduced-order model simulation. Optimization algorithms include particle swarm optimization, ant colony optimization, and the second-generation non-dominated solution sorting genetic algorithm (NSGA-II).
[0018] Artificial neural networks offer the advantages of easy scalability, fast computation, and high prediction accuracy, while genetic optimization algorithms offer strong search capabilities and robustness for complex problems. Combining these two approaches can significantly improve the efficiency and effectiveness of PCHE optimization design, further enhancing its overall performance and facilitating its practical engineering applications. However, current research on this topic is insufficient. The optimization objective primarily focuses on the flow and heat transfer performance of the PCHE, with less attention paid to factors such as compactness.
[0019] Through the above analysis, the problems and defects of the existing technology are as follows:
[0020] Current research is insufficient. The optimization objective is mainly focused on the flow and heat transfer performance of PCHE, with less attention paid to compactness and other factors. Summary of the Invention
[0021] In response to the problems existing in the prior art, the present invention provides a step-by-step iterative multi-objective optimization design method and system for a heat exchanger.
[0022] The present invention is implemented as follows: a step-by-step iterative multi-objective optimization design method for a heat exchanger includes:
[0023] Step 1, establish a substitution model;
[0024] First, a fully parameterized geometric model of a wing-shaped fin channel (PCHE) was established. Then, the orthogonal design method was used to combine different parameter values and perform numerical simulations. The sensitivity of the heat transfer rate per unit volume, heat transfer characteristics, and flow characteristics to each parameter was analyzed based on the obtained results, and the optimized parameters were screened. Next, the central composite design method was used to expand the parameter range and perform numerical simulations to form a representative dataset of optimized variables and thermal-hydraulic characteristic indicators. The resulting dataset was then used to train a neural network to achieve rapid prediction of the thermal-hydraulic characteristic parameters of the AFPH.
[0025] Step 2: Carry out the first optimization design; use the neural network and genetic optimization algorithm to carry out the first multi-objective optimization design, and determine the secondary optimization area based on the decision points and extreme points;
[0026] Step 3: Carry out the second optimization design; according to the secondary optimization area, determine the parameter range, supplement the numerical simulation calculation, and iteratively update the deep neural network model, and finally carry out the second optimization design.
[0027] Furthermore, the alternative model:
[0028] (1) Parametric geometric model;
[0029] A pair of hot and cold channels were selected for modeling and simulation. The airfoil fin chord length is L, and the cross-sectional shape is modeled based on the NACA00ZZ series symmetrical airfoil, where ZZ represents the percentage of the maximum half-thickness to the chord length. The total channel length is 130 mm, the number of fins per row is N, the longitudinal spacing of the fins is Lx, and the transverse spacing is δ. The fins run through the channel, and the hot and cold channels are arranged in a countercurrent manner.
[0030] (2) Flow and heat transfer control equations and boundary conditions;
[0031] (3) Model validation;
[0032] (4) Performance indicators;
[0033] (5) Deep neural networks;
[0034] (6) Multi-objective genetic optimization algorithm.
[0035] Furthermore, the flow heat transfer control equation and boundary conditions are:
[0036] The flow and heat transfer process of S-CO2 in PCHE satisfies the conservation of mass, momentum, and energy, considering the following assumptions: (1) gravity is ignored; (2) heat loss is ignored; (3) the flow process is a three-dimensional steady-state turbulent flow; the governing equations of the flow and heat transfer process are shown in Equations (1)-(3);
[0037] Continuity equation:
[0038]
[0039] Momentum equation:
[0040]
[0041] Energy equation:
[0042]
[0043] Where ρ is density (kg / m3), u is velocity (m / s), μ is dynamic viscosity (Pa·s), μt is turbulent viscosity (Pa·s), p is pressure (Pa), cp is specific heat capacity at constant pressure (J / (kg·K)), T is temperature (K), λ is thermal conductivity (W / (m·K)), and Φ is turbulent energy dissipation rate (W / m3).
[0044] The governing equation for heat conduction in a metal matrix is a three-dimensional steady-state heat conduction differential equation without an internal heat source, as shown in Equation (4);
[0045]
[0046] The above control equations are solved using Fluent software, with the convergence residual of the energy equation set to 10-9 and the convergence residual of other equations set to 10-6.
[0047] Furthermore, the model verifies that:
[0048] 1) Grid independence verification
[0049] Fluent-meshing was used to mesh the established geometric model with a polyhedra type mesh. This type of mesh has good adaptability to complex geometric structures and can reduce the total number of meshes compared to tetrahedral meshes, resulting in higher computational efficiency. Boundary layer meshes were created in the fluid region near the fluid-solid interface, with the first layer height set to 0.001 mm.
[0050] 2) Verification of numerical simulation accuracy
[0051] The calculation results of the convective heat transfer coefficient h, which characterizes the heat transfer characteristics, and the Fanning friction coefficient f, which characterizes the flow characteristics, were compared.
[0052] Furthermore, the performance indicators:
[0053] The airfoil fins are dispersed, so the interface shape of the S-CO2 flow area changes with position. The hydraulic diameter Dh is defined based on the fluid area volume V and the fluid-solid interface area S:
[0054]
[0055] According to the inlet flow The physical properties of S-CO2 can be determined by the inlet temperature, and then the Reynolds number Rein, which characterizes the inlet flow state, can be obtained: Re symbols are italicized in the text, but in normal form in the equations; generally, symbols are italicized and units are normal form;
[0056]
[0057] Where Ain is the channel inlet cross-sectional area (m2), μ is the dynamic viscosity (Pa·s);
[0058] The local Nusselt number NuL is defined as follows:
[0059]
[0060] Where λ is the thermal conductivity, hl is the local convection heat transfer coefficient, and hl is calculated using formula (8); Tw,L and Tb,L are the average boundary heat flux density (W / m2), wall temperature (K), and cross-sectional average temperature (K), respectively. A series of cross sections parallel to the YZ plane are taken along the flow direction. Local parameters are calculated at each cross section, and the global parameter is the average of all local parameters. Local parameters are denoted by the subscript L.
[0061] The Fanning friction coefficient f is defined as follows:
[0062]
[0063] Where pin, pout, uin, ρin, and Lf represent the inlet pressure (Pa), outlet pressure (Pa), inlet cross-sectional velocity (m / s), inlet cross-sectional density (kg / m3), and channel length (m), respectively;
[0064] The heat transfer per unit volume qv is defined as follows:
[0065]
[0066] Where Q is the heat transfer capacity (W) and V is the volume of the heat exchanger (m3).
[0067] Furthermore, the multi-objective genetic optimization algorithm:
[0068] The multi-objective optimization problem is shown in formula (11);
[0069]
[0070] Another object of the present invention is to provide a step-by-step iterative multi-objective optimization design system for a heat exchanger, comprising:
[0071] The surrogate model building module is used to establish a fully parameterized geometric model of the airfoil-shaped fin channel PCHE. Then, the orthogonal design method is used to combine different values of each parameter and conduct numerical simulation calculations. Based on the obtained results, the sensitivity of the heat transfer rate per unit volume, heat transfer characteristics, and flow characteristics to each parameter is analyzed to screen the optimized parameters. Next, the central composite design method is used to expand the parameter value range and conduct numerical simulation calculations to form a representative data set of optimization variables and thermal-hydraulic characteristic indicators. The obtained data set is then used to train a neural network to achieve rapid prediction of the thermal-hydraulic characteristic parameters of the AFPH.
[0072] The first optimization design module is used to carry out the first optimization design; it uses a combination of neural network and genetic optimization algorithms to carry out the first multi-objective optimization design, and determines the secondary optimization area based on the decision points and extreme points;
[0073] The second optimization design module is used to carry out the second optimization design; according to the secondary optimization area, the parameter range is determined, the numerical simulation calculation is supplemented, and the deep neural network model is iteratively updated, and finally the second optimization design is carried out.
[0074] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the step-by-step iterative multi-objective optimization design method for the heat exchanger.
[0075] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the step-by-step iterative multi-objective optimization design method for the heat exchanger.
[0076] Another object of the present invention is to provide an information data processing terminal, which is used to implement the step-by-step iterative multi-objective optimization design system for the heat exchanger.
[0077] In combination with the above technical solutions and the technical problems solved, please analyze the advantages and positive effects of the technical solutions to be protected by the present invention from the following aspects:
[0078] The present invention studies its thermal-hydraulic characteristics and proposes a step-by-step iterative multi-objective optimization strategy. It comprehensively utilizes experimental design methods, numerical simulation, deep neural networks, and genetic optimization algorithms to carry out multi-objective optimization design, obtains the Pareto frontier, and uses LINMAP and TOPSIS methods to make decisions. The performance indicators of the decision points were numerically verified. The maximum error of the results obtained from the first optimization design was 25.3%, and the maximum error of the results obtained from the second optimization design was only 0.45%. Calculations show that the step-by-step iterative optimization strategy effectively screens out key optimization variables and gradually narrows the optimization range, achieving low-cost, high-precision multi-objective optimization design within a wide range of multiple parameters. From the perspective of the three performance indicators of heat transfer per unit volume, average Nusselt number, and average friction coefficient, large flow and small fins have better performance; reducing the lateral spacing of the fins can increase the heat transfer per unit volume, and increasing the lateral spacing of the fins can reduce the friction coefficient. The method and conclusions of the present invention provide new design calculation methods and theoretical support for heat exchanger design and optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0079] Figure 1 It is a flow chart of the step-by-step iterative multi-objective optimization design method for a heat exchanger provided by an embodiment of the present invention.
[0080] Figure 2 It is a structural block diagram of the heat exchanger step-by-step iterative multi-objective optimization design system provided by an embodiment of the present invention.
[0081] Figure 3 This is a step-by-step iterative optimization flowchart based on a deep neural network provided by an embodiment of the present invention.
[0082] Figure 4 It is a schematic diagram of a geometric model provided by an embodiment of the present invention.
[0083] Figure 5 Schematic diagram of a boundary layer grid provided by an embodiment of the present invention.
[0084] Figure 6 3 is a graph showing the outlet temperature and pressure drop changing with the number of grids, as provided in an embodiment of the present invention.
[0085] Figure 7 This is a comparison chart of the numerical simulation results provided by the embodiment of the present invention and the results of the literature
[56] .
[0086] Figure 8 Schematic diagram of a deep neural network provided by an embodiment of the present invention.
[0087] Figure 9 4 is a factor level trend diagram of the surface heat transfer coefficient provided by an embodiment of the present invention.
[0088] Figure 10This is a factor level trend diagram of the inlet and outlet pressure drops of the cold and hot channels provided by an embodiment of the present invention.
[0089] Figure 11 This is a factor level trend diagram of the heat transfer per unit volume and the heat transfer area per unit volume provided by an embodiment of the present invention.
[0090] Figure 12 This is a comparison chart of the DNN prediction results and CFD numerical simulation results of Q / V provided by an embodiment of the present invention.
[0091] Figure 13 This is a comparison chart of the DNN prediction results and CFD numerical simulation results of Nu provided by an embodiment of the present invention.
[0092] Figure 14 This is a comparison chart of the DNN prediction results and simulation results for f provided by an embodiment of the present invention.
[0093] Figure 15 This is the Pareto front and decision point diagram of the first optimization design provided by an embodiment of the present invention.
[0094] Figure 16 This is a Pareto frontier and decision point diagram of the second optimization design provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0095] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0096] like Figure 1 As shown, a step-by-step iterative multi-objective optimization design method for a heat exchanger provided by an embodiment of the present invention includes the following steps:
[0097] S101, establish a substitution model;
[0098] First, a fully parameterized geometric model of a wing-shaped fin channel (PCHE) was established. Then, the orthogonal design method was used to combine different parameter values and perform numerical simulations. The sensitivity of the heat transfer rate per unit volume, heat transfer characteristics, and flow characteristics to each parameter was analyzed based on the obtained results, and the optimized parameters were screened. Next, the central composite design method was used to expand the parameter range and perform numerical simulations to form a representative dataset of optimized variables and thermal-hydraulic characteristic indicators. The resulting dataset was then used to train a neural network to achieve rapid prediction of the thermal-hydraulic characteristic parameters of the AFPH.
[0099] S102, carry out the first optimization design; use the neural network and genetic optimization algorithm to carry out the first multi-objective optimization design, and determine the secondary optimization area based on the decision points and extreme points;
[0100] S103, carry out the second optimization design; according to the secondary optimization area, determine the parameter range, supplement the numerical simulation calculation, and iteratively update the deep neural network model, and finally carry out the second optimization design.
[0101] The alternative model provided by the embodiment of the present invention is:
[0102] (1) Parametric geometric model;
[0103] A pair of hot and cold channels were selected for modeling and simulation. The airfoil fin chord length is L, and the cross-sectional shape is modeled based on the NACA00ZZ series symmetrical airfoil, where ZZ represents the percentage of the maximum half-thickness to the chord length. The total channel length is 130 mm, the number of fins per row is N, the longitudinal spacing of the fins is Lx, and the transverse spacing is δ. The fins run through the channel, and the hot and cold channels are arranged in a countercurrent manner.
[0104] (2) Flow and heat transfer control equations and boundary conditions;
[0105] (3) Model validation;
[0106] (4) Performance indicators;
[0107] (5) Deep neural networks;
[0108] (6) Multi-objective genetic optimization algorithm.
[0109] The flow heat transfer control equations and boundary conditions provided by the embodiments of the present invention are:
[0110] The flow and heat transfer process of S-CO2 in PCHE satisfies the conservation of mass, momentum, and energy, considering the following assumptions: (1) gravity is ignored; (2) heat loss is ignored; (3) the flow process is a three-dimensional steady-state turbulent flow; the governing equations of the flow and heat transfer process are shown in Equations (1)-(3);
[0111] Continuity equation:
[0112]
[0113] Momentum equation:
[0114]
[0115] Energy equation:
[0116]
[0117] Where ρ is density (kg / m3), u is velocity (m / s), μ is dynamic viscosity (Pa·s), μt is turbulent viscosity (Pa·s), p is pressure (Pa), cp is specific heat capacity at constant pressure (J / (kg·K)), T is temperature (K), λ is thermal conductivity (W / (m·K)), and Φ is turbulent energy dissipation rate (W / m3).
[0118] The governing equation for heat conduction in a metal matrix is a three-dimensional steady-state heat conduction differential equation without an internal heat source, as shown in Equation (4);
[0119]
[0120] The above control equations are solved using Fluent software, with the convergence residual of the energy equation set to 10-9 and the convergence residual of other equations set to 10-6.
[0121] Model verification provided by the embodiment of the present invention:
[0122] 1) Grid independence verification
[0123] Fluent-meshing was used to mesh the established geometric model with a polyhedra type mesh. This type of mesh has good adaptability to complex geometric structures and can reduce the total number of meshes compared to tetrahedral meshes, resulting in higher computational efficiency. Boundary layer meshes were created in the fluid region near the fluid-solid interface, with the first layer height set to 0.001 mm.
[0124] 2) Verification of numerical simulation accuracy
[0125] The calculation results of the convective heat transfer coefficient h, which characterizes the heat transfer characteristics, and the Fanning friction coefficient f, which characterizes the flow characteristics, were compared.
[0126] The performance indicators provided by the embodiments of the present invention are:
[0127] The airfoil fins are dispersed, so the interface shape of the S-CO2 flow area changes with position. The hydraulic diameter Dh is defined based on the fluid area volume V and the fluid-solid interface area S:
[0128]
[0129] According to the inlet flow The physical properties of S-CO2 can be determined by the inlet temperature, and then the Reynolds number Rein, which characterizes the inlet flow state, can be obtained: Re symbols are italicized in the text, but in normal form in the equations; generally, symbols are italicized and units are normal form;
[0130]
[0131] Where Ain is the channel inlet cross-sectional area (m2), μ is the dynamic viscosity (Pa·s);
[0132] The local Nusselt number NuL is defined as follows:
[0133]
[0134] Where λ is the thermal conductivity, hl is the local convection heat transfer coefficient, and hl is calculated using formula (8); Tw,L and Tb,L are the average boundary heat flux density (W / m2), wall temperature (K), and cross-sectional average temperature (K), respectively. A series of cross sections parallel to the YZ plane are taken along the flow direction. Local parameters are calculated at each cross section, and the global parameter is the average of all local parameters. Local parameters are denoted by the subscript L.
[0135] The Fanning friction coefficient f is defined as follows:
[0136]
[0137] Where pin, pout, uin, ρin, and Lf represent the inlet pressure (Pa), outlet pressure (Pa), inlet cross-sectional velocity (m / s), inlet cross-sectional density (kg / m3), and channel length (m), respectively;
[0138] The heat transfer per unit volume qv is defined as follows:
[0139]
[0140] Where Q is the heat transfer capacity (W) and V is the volume of the heat exchanger (m3).
[0141] The multi-objective genetic optimization algorithm provided by the embodiment of the present invention:
[0142] The multi-objective optimization problem is shown in formula (11);
[0143]
[0144] like Figure 2 As shown, an embodiment of the present invention provides a heat exchanger step-by-step iterative multi-objective optimization design system comprising:
[0145] The surrogate model building module is used to establish a fully parameterized geometric model of the airfoil-shaped fin channel PCHE. Then, the orthogonal design method is used to combine different values of each parameter and conduct numerical simulation calculations. Based on the obtained results, the sensitivity of the heat transfer rate per unit volume, heat transfer characteristics, and flow characteristics to each parameter is analyzed to screen the optimized parameters. Next, the central composite design method is used to expand the parameter value range and conduct numerical simulation calculations to form a representative data set of optimization variables and thermal-hydraulic characteristic indicators. The obtained data set is then used to train a neural network to achieve rapid prediction of the thermal-hydraulic characteristic parameters of the AFPH.
[0146] The first optimization design module is used to carry out the first optimization design; it uses a combination of neural network and genetic optimization algorithms to carry out the first multi-objective optimization design, and determines the secondary optimization area based on the decision points and extreme points;
[0147] The second optimization design module is used to carry out the second optimization design; according to the secondary optimization area, the parameter range is determined, the numerical simulation calculation is supplemented, and the deep neural network model is iteratively updated, and finally the second optimization design is carried out.
[0148] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the step-by-step iterative multi-objective optimization design method for the heat exchanger.
[0149] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the step-by-step iterative multi-objective optimization design method for the heat exchanger.
[0150] Another object of the present invention is to provide an information data processing terminal, which is used to implement the step-by-step iterative multi-objective optimization design system for the heat exchanger.
[0151] The present invention is specifically implemented:
[0152] Based on the above research, the present invention conducts numerical research on the thermal-hydraulic characteristics of the AirfoilFins Printed Circuit Heat Regenerator (AFPCHR) and carries out multi-objective optimization design based on the Deep Neural Network (DNN).
[0153] Step-by-step iterative multi-objective optimization strategy
[0154] The accuracy of multi-objective optimization design results is directly related to the accuracy of DNN prediction results, and the accuracy of DNN depends on the training data. However, CFD calculations take a long time, and the amount of training data is limited by computing power. In order to achieve low-cost, high-precision multi-objective optimization design, a method such as Figure 3 The step-by-step iterative optimization strategy shown.
[0155] The first step is to establish an alternative model. First, a fully parameterized wing-shaped fin channel PCHE geometric model is established. Then, the orthogonal design method is used to combine different value levels of each parameter, and numerical simulation calculations are carried out. Based on the results, the sensitivity of the unit volume heat transfer, heat transfer characteristics and flow characteristics to each parameter is analyzed to screen out the optimized parameters. Then, the central composite design method is used to expand the parameter value range, and numerical simulation calculations are carried out to form a representative data set of optimization variables-thermal-hydraulic characteristic indicators. Subsequently, the obtained data set is used to train a neural network to achieve rapid prediction of the AFPH thermal-hydraulic characteristic parameters.
[0156] The second step is to carry out the first optimization design. The neural network and genetic optimization algorithm are used to carry out the first multi-objective optimization design, and the secondary optimization area is determined based on the decision points and extreme points.
[0157] The third step is to conduct a second optimization design. Based on the second optimization area, the parameter range is determined, numerical simulation calculations are supplemented, and the deep neural network model is iteratively updated before the second optimization design is carried out.
[0158] Mathematical Model
[0159] 1.1 Parametric geometric model
[0160] PCHE contains a large number of hot and cold channels arranged at intervals, which has good symmetry. In order to reduce the simulation calculation cost, the following Figure 4 The pair of hot and cold channels shown in the figure were modeled and simulated. The airfoil fin chord length is L, and the cross-sectional shape is based on the NACA00ZZ series symmetrical airfoil, where ZZ represents the maximum half-thickness as a percentage of the chord length. The total channel length is 130 mm, the number of fins per row is N, the longitudinal spacing between the fins is Lx, and the transverse spacing is δ. The fins extend through the channels, and the hot and cold channels are arranged in countercurrent flow.
[0161] 1.2 Flow and heat transfer control equations and boundary conditions
[0162] The flow and heat transfer process of S-CO2 in the PCHE satisfies the conservation of mass, momentum, and energy, considering the following assumptions: (1) gravity is neglected; (2) heat loss is neglected; and (3) the flow process is a three-dimensional steady-state turbulent flow. The governing equations for the flow and heat transfer process are shown in Equations (1)-(3).
[0163] Continuity equation:
[0164]
[0165] Momentum equation:
[0166]
[0167] Energy equation:
[0168]
[0169] where ρ is density (kg / m³), u is velocity (m / s), μ is dynamic viscosity (Pa·s), μt is turbulent viscosity (Pa·s), p is pressure (Pa), cp is specific heat capacity at constant pressure (J / (kg·K)), T is temperature (K), λ is thermal conductivity (W / (m·K)), and Φ is turbulent energy dissipation rate (W / m³).
[0170] During the flow, the S-CO2 is disturbed by the wing-shaped fins, which produces strong turbulence. The SST k-ω turbulence model is used for simulation. This model takes into account the stability of the k-ω turbulence model near the wall and the accuracy of the k-ε turbulence model far from the wall, and is suitable for the turbulent flow of S-CO2.
[0171] The governing equation for heat conduction in a metal matrix is a three-dimensional steady-state heat conduction differential equation without an internal heat source, as shown in Equation (4).
[0172]
[0173] The boundary condition settings are shown in Table 1.
[0174] Table 1 Boundary conditions
[0175]
[0176] The above control equations were solved using Fluent software, and the convergence residual of the energy equation was set to 10 -9 , the convergence residuals of other equations are set to 10 -6 .
[0177] 1.3 Model Validation
[0178] 1.3.1 Grid independence verification
[0179] Fluent-meshing is used to divide the established geometric model into polyhedra type meshes. This type of mesh has good adaptability to complex geometric structures and can reduce the total number of meshes compared to tetrahedral meshes, which is more efficient. The boundary layer mesh is divided in the fluid area near the fluid-solid interface, and the first layer height is set to 0.001mm. Figure 5 shown.
[0180] Reducing the grid size can improve the calculation accuracy, but the total number of grids will increase rapidly, resulting in higher computational costs. Figure 6 The trend of outlet temperature and pressure drop with the number of grids is shown. It can be seen that when the number of grids reaches 2.76×10 6 After that, the outlet temperature and pressure drop change little. If we continue to increase the number of grids, the result changes within 0.8%, that is, the number of grids has little effect on the calculation results. Considering the calculation accuracy and cost, we choose 2.76×10 6 The numerical simulation is carried out with the mesh size corresponding to the mesh.
[0181] 1.3.2 Verification of numerical simulation accuracy
[0182] The same geometric model and conditions as those in the literature
[56] were used for simulation, and the calculation results of the convective heat transfer coefficient h, which characterizes the heat transfer characteristics, and the Fanning friction coefficient f, which characterizes the flow characteristics, were compared. Figure 7 The maximum error of h is 2.24%, and the maximum error of f is 0.63%. The results are in good agreement, which verifies the reliability of the numerical calculation method of the present invention.
[0183] Literature
[56] Zhang H, Guo J, Cui X, et al. Experimental and numerical investigations of thermal-hydraulic characteristics in a novel airfoil finheat exchanger[J]. International Journal of Heat and Mass Transfer, 2021,175:121333.
[0184] 1.4 Performance Indicators
[0185] The airfoil fins are dispersed, so the interface shape of the S-CO2 flow area changes with position. The hydraulic diameter D is defined based on the fluid area volume V and the fluid-solid interface area S. h for:
[0186]
[0187] According to the inlet flow The physical properties of S-CO2 can be determined by the inlet temperature, and then the Reynolds number Re that characterizes the inlet flow state can be obtained. in :Re symbols are italicized in the text, but in regular font in equations. In general, symbols are italicized and units are regular font.
[0188]
[0189] Where A in is the channel entrance cross-sectional area (m 2 ), μ is the dynamic viscosity (Pa·s).
[0190] Local Nusselt number Nu L The definition is as follows:
[0191]
[0192] Where λ is the thermal conductivity, h l is the local convective heat transfer coefficient, h l Calculate using formula (8). T w,L and T b,L The average heat flux density at the boundary (W / m 2 ), wall temperature (K), and cross-sectional average temperature (K). A series of cross sections parallel to the YZ plane are taken along the flow direction. Local parameters are calculated at each cross section, and the global parameter is the average of all local parameters. Local parameters are denoted by the subscript L.
[0193] The Fanning friction coefficient f is defined as follows:
[0194]
[0195] Among them, p in 、p out 、u in , ρ in and L f They represent inlet pressure (Pa), outlet pressure (Pa), inlet cross-sectional velocity (m / s), inlet cross-sectional density (kg / m 3 ) and channel length (m).
[0196] Heat transfer per unit volume q v The definition is as follows:
[0197]
[0198] Where Q is the heat transfer capacity (W), V is the volume of the heat exchanger (m 3 ).
[0199] 1.5 Deep Neural Networks
[0200] Deep Neural Network (DNN) is a machine learning technology based on artificial neural networks. It imitates the working mode of neurons in the human brain and realizes feature learning and pattern recognition of complex data by constructing a network structure containing multiple levels.
[0201] A deep neural network consists of an input layer, multiple hidden layers, and an output layer. In the input layer, data is fed into the network. The hidden layers then abstract and extract features from the input data, with each layer learning higher-level, more abstract features based on the previous layer. Finally, the processed results are obtained at the output layer. This multi-layered structure gives deep neural networks powerful expressive power, enabling them to learn complex patterns and relationships in data.
[0202] The present invention is based on Keras to build Figure 8 The deep neural network shown in the figure has geometric and operational parameters as input layers, multiple hidden layers in between, and outputs the heat transfer per unit volume, the average Nusselt number of the hot and cold channels, and the average friction coefficient. Training the neural network using a dataset derived from numerical simulations extracts key features from the data and establishes a nonlinear mapping between input parameters and performance indicators. The trained neural network can quickly derive performance indicators for the AFPH based on the input parameters, providing strong support for its optimized design.
[0203] 1.6 Multi-objective genetic optimization algorithm
[0204] The three performance indicators, heat transfer per unit volume (indicating compactness), the Nusselt number (indicating heat transfer performance), and the friction coefficient (indicating flow properties), are not fully synchronized. Traditional single-objective optimization methods struggle to comprehensively consider the optimal compromise between these indicators. Simultaneous optimization of these three performance indicators is necessary. The optimization variables are operating parameters and geometric parameters. The multi-objective optimization problem studied in this paper is shown in Equation (11).
[0205]
[0206] The multi-objective genetic optimization algorithm is an optimization algorithm based on the principles of natural selection and genetic mutation. It can be used to solve the problem of finding the optimal solution among multiple conflicting objective functions. By simulating the biological processes of inheritance, crossover, and mutation, it searches for the optimal solution set that meets multiple objectives in the solution space.
[0207] NSGA-II (Non-dominated Sorting Genetic Algorithm II) is a widely used and representative multi-objective genetic optimization algorithm. Its core features lie in its fast non-dominated sorting and congestion calculation. Through these two mechanisms, NSGA-II balances convergence and diversity during the search process, efficiently finding a uniformly distributed Pareto optimal set of solutions. Within the range of input parameter values, it can search for a range of optimization solutions that meet different performance requirements, providing a wide range of options.
[0208] 1. Results Analysis
[0209] 1.1 Analysis of orthogonal experimental results
[0210] In order to fully study the influence of geometric parameters and operating parameters on the thermal hydraulic characteristics of AFPH, the present invention selects L, ZZ, N, δ and The five parameters are analyzed and the values of each parameter are shown in Table 2. If a full factorial experimental design is used, there are 3125 parameter combinations, which is too expensive to calculate. In order to reduce the calculation cost, representative parameters are selected for combination. 25 (5 6 ) orthogonal array for combination design.
[0211] Table 2 Parameter values
[0212]
[0213] For each parameter combination, the flow and heat transfer process of S-CO2 in AFPH was numerically simulated. Based on the simulation results, the heat transfer characteristics, flow characteristics and heat transfer per unit volume as a function of each parameter were analyzed, and intuitive analysis and variance analysis were carried out in combination with orthogonal experimental design.
[0214] 1.1.1 Analysis of heat transfer characteristics
[0215] AFPH is a partition-type heat exchanger. The heat transfer resistance is mainly reflected in the convection heat transfer between the cold and hot fluids and the wall surface. Therefore, the focus is on analyzing the surface heat transfer coefficients of the cold and hot channels. Figure 9 The factor level trend of the surface heat transfer coefficient is given. It can be seen that the change trends of the cold channel, hot channel and average surface heat transfer coefficients are the same.
[0216] For geometric parameters, h decreases with the increase of L and δ, but is less affected by ZZ and N. This is mainly because the increase of L and δ will lead to a larger channel spacing and an increase in the fluid flow cross section. Under the condition of , the fluid velocity will decrease, so the convective heat transfer intensity will decrease and h will decrease. ZZ and N have little effect on the channel cross section, so their effect on h is also small. Increase and grow rapidly For the average surface heat transfer coefficient, Figure 9 The ranges of the parameters are 706.4, 94.3, 197.4, 769.8 and 1326.4 W / (m 2 ·K), and the order of their influence on the surface heat transfer coefficient is: Comparing the hot and cold channels, we can find that the surface heat transfer coefficient of the hot channel is higher than that of the cold channel. This is because the fluid temperature in the hot channel is high, the pressure is low, and the density is small. Under the condition of high flow rate, the heat exchange intensity is greater.
[0217] Table 3 shows the variance analysis results of the average surface heat transfer coefficient. The results show that L, δ and The impact on it is significant.
[0218] Table 3 Variance analysis of average surface heat transfer coefficient
[0219]
[0220] 1.1.2 Flow characteristics analysis
[0221] Figure 10 The factor-level trends for the inlet and outlet pressure drops of the cold and hot channels are presented, showing that the pressure drops within the cold and hot channels follow the same trend. The inlet and outlet pressure drops are primarily influenced by the fluid velocity and the contact area between the fluid and the solid. While increasing L and δ increases the contact area, it also increases the fluid flow cross-section, decreasing the flow velocity and, overall, reducing the pressure drop. Increasing ZZ and N increases the contact area while having little effect on the flow velocity. Therefore, the pressure drop generally increases with increasing ZZ and N. The increase in flow rate increases, so the pressure drop increases Comparing the cold and hot channels, we can find that the pressure drop in the hot channel is higher, up to 3.4 times that of the cold channel. This is because the fluid in the hot channel has a high temperature, low pressure, and low density. In order to comprehensively consider the impact of various factors on the pressure drop of cold and hot channels, the total pressure drop is analyzed. Figure 10 The ranges of the parameters are 8580.6, 3301.3, 4497.9, 8609.7 and 16525.9 Pa, respectively, and the order of their influence on the total pressure drop is:
[0222] Table 4 shows the variance analysis results of the total pressure drop of the cold and hot channels. The impact on it is significant.
[0223] Table 4 Variance analysis of total pressure drop in cold and hot channels
[0224]
[0225] 1.1.3 Analysis of heat transfer per unit volume
[0226] While ensuring the heat transfer rate, reducing the volume of the heat exchanger can effectively increase the heat transfer rate per unit volume of the SCBC, fully leveraging its compact structure and applying it to confined spaces such as ships. To this end, this paper studies the heat transfer rate per unit volume as an indicator of the compactness of the AFPH. Figure 11 The factor level trends of heat transfer per unit volume and heat transfer area per unit volume are given. The heat transfer per unit volume depends on the convective heat transfer intensity and heat transfer area. The heat transfer per unit volume varies with the fin chord length L, δ and mass flow rate. The change is large, while the fin thickness ZZ and the number of fins N have little effect on it. As L increases, the heat transfer area per unit volume decreases, while the fluid flow cross-section increases, the flow velocity decreases, and the convective heat transfer intensity decreases, which generally reduces the heat transfer per unit volume. As δ increases, the effect on the heat transfer area per unit volume is relatively small, but it will increase the fluid flow cross-section, reduce the flow velocity, and reduce the convective heat transfer intensity, resulting in a decrease in the heat transfer per unit volume. As the value increases, the heat transfer per unit volume increases. Figure 11 The ranges of the parameters are 81.3, 17.4, 23.3, 73.0 and 123.8 W / cm 3 , the impact degree is ranked as follows
[0227] Table 5 shows the variance analysis of the heat transfer per unit volume. The results show that L, δ and The impact on it is significant.
[0228] Table 5 Variance analysis of heat transfer per unit volume
[0229]
[0230] According to the results of orthogonal experimental analysis, for the three indicators of surface heat transfer coefficient, pressure drop of hot and cold channels and heat transfer per unit volume, L, δ and The influence of the three parameters is more significant, so these three parameters are selected as the key variables for the optimization design of AFPH. A representative parameter combination can be designed through the orthogonal table, but it cannot meet the amount of data required for training the neural network. Therefore, L, δ and The three parameters are used as variables, and the central composite design method is further used to supplement the parameter combination and conduct numerical simulations to form a sufficient data set of "optimization variables-thermal-hydraulic performance indicators".
[0231] Deep Neural Network Prediction
[0232] The simulation results were used to establish an input-output data set; the data set was normalized according to the standard normal distribution and randomly divided into a training data set and a test data set in a ratio of 9:1; the training data set was used to train the node parameters in the neural network. During the training process, the mean square error was selected as the error function; and finally, an alternative model of the flow and heat transfer characteristics of AFPH was established. The prediction results of the deep neural network for AFPH performance indicators were compared with the CFD numerical simulation results. Figure 12-14 shown.
[0233] Figure 12 Comparisons of DNN predictions of Q / V with CFD simulations are shown. The relative error of the DNN predictions compared to the simulation results is within 6.5%. The maximum relative error of Q / V on the training set is 5.69%, and on the test set is 2.55%.
[0234] Figure 13 The DNN predictions for the Nu number of cold and hot aisles were compared with simulation results. The relative error of the DNN predictions compared to the simulation results was within 6.5%. The maximum relative error for the Nu number of cold aisles was 0.58% on the training set and 1.10% on the test set. The maximum relative error for the Nu number of hot aisles was 6.14% on the training set and 4.48% on the test set.
[0235] Figure 14 Comparisons between DNN predictions and numerical simulations for the cold and hot channels f are presented. The relative errors of the DNN predictions compared to the simulation results are all within 6.5%. The maximum relative error for the cold channel f is 0.15% on the training set and 2.42% on the test set. The maximum relative error for the hot channel f is 0.23% on the training set and 6.07% on the test set.
[0236] The above results show that the substitution model established by the present invention using neural network can effectively and quickly predict the heat transfer per unit volume of AFPH, Nu and f of the hot and cold channels.
[0237] Analysis of multi-objective optimization design results
[0238] Multi-objective optimization design is carried out by combining the NSGA-II algorithm and deep neural networks. The NSGA-II algorithm is used to plan the optimal path for parameter combinations, while the trained deep neural network rapidly calculates the objective function. Through iterative calculations, a Pareto optimal solution set is ultimately generated. Based on this Pareto optimal solution set, the extreme points of a single objective can be determined, and the TOPSIS and LINMAP methods can be used to generate multi-objective decision points.
[0239] The results of the first multi-objective optimization design are as follows Figure 15 The input parameters and performance indicators of the extreme points of each performance index and the TOPSIS and LINMAP decision points are shown in Table 6. For the single target extreme point, Nu ave The maximum point coincides with the Q / V maximum point. This is because within the parameter range of the present invention, under the condition of the same working fluid mass flow rate, the smaller the channel volume, the faster the working fluid flow rate, the higher the convective heat transfer intensity, the greater the Nusselt number and heat transfer, and the greater the heat transfer per unit volume. At this time, the heat transfer per unit volume can reach 488.38W / cm 3 The average Nusselt number of the cold and hot channels is 374.03, but the friction coefficient is as high as 0.00762, which is higher than f ave The minimum point is 22.31% higher. ave At the minimum point, the friction coefficient is the smallest, but at the same time Nu ave The Q / V ratio is also small, and the heat transfer per unit volume is only 30.82% of the maximum value of Q / V, and the average Nusselt number is only Nu ave The maximum point is 38.54%.
[0240] Table 6 shows that the mass flow rate at the extreme points of each performance indicator, as well as the TOPSIS and LINMAP decision points, is 5 g / s. The main differences lie in the AFPH fin chord length and longitudinal spacing. This is because, within the parameter range studied in this paper, a higher flow rate, a faster working fluid velocity, and a larger Reynolds number enhance heat transfer and reduce the friction coefficient. Therefore, from the perspective of heat transfer per unit volume, average Nusselt number, and average friction factor, a higher mass flow rate is better.
[0241] Table 6 Input parameters and performance indicators of the first optimization design decision point
[0242]
[0243] Numerical simulations were performed for the input parameters of the three extreme points and the TOPSIS and LINMAP decision points. The relative errors between the DNN prediction results and the CFD simulation results are shown in Table 6. Among the performance indicators, the maximum error was 25.35% for the extreme points, 6.55% for the TOPSIS decision points, and 3.47% for the LINMAP decision points. It can be seen that the DNN prediction results for the performance indicators at the extreme points have large errors. This is mainly because when a single performance indicator reaches an extreme value, the input parameter is closer to the boundary of the parameter range. There are fewer calculation points at the boundary, resulting in insufficient prediction ability of the trained DNN in areas near the boundary.
[0244] Based on the primary optimization design, the target area is determined, and the value range of the input parameters is narrowed according to the target area. The parameter combination is given through the experimental design method and CFD calculations are supplemented. The CFD calculation results are used to iteratively update the DNN replacement model, and further secondary optimization design is carried out.
[0245] In order to improve the accuracy of the DNN replacement model and further improve the accuracy of the multi-objective optimization design results, the following Figure 15 The target region for the secondary iterative optimization design is shown in the cube. The target region for single-objective optimization was determined with the extreme point as the center and ±25% of the range of each performance indicator on the Pareto front as the standard. The target region for multi-objective optimization was determined with the LINMAP decision point as the center and ±10% of the range of each performance indicator on the Pareto front as the standard. The range of input parameters was determined based on the input parameters at each point within the target region. Using the central composite design method, parameter combinations were supplemented and numerical simulations were performed to enrich the simulation dataset within the target region. Furthermore, the established deep neural network surrogate model was trained for a second time. Finally, the updated deep neural network was combined with the NSGA-II algorithm to carry out secondary multi-objective optimization design.
[0246] The second optimization result is as follows Figure 16 As shown in the figure, the CFD calculation data of the parameter combination near the extreme point is added, the extreme value prediction of each performance indicator is more accurate, and the position of the positive and negative ideal points has also changed significantly. The LIMMAP and TOPSIS decision processes are determined according to the distance between the positive and negative ideal points. Therefore, the position of the LINMAP and TOPSIS decision points has also changed significantly and is located in the middle of the Pareto frontier.
[0247] Detailed data for extreme points and decision points are given in Table 7. The maximum error between the performance indicators of extreme points and LINMAP and TOPSIS decision points and the CFD verification results is only 0.45%, which is a significant improvement compared to the accuracy of the first optimization design.
[0248] Table 7 Input parameters and performance indicators of the second optimization design decision point
[0249]
[0250] According to the optimization results, from the perspective of the three performance indicators of heat transfer per unit volume, average Nusselt number and average friction coefficient, large flow and small fins have better performance; if you want to increase the heat transfer per unit volume, you can reduce the lateral spacing of the fins, but if you want to reduce the friction coefficient, you should increase the lateral spacing of the fins. On the Pareto frontier of the secondary optimization results, the LINMAP and TOPSIS decision points are relatively close. The TOPSIS decision point is a decision solution on the Pareto frontier that is close to the positive ideal point and far away from the negative ideal point. Q / V is 74.30% of its maximum value, Nu ave 73.15% of its maximum value, f ave The LINMAP decision point is a decision solution on the Pareto frontier that is close to the positive ideal point and far away from the negative ideal point. Q / V is 75.74% of its maximum value. Nu ave 73.03% of its maximum value, f ave 115.34% of its minimum value.
[0251] 2. Conclusion
[0252] This paper investigates the thermal-hydraulic characteristics and multi-objective optimization design of AFPH, proposing a step-by-step, iterative multi-objective optimization strategy. This strategy utilizes a combination of parametric geometric models, experimental design, deep neural networks, and genetic optimization algorithms to achieve optimal design. This step-by-step optimization approach allows for the identification of key optimization variables and a gradual narrowing of the optimization range, enabling the realization of high-precision, multi-objective optimization design across multiple parameters and a wide range at low cost. The main conclusions are as follows:
[0253] (1) The orthogonal experimental results show that L, δ and The influence on the surface heat transfer coefficient is significant. The total pressure drop of the cold and hot channels is significantly affected, L, δ and It has a significant impact on the heat transfer per unit volume. Therefore, L, δ and The three parameters are key parameters for the optimized design of AFPH.
[0254] (2) The constructed AFPH thermal hydraulic characteristics prediction agent model can be used to predict the thermal hydraulic characteristics of the AFPH based on L, δ and The heat transfer per unit volume, average Nusselt number and average friction coefficient are effectively predicted, and the relative error of each parameter on the test set is within 6.5%.
[0255] (3) The results of the multi-objective optimization design show that, from the perspective of the three performance indicators of heat transfer per unit volume, average Nusselt number and average friction coefficient, the performance of large flow and small fins is better; if the heat transfer per unit volume is to be increased, the lateral spacing of the fins can be reduced, and if the friction coefficient is to be reduced, the lateral spacing of the fins should be increased.
[0256] (4) The accuracy of the optimization design results can be greatly improved through step-by-step iterative optimization. Compared with the CFD numerical calculation results, the maximum relative error of the results obtained from the first optimization design is 25.3%, while the maximum relative error of the results obtained from the second optimization design is only 0.45%.
[0257] Symbol Table
[0258]
[0259] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.
[0260] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.
Claims
1. A step-by-step iterative multi-objective optimization design method for a heat exchanger, characterized in that: The following steps are involved: Step 1, establish a substitution model; First, a fully parameterized geometric model of a wing-shaped fin channel (PCHE) was established. Then, the orthogonal design method was used to combine different parameter values and perform numerical simulations. The sensitivity of the heat transfer rate per unit volume, heat transfer characteristics, and flow characteristics to each parameter was analyzed based on the obtained results, and the optimized parameters were screened. Next, the central composite design method was used to expand the parameter range and perform numerical simulations to form a representative dataset of optimized variables and thermal-hydraulic characteristic indicators. The resulting dataset was then used to train a neural network to achieve rapid prediction of the thermal-hydraulic characteristic parameters of the AFPH. Step 2: Carry out the first optimization design; use the neural network and genetic optimization algorithm to carry out the first multi-objective optimization design, and determine the secondary optimization area based on the decision points and extreme points; Step 3: Carry out the second optimization design; according to the secondary optimization area, determine the parameter range, supplement the numerical simulation calculation, and iteratively update the deep neural network model, and finally carry out the second optimization design.
2. The step-by-step iterative multi-objective optimization design method for a heat exchanger according to claim 1, characterized in that: The alternative model: (1) Parametric geometric model; A pair of hot and cold channels were selected for modeling and simulation. The airfoil fin chord length is L, and the cross-sectional shape is modeled based on the NACA00ZZ series symmetrical airfoil, where ZZ represents the percentage of the maximum half-thickness to the chord length. The total channel length is 130 mm, the number of fins per row is N, the longitudinal spacing of the fins is Lx, and the transverse spacing is δ. The fins run through the channel, and the hot and cold channels are arranged in a countercurrent manner. (2) Flow and heat transfer control equations and boundary conditions; (3) Model validation; (4) Performance indicators; (5) Deep neural networks; (6) Multi-objective genetic optimization algorithm.
3. The step-by-step iterative multi-objective optimization design method for a heat exchanger according to claim 2, characterized in that: The flow heat transfer control equations and boundary conditions are: The flow and heat transfer process of S-CO2 in PCHE satisfies the conservation of mass, momentum, and energy, considering the following assumptions: (1) gravity is ignored; (2) heat loss is ignored; (3) the flow process is a three-dimensional steady-state turbulent flow; the governing equations of the flow and heat transfer process are shown in Equations (1)-(3); Continuity equation: Momentum equation: Energy equation: Where ρ is density (kg / m3), u is velocity (m / s), μ is dynamic viscosity (Pa·s), μt is turbulent viscosity (Pa·s), p is pressure (Pa), cp is specific heat capacity at constant pressure (J / (kg·K)), T is temperature (K), λ is thermal conductivity (W / (m·K)), and Φ is turbulent energy dissipation rate (W / m3). The governing equation for heat conduction in a metal matrix is a three-dimensional steady-state heat conduction differential equation without an internal heat source, as shown in Equation (4); The above control equations are solved using Fluent software, with the convergence residual of the energy equation set to 10-9 and the convergence residual of other equations set to 10-6.
4. The step-by-step iterative multi-objective optimization design method for a heat exchanger according to claim 2, characterized in that: The model validation: 1) Grid independence verification Fluent-meshing was used to mesh the established geometric model with a polyhedra type mesh. This type of mesh has good adaptability to complex geometric structures and can reduce the total number of meshes compared to tetrahedral meshes, resulting in higher computational efficiency. Boundary layer meshes were created in the fluid region near the fluid-solid interface, with the first layer height set to 0.001 mm. 2) Verification of numerical simulation accuracy The calculation results of the convective heat transfer coefficient h, which characterizes the heat transfer characteristics, and the Fanning friction coefficient f, which characterizes the flow characteristics, were compared.
5. The step-by-step iterative multi-objective optimization design method for a heat exchanger according to claim 2, characterized in that: The performance indicators are: The airfoil fins are dispersed, so the interface shape of the S-CO2 flow area changes with position. The hydraulic diameter Dh is defined based on the fluid area volume V and the fluid-solid interface area S: According to the inlet flow The physical properties of S-CO2 can be determined by the inlet temperature, and then the Reynolds number Rein, which characterizes the inlet flow state, can be obtained: Re symbols are italicized in the text, but in normal form in the equations; generally, symbols are italicized and units are normal form; Where Ain is the channel inlet cross-sectional area (m2), μ is the dynamic viscosity (Pa·s); The local Nusselt number NuL is defined as follows: Where λ is the thermal conductivity, hl is the local convection heat transfer coefficient, and hl is calculated using formula (8); Tw,L and Tb,L are the average boundary heat flux density (W / m2), wall temperature (K), and cross-sectional average temperature (K), respectively. A series of cross sections parallel to the YZ plane are taken along the flow direction. Local parameters are calculated at each cross section, and the global parameter is the average of all local parameters. Local parameters are denoted by the subscript L. The Fanning friction coefficient f is defined as follows: Where pin, pout, uin, ρin, and Lf represent the inlet pressure (Pa), outlet pressure (Pa), inlet cross-sectional velocity (m / s), inlet cross-sectional density (kg / m3), and channel length (m), respectively; The heat transfer per unit volume qv is defined as follows: Where Q is the heat transfer capacity (W) and V is the volume of the heat exchanger (m3).
6. The step-by-step iterative multi-objective optimization design method for a heat exchanger according to claim 2, characterized in that: The multi-objective genetic optimization algorithm: The multi-objective optimization problem is shown in formula (11); 7. A heat exchanger step-by-step iterative multi-objective optimization design system for implementing the heat exchanger step-by-step iterative multi-objective optimization design method according to any one of claims 1 to 6, characterized in that: The heat exchanger step-by-step iterative multi-objective optimization design system includes: The surrogate model building module is used to establish a fully parameterized geometric model of the airfoil-shaped fin channel PCHE. Then, the orthogonal design method is used to combine different values of each parameter and conduct numerical simulation calculations. Based on the obtained results, the sensitivity of the heat transfer rate per unit volume, heat transfer characteristics, and flow characteristics to each parameter is analyzed to screen the optimized parameters. Next, the central composite design method is used to expand the parameter value range and conduct numerical simulation calculations to form a representative data set of optimization variables and thermal-hydraulic characteristic indicators. The obtained data set is then used to train a neural network to achieve rapid prediction of the thermal-hydraulic characteristic parameters of the AFPH. The first optimization design module is used to carry out the first optimization design; it uses a combination of neural network and genetic optimization algorithms to carry out the first multi-objective optimization design, and determines the secondary optimization area based on the decision points and extreme points; The second optimization design module is used to carry out the second optimization design; according to the secondary optimization area, the parameter range is determined, the numerical simulation calculation is supplemented, and the deep neural network model is iteratively updated, and finally the second optimization design is carried out.
8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the step-by-step iterative multi-objective optimization design method for a heat exchanger as described in any one of claims 1 to 6.
9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the step-by-step iterative multi-objective optimization design method for a heat exchanger according to any one of claims 1 to 6.
10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the heat exchanger step-by-step iterative multi-objective optimization design system as described in claim 7.
Citation Information
Cited By
Finned tube heat exchanger multi-parameter collaborative optimization system and method
CN121351718A
Multi-objective optimization design method and system for liquid helium low-temperature subcooler
CN121936318A
