Design method of an intelligent tesla valve and its proton exchange membrane fuel cell
Patent Information
- Application Number
- TW114111622
- Authority / Receiving Office
- TW · TW
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-03-26
- Publication Date
- 2026-10-01
- Estimated Expiration
- 2045-03-25
AI Technical Summary
Proton exchange membrane fuel cells (PEMFCs) face performance, efficiency, and durability issues due to temperature fluctuations and uneven gas distribution caused by suboptimal flow channel geometries in bipolar plates, leading to potential membrane rupture and inefficiencies.
An intelligent Tesla valve design method using variational autoencoders and machine learning to optimize flow channel geometry, incorporating finite element analysis and reinforcement learning to enhance temperature control and gas uniformity, reducing pressure loss and improving thermal management.
The optimized flow channel design stabilizes temperature below 80°C, enhances gas distribution uniformity, and increases power density, resulting in improved PEMFC performance and durability.
Smart Images
Figure TWG2TA001077896_001 
Figure TWG2TA001077896_002 
Figure TWG2TA001077896_003
Abstract
Description
Technical Field
[0001] A design method for a Tesla valve, particularly a design method for the gas flow channel of an intelligent Tesla valve, and its application in a proton exchange membrane fuel cell. Prior Technology
[0002] Proton exchange membrane fuel cells (PEMFCs) primarily use pure hydrogen and pure oxygen as fuel, but for safety reasons, some water is added to the fuel to prevent explosions. The reaction process involves the flow channels in the bipolar plates transporting the fuel to the membrane electrode assembly, where an electrochemical reaction generates electrical and thermal energy. However, PEMFCs are low-temperature batteries, with an operating range of 60 to 80 degrees Celsius. Exceeding this range can lead to membrane rupture and other problems.
[0003] The electrochemical reaction temperature is closely related to the flow channel geometry and characteristics in the bipolar plate. Therefore, different flow channel geometries will affect the gas uniformity inside the battery, resulting in different temperature changes, which in turn change the reaction rate between fuels, leading to a decrease in the performance, efficiency and durability of the proton exchange membrane fuel cell.
[0004] Besides factors such as temperature, pressure, humidity, gas flow rate, and water management affecting fuel cell performance during operation, the gas flow channel design within the bipolar plates is crucial. Different flow channel designs influence gas flow within the cell. Gas flow resistance is affected by the flow channel design; poor design can lead to uneven gas distribution within the flow channel, resulting in concentration differences. This causes high-concentration areas to accelerate reactions due to high temperatures, releasing excessive heat and potentially leading to membrane rupture. Furthermore, pressure loss at the flow channel inlet and outlet is also affected by flow resistance. Therefore, resistance increases when the fluid passes through bends or areas with large angles. Simultaneously, higher fluid velocity results in lower pressure, while decreased velocity leads to increased pressure, creating a pressure difference that alters the gas velocity distribution within the flow channel. This results in slower gas velocity in high-pressure areas and vice versa in low-pressure areas. Therefore, to improve battery efficiency and ensure stability, PEMFC flow channel geometry optimization, appropriate temperature control, and battery thermal management are necessary.
[0005] The process of this invention can quickly generate high-performance flow channel designs, and optimize and verify them by combining real simulation results, ultimately providing design solutions that can be practically applied to fuel cells or other fluid systems. Summary of the Invention
[0006] To optimize the flow channel geometry of a proton exchange membrane fuel cell (PEMFC) for appropriate temperature control and thermal management, thereby improving the overall performance, efficiency, and durability of the PEMFC, this invention provides a design method for an intelligent Tesla valve, comprising the following steps: Step 1: Provide a flow channel model of a Tesla valve in a primitive proton exchange membrane fuel cell; Step S2: Perform finite element analysis on the flow channel model to obtain a cell reaction result; wherein, the cell reaction result includes the temperature distribution, gas reaction rate, pressure loss distribution, pressure drop, gas flow rate uniformity and / or temperature distribution in the flow channel model; Step 3: Input the battery reaction result into a machine learning module for judgment and determine the area in the flow channel model that needs to be modified, and at the same time parameterize the area that needs to be modified; Step 4: Perform performance prediction or testing on the newly modified flow channel model to obtain the optimal solution for the flow channel design.
[0007] In step 1, a variational autoencoder is used to extract key design parameters from the simulation data of the flow channel model, and a new parameter vector is generated by modeling with the variational autoencoder.
[0008] Next, multiple candidate design schemes are generated through the variational autoencoder. These candidate schemes are initially screened according to a lightweight screening model to eliminate designs with unreasonable structures or performance prediction values that are obviously not up to standard.
[0009] In step 4, the newly modified flow channel model is used to quickly evaluate the performance of various designs that generate the flow channel model using a performance prediction module. Then, a reinforcement learning module is used to iteratively optimize the generated flow channel model design based on the evaluation results of the performance prediction module. Alternatively, step 4 can also involve performing finite element analysis on the newly modified flow channel model again to obtain the modified battery reaction results.
[0010] Preferably, the aforementioned steps further include step 5 verifying the optimized flow channel model design and / or step 6 repeating steps 2-4 to fine-tune and re-optimize the design.
[0011] In step 1, the variational autoencoder is used to design and optimize the flow channel model structure.
[0012] The machine learning module determines at least whether the highest temperature of the flow channel model is greater than 80 °C, whether the reaction rate reaches a threshold, and / or whether the valve inlet pressure minus the outlet pressure of the pressure loss distribution is greater than 60%.
[0013] The machine learning module modifies the flow channel in the flow channel model by adding, reducing, lengthening, shortening, widening, shrinking, or arcing the flow channel in a specific area, changing the flow channel cross-section, increasing or decreasing the flow channel sidewall angle, and generating new branches.
[0014] The cross-sections of the different flow channels include square, rhomboid and / or circular.
[0015] The present invention also provides a proton exchange membrane fuel cell comprising one of the Tesla valves obtained by the aforementioned intelligent Tesla valve design method.
[0016] As can be seen from the above description, the present invention has the following beneficial effects and advantages:
[0017] This invention analyzes the internal structure of a proton exchange membrane fuel cell (PEMFC) using computational fluid dynamics (CFD). Beyond basic cell performance analysis, it includes the distribution of gas within the flow field and the physical differences in the flow channel structure. The overall performance of the PEMFC is evaluated through output performance, oxygen distribution, current density distribution, and pressure drop. The results show that the oxygen mass transfer in the novel Tesla valve flow channel is higher than that in the traditional flow channel. When switching from the traditional flow channel to the novel Tesla valve gas flow channel, the improved mass transfer results in a significant increase in the power density of the Tesla valve application.
[0018] Under the same operating voltage, changes in voltage drop often result in excessively low voltage drop in the Z-type and U-type flow channels of traditional Tesla valves, leading to poor battery efficiency. After improvement, the voltage drop is significantly increased. In contrast, excessively high voltage drop often occurs in the serpentine flow channel, resulting in severe energy loss. After modification, the voltage drop is significantly reduced.
[0019] Regarding temperature control, the battery temperature should be kept below 80°C to avoid problems such as decreased battery safety performance caused by high temperatures. This invention compares temperature changes under different flow channel geometries and maintains the operating temperature between 49°C and 52°C. Even when the power density reaches its peak, the battery temperature rise is not too high and poses a danger, demonstrating that the Tesla valve gas flow channel designed in this invention can indeed stabilize the battery structure and improve efficiency. Simple Explanation of the Diagram
[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are merely some examples or embodiments of the present invention and are not intended to absolutely limit the technical scope of the present invention. Unless obvious from the context or otherwise stated, the same reference numerals in the drawings represent the same structures or operations. Wherein: Figure 1 is a flowchart of the steps of a preferred embodiment of the design method of the intelligent Tesla valve of the present invention. Figure 2 is a flowchart of step 4 of a preferred embodiment of the design method of the intelligent Tesla valve of the present invention. Figures 3A to 3C are schematic diagrams of the temperature distribution, gas reaction rate, and / or pressure loss distribution of the battery reaction results before modification. Figures 4A to 4C are schematic diagrams of the temperature distribution, gas reaction rate, and / or pressure loss distribution of the modified battery reaction results. Figure 5 shows a schematic diagram of the new serpentine flow channel design. Implementation
[0021] The present invention will be described in detail below with reference to several preferred embodiments. The accompanying drawings are merely some exemplary representations or embodiments of the present invention. For those skilled in the art, the present invention can be applied to other similar situations based on these drawings without any further effort.
[0022] The terms “system,” “apparatus,” “unit,” and / or “module” used in this invention are to distinguish different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions. As illustrated in this invention, unless the context clearly indicates otherwise, words such as “a,” “an,” “an,” and / or “the” do not specifically refer to the singular and may also include the plural. Generally, the terms “comprising” and “including” only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list; the method or apparatus may also include other steps or elements.
[0023] This invention uses flowcharts to illustrate the operations performed by the system according to embodiments of the invention. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.
[0024] Please refer to Figures 1 and 2, which are flowcharts of the preferred embodiment of the design method of the intelligent Tesla valve of the present invention. The steps include:
[0025] Step S1: Provide a flow channel model 10 of a Tesla valve in a primitive proton exchange membrane fuel cell; preferably, the parameters of the flow channel model 10 include at least the original flow channel width, height, bending radius, and branch flow channel spacing, forming an initial design template; for example, in this embodiment, a variational autoencoder (VAE) can be used to extract key design parameters from the simulation data of the flow channel model 10, and new parameter vectors can be generated through variational autoencoder modeling. These parameter vectors represent different candidate design schemes. Preferably, multiple candidate design schemes are generated through the VAE module. These candidate schemes are initially screened according to a lightweight screening model to eliminate designs with unreasonable structures or performance prediction values that are obviously unsatisfactory, reducing the computational burden of subsequent performance evaluation;
[0026] Step S2: Perform finite element analysis on the flow channel model 10 to obtain a cell reaction result; wherein, the cell reaction result includes the temperature distribution, gas reaction rate, and / or pressure loss distribution in the flow channel model 10;
[0027] Step S3: Input the battery reaction result into a machine learning module for judgment and determine the area in the flow channel model 10 that needs to be modified, and parameterize the area to be modified at the same time. This step mainly involves continuously optimizing the design scheme through the enhanced machine learning module, including but not limited to performance indicators such as temperature distribution, gas reaction rate, pressure loss distribution, pressure drop, gas flow uniformity and / or temperature distribution, to perform reward calculations and strategy updates, and finally generate a high-efficiency flow channel structure recommendation version with low pressure drop, uniform gas flow and stable temperature distribution. For example, the machine learning module at least determines whether the highest temperature of the flow channel model 10 in the battery reaction result is greater than 80 °C, whether the reaction rate reaches a threshold and / or whether the valve inlet pressure minus the outlet pressure of the pressure loss distribution is greater than 60%, as shown in Figure 2C.
[0028] Step S4: Perform performance prediction or testing on the newly modified flow channel model 10 to obtain the optimal solution; Step S4 includes the following two different embodiments:
[0029] Step S4-1: Quickly evaluate the performance of various designs that generate the newly modified flow channel model 10 using a performance prediction module, and then use a reinforcement learning module to iteratively optimize the generated flow channel model 10 design based on the evaluation results of the performance prediction module; or
[0030] Step S4-2: Perform finite element analysis on the newly modified flow channel model 10 again to obtain the modified battery reaction results, and optionally repeat the aforementioned steps S2~S4 to obtain the optimal solution; and
[0031] Step S5 (optional): Verify the optimized flow channel model 10 design to ensure the feasibility and stability of the design under real physical conditions. Preferably, detailed simulation and verification can be performed using fluid dynamics tools to ensure the feasibility and stability of the design under real physical conditions;
[0032] Step S6 (optional): Design fine-tuning, repeat the aforementioned steps S2~S4 to perform design fine-tuning and re-optimization.
[0033] Specifically, in step S1, the preferred embodiment uses a variational autoencoder to perform the following structural design and optimization steps for the flow channel model 10:
[0034] An encoder is used to encode a parameter vector into a latent space vector using a multilayer perceptron (MLP). This process can automatically learn the implicit nonlinear characteristics in the design, ensuring the diversity and rationality of the generated design schemes.
[0035] A decoder uses an MLP to decode the latent space vector back into a parameter vector; this process reconstructs the complete design parameters through hierarchical mapping and ensures that the generated design meets structural constraints.
[0036] The loss function of VAE is calculated, which includes two parts: reconstruction loss and KL divergence loss. The reconstruction loss uses mean squared error (MSE) to measure the reconstruction accuracy, while the KL divergence loss is used to regularize the latent space to make it close to the standard normal distribution, thereby improving the stability and generalization ability of the subsequent generative model of this machine learning module.
[0037] Next, in step S4-1, the performance prediction module is used to quickly evaluate the performance of the generated flow channel model 10 design. This performance prediction module is built based on the machine learning module. Specifically, the performance prediction module includes a deep neural network prediction module that uses a multilayer perceptron (MLP) input as a parameter vector and outputs three performance indicators: pressure drop, gas reaction rate, and temperature distribution. A lightweight screening model is then introduced before the main evaluation model to eliminate unreasonable designs based on simple physical rules, reducing the computational load of the main model. Finally, the performance of various generated flow channel model 10 designs is evaluated using MSE as the loss function. Preferably, for more accurate evaluation, fluid dynamics (CFD) tools can be used directly for simulation to output the flow channel design performance indicators, ensuring the feasibility of the optimized design in a real physical environment.
[0038] In step S4-1, the reinforcement learning module iteratively optimizes the generated flow channel design based on the evaluation results of the performance prediction module. Specifically, the reinforcement learning module includes a policy network that takes a multilayer perceptron (MLP) as the input to the VAE's latent vector and outputs a new latent vector. It adjusts the latent vector according to a reward function to generate a better design scheme. The reward function is calculated based on the evaluation results, and its specific formula is as follows (1):
[0039] …Equation (1); where:
[0040] R: Reward value, the objective used in reinforcement learning for policy updates;
[0041] Pressure drop, measured in Pa. A smaller pressure drop is better, hence the negative sign.
[0042] Gas flow velocity uniformity index, range The closer to 1, the better;
[0043] Standard deviation of temperature distribution, in K. The smaller the value, the more uniform the temperature, hence the negative sign.
[0044] Weight coefficients are adjusted dynamically according to the optimization objective and at different training stages. The value of . For example, in the early stages of optimization, the main focus is on voltage drop, so increase . The value gradually increases in the later stages. The value of .
[0045] This reinforcement learning module also uses reinforcement learning algorithms and adopts a hybrid optimization strategy, combining deep deterministic policy gradient (DDPG), proximal policy optimization (PPO), and genetic algorithm (GA) for policy updates, thereby improving global search capability and convergence speed.
[0046] In step S2 and step S4, a preferred embodiment of the temperature distribution in the flow channel model 10 of the battery reaction result is shown in Figure 3A. The dark part of the flow channel model 10 of the Tesla valve shows that the temperature of some areas is greater than 80 °C. Therefore, the machine learning module in step S3 modifies the dark part of the area to reduce the temperature during the reaction.
[0047] Specifically, the gas reaction rate is shown in Figure 3B. This preferred embodiment is analyzed using the reaction rate diagram of the flow channel model 10. As shown in Figure 3B, the dark area is the region with poor flow rate. Therefore, if the total area is subtracted from the dark area and the value of the dark area is at least 50% (total area - dark area / dark area > / = 50%), the machine learning module will modify the dark area to improve the reaction rate.
[0048] The machine learning module modifies, but is not limited to, the areas of the flow channel in the flow channel model 10 that require modification, including but not limited to adding, reducing, lengthening, shortening, widening, shrinking, or arcing the flow channel in specific areas; changing the flow channel cross-section (including square, rhomboid, or circular); increasing or decreasing the flow channel sidewall angle; generating new branches; adjusting the flow channel bending radius and angle; adjusting the spacing and number of branch channels; or determining the inlet and outlet positions. The machine learning module parameterizes the areas to be modified in this embodiment. The novel serpentine design shown in Figure 5 mainly includes, but is not limited to, reducing, arcing, and generating new branches. Next, as shown in Figures 4A, 4B, and 4C, finite element analysis is performed on the modified flow channel model 10 to obtain the modified battery reaction results. After at least the temperature distribution, gas reaction rate, and / or pressure loss distribution are improved, the optimized battery reaction results are saved.
[0049] Comparing Figures 3A-3C and 4A-4C, it can be confirmed that the intelligent Tesla valve design method provided by the present invention can indeed optimize the battery reaction results, improve drainage by increasing pressure drop, achieve more uniform gas distribution, reduce low concentration areas and energy loss, and improve and stabilize the performance of the proton exchange membrane fuel cell.
[0050] Furthermore, the present invention also provides a proton exchange membrane fuel cell comprising a Tesla valve obtained by the aforementioned design method.
[0051] Furthermore, unless explicitly stated herein, the use of elements, sequences, numbers, letters, or other names in this invention is not intended to limit the order of the processes and methods of this invention. Although various examples have been discussed above to illustrate some embodiments of the invention that are currently considered useful, it should be understood that such details are for illustrative purposes only, and the invention is not limited to the disclosed embodiments, but also includes equivalent related variations and modifications.
[0052] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of the invention are approximate values, in specific embodiments, such values are set as precisely as feasible.
[0053] 10: Flow channel model S1~S5: Steps
Claims
1. A design method for an intelligent Tesla valve, comprising the following steps: Step S1: Providing a flow channel model of a Tesla valve for a primitive proton exchange membrane fuel cell; extracting key design parameters from the simulation data of the flow channel model using a variational autoencoder, and generating a new parameter vector through variational autoencoder modeling; Step S2: Performing finite element analysis on the flow channel model to obtain a cell reaction result; wherein, The battery reaction results include the temperature distribution, gas reaction rate, pressure loss distribution, pressure drop, gas flow rate uniformity, and / or temperature distribution in the flow channel model; Step 3: Input the battery reaction results into a machine learning module for judgment and determine the area in the flow channel model that needs to be modified, and parameterize the area that needs to be modified; Step 4: Perform performance prediction or detection on the newly modified flow channel model to obtain the optimal solution for the flow channel design.
2. The design method of the intelligent Tesla valve as described in claim 1, wherein: multiple candidate design schemes are then generated by the variational autoencoder, and these candidate schemes are initially screened according to a lightweight screening model to eliminate designs with unreasonable structures or performance prediction values that are obviously not up to standard.
3. The design method of the intelligent Tesla valve as described in claim 1, wherein: step 4 is to quickly evaluate the performance of various designs that generate the newly modified flow channel model using a performance prediction module, and then use a reinforcement learning module to iteratively optimize the generated flow channel model design based on the evaluation results of the performance prediction module.
4. The design method of the intelligent Tesla valve as described in claim 1, wherein: step 4 involves performing finite element analysis on the newly modified flow channel model again to obtain the modified battery reaction result.
5. The design method of the intelligent Tesla valve as described in claim 1, further comprising step 5 verifying the optimized flow channel model design and / or step 6 repeating steps 2 to 4 to fine-tune and re-optimize the design.
6. The design method of the intelligent Tesla valve as described in claim 2, wherein: in step 1, the variational autoencoder is used to perform the following flow channel model structure design and optimization steps: using an encoder and a multilayer perceptron to encode the parameter vector into a latent space vector, the multilayer perceptron automatically learns the implicit nonlinear features in the design to ensure the diversity and rationality of the generated design scheme; using a decoder to decode the latent space vector back into a parameter vector using the multilayer perceptron, and then reconstructing the complete design parameters through hierarchical mapping, and ensuring that the generated design meets the structural constraints; and calculating the loss function of a variational autoencoder, including reconstruction loss and KL divergence loss, thereby improving the stability and generalization ability of the generated model of the subsequent machine learning module.
7. The design method of the intelligent Tesla valve as described in claim 1, wherein: the machine learning module at least determines whether the highest temperature of the flow channel model is greater than 80°C, whether the reaction rate reaches a threshold, and / or whether the valve inlet pressure minus the outlet pressure of the pressure loss distribution is greater than 60% in the battery reaction result.
8. The design method of the intelligent Tesla valve as described in claim 1, wherein: the machine learning module modifies the flow channel in the flow channel model by adding, reducing, lengthening, shortening, widening, shrinking, arcing, changing the flow channel cross-section, increasing or decreasing the flow channel sidewall angle, or generating new branches in the specific area.
9. A proton exchange membrane fuel cell comprising one of the Tesla valves obtained by the design method of the intelligent Tesla valve described in any one of claims 1 to 8.