A method for predicting water content in a cathode flow channel of a visualized cathode fuel cell device

Abstract

This invention discloses a method for predicting the water content of the cathode flow channel in a visualized cathode fuel cell device, belonging to the field of fuel cell visualization. The method includes an anode unit, a cathode unit, and an MEA (Mechanical and Electrical Assembly). The cathode unit has a transparent end plate with a hollow cavity. Circulating hot water is introduced to eliminate the temperature difference between the two sides of the end plate, preventing fogging of the viewing window. Images of the cathode flow channel with water content and images of the dry flow channel are acquired. A difference image is obtained through difference calculation, and then a binary image is obtained through adaptive binarization. The water coverage rate is calculated using a mask image as a quantitative representation of the water content. A convolutional neural network is constructed, and the model is trained using the water coverage rate as input. The Sobol sensitivity analysis method is used to optimize the hyperparameters of the learning rate, dropout rate, and batch size. The optimized CNN model is used to predict the water content in the cathode flow channel in real time, providing a basis for flood warning. This invention achieves non-invasive quantitative prediction of the water content in the cathode flow channel, improving prediction accuracy and robustness.

Description

Technical Field

[0001] This invention belongs to the field of fuel cell visualization, and particularly relates to a method for predicting the water content of the cathode flow channel in a visualized cathode fuel cell device. Background Technology

[0002] Proton exchange membrane fuel cells (PEMFCs) have broad application prospects in electric vehicles, stationary power sources, and other fields due to their advantages such as high energy conversion efficiency, low operating temperature, and fast start-up speed. However, water management within the battery directly affects its performance and lifespan, especially since excessive liquid water in the cathode channel can lead to "flooding," hindering the transport of reactant gases. To address this, researchers are using optical visualization technology to observe the distribution and transport behavior of water within the channel, providing a basis for channel design and water management strategies. Currently, some transparent fuel cell devices allow for direct observation of the liquid water within the channel through transparent endplates.

[0003] Although existing visualization devices can qualitatively observe the generation and movement of water, the following technical problems still exist: First, temperature differences easily occur on both sides of traditional transparent end plates, causing fogging of the viewing window and affecting the image acquisition quality; Second, most devices can only perform qualitative analysis and lack quantitative detection methods for water content; Third, existing methods cannot accurately predict the changing trend of water content in the flow channel based on real-time images, making it difficult to provide early warning of flooding risks; Fourth, there is a lack of prediction models and hyperparameter optimization methods for water content in the cathode flow channel, resulting in insufficient model prediction accuracy. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for predicting the water content of the cathode flow channel in a visualized cathode fuel cell device, comprising: Based on the water-containing and dry flow channel images of the cathode channel acquired from the transparent endplate of the visualized cathode fuel cell device, a difference image is obtained. Based on the difference image, a binarized image is obtained; based on the binarized image and the mask image, the water coverage is obtained; based on the water coverage, Construct a training dataset for a convolutional neural network; train the convolutional neural network based on the training dataset; The water content of the cathode channel is predicted based on the trained convolutional neural network.

[0005] Optionally, the visualized cathode fuel cell device includes an anode unit, a cathode unit, and an MEA; wherein, the cathode unit includes a cathode side pressure plate, a sealing ring, a cathode transparent end plate, and a metal perforated flow field plate; the anode unit includes an anode end plate, an anode current collector plate, and an anode graphite plate flow field plate.

[0006] Optionally, before obtaining the difference image based on the cathode flow channel water-containing image and dry flow channel image acquired by the cathode transparent end plate, the method further includes: cropping the region of interest based on the original acquired image to obtain a cropped image; and creating a mask image based on the cropped image, wherein the mask image is used to mask the metal flow channel portion.

[0007] Optionally, a difference image is obtained based on the water-bearing image and the dry flow channel image of the cathode transparent end plate. Specifically, this includes: calculating the difference between the pixel value of each pixel in the water-bearing image and the pixel value of the corresponding pixel in the dry flow channel image to obtain the difference image. Obtaining a binarized image based on the difference image specifically includes: performing adaptive binarization processing on the difference image to obtain the binarized image.

[0008] Optionally, the water coverage rate is obtained based on the binarized image and the mask image, specifically by dividing the number of white pixels in the binarized image by the number of white pixels in the mask image to obtain the water coverage rate.

[0009] Optionally, before constructing the training dataset for the convolutional neural network based on the water coverage rate, the method further includes: segmenting the original image into multiple sub-images; and obtaining the water coverage rate of each sub-image based on the binarized image and mask image corresponding to each sub-image.

[0010] Optionally, a convolutional neural network is trained based on the training dataset, specifically including: taking the water coverage rate as input, processing it through the input layer, convolutional layer, pooling layer, fully connected layer and output layer of the convolutional neural network, and training the model using the Adam optimizer.

[0011] Optionally, before training the convolutional neural network, the process includes: optimizing the hyperparameters of the convolutional neural network based on the Sobol sensitivity analysis method; and determining the optimization order of the hyperparameters as learning rate, dropout rate, and batch size based on the optimization results.

[0012] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0013] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0014] Compared with the prior art, the present invention has the following advantages and technical effects: This invention provides a method for predicting the water content of the cathode flow channel in a visualized cathode fuel cell device. Circulating hot water is introduced into the hollow cavity of the transparent endplate of the cathode, eliminating the temperature difference between the two sides of the endplate, effectively preventing fogging of the visualization window, and ensuring the clarity of image acquisition. Based on the difference calculation and binarization of the acquired water-content image and dry image, and combined with a mask image to obtain the water coverage rate, a quantitative characterization of the water content in the cathode flow channel is achieved. Furthermore, a convolutional neural network is constructed, and the Sobol method is used to optimize hyperparameters such as learning rate, dropout rate, and batch size, significantly improving the accuracy and robustness of water content prediction. Finally, the optimized model is used to predict the water content in the cathode flow channel in real time, providing early warning before flooding occurs, thereby avoiding a sudden drop in battery performance and extending the service life of the fuel cell. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is an exploded view of the structure of a transparent fuel cell according to an embodiment of the present invention; Figure 2 This is a schematic diagram comparing battery performance at RH=40%, 60%, 80%, and 100% according to an embodiment of the present invention. Figure 3 This is a schematic diagram comparing battery performance at temperatures of 40°C, 50°C, 60°C, 70°C, and 80°C, according to an embodiment of the present invention. Figure 4 This is an image processing flowchart of an embodiment of the present invention, wherein (a) is a dry channel image; (b) is a mask; (c) is a water-bearing channel image; and (d) is a binarized water-bearing image obtained after difference calculation. Figure 5 This is a schematic diagram illustrating how the original image and mask are divided into nine smaller parts according to an embodiment of the present invention. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0018] Example 1 This embodiment provides a method for predicting the water content of the cathode flow channel in a visualized cathode fuel cell device, including: Based on the water-containing and dry flow channel images of the cathode channel acquired from the transparent endplate of the visualized cathode fuel cell device, a difference image is obtained. Based on the difference image, a binarized image is obtained; based on the binarized image and the mask image, the water coverage is obtained; based on the water coverage, Construct a training dataset for a convolutional neural network; train the convolutional neural network based on the training dataset; The water content of the cathode channel is predicted based on the trained convolutional neural network.

[0019] Specifically, it includes: The visualized cathode fuel cell device includes an anode unit, a cathode unit, and a MEA. The anode unit consists of an anode end plate, an anode current collector plate, and an anode graphite plate flow field plate; the cathode unit consists of a cathode side pressure plate, a sealing ring, a cathode transparent end plate, and a metal perforated flow field plate.

[0020] In the anode unit, the inner side of the anode end plate is tightly attached to the anode current collector plate, and the other side of the anode current collector plate is tightly attached to the anode graphite flow field plate. The flow field in the anode graphite flow field plate is a serpentine channel with a channel width of 0.9 mm, a rib width of 0.9 mm, and a depth of 1 mm. Heating plates are attached to the anode end plate for heating the anode side.

[0021] In the cathode unit, the cathode-side pressure plate and the cathode transparent end plate are sealed by a sealing ring. A metal flow field plate is tightly attached to the other side of the cathode transparent end plate. This metal flow field plate is made of gold-plated copper plate with a serpentine flow channel, and heating elements are attached to the plate for heating the cathode side of the battery. A hollow chamber is provided inside the cathode transparent end plate, through which circulating hot water is introduced to eliminate temperature differences between the two sides of the transparent end plate and prevent fogging of the visualization window.

[0022] The MEA has a total thickness of 0.52 mm, a frame thickness of 0.06 mm, and an effective area of ​​5 cm × 5 cm.

[0023] The anode unit, MEA, cathode unit and their internal components are fixedly stacked and connected by fastening nuts and bolts.

[0024] The establishment of the cathode channel water content prediction model includes three parts: image acquisition and processing, construction of a CNN network, and optimization of CNN hyperparameters.

[0025] The image acquisition and processing workflow is as follows: First, the ROI region is cropped, retaining only the region of interest in the image. The cropped image size is 1320×1059. Then, a mask image is created to control which parts of the image are processed or ignored, thus masking the metal flow channel and retaining only the liquid water portion of the flow channel region. Next, an interpolated image is obtained: the difference between each pixel value of the water-containing image in the flow channel and the corresponding pixel value in a pre-captured dry flow channel image is calculated to obtain an image containing only liquid water. The calculation formula is as follows: ,in, The pixel value of each pixel in the obtained difference image; This represents the pixel value of each pixel in the water content image within the flow channel. This represents the pixel value of each point in the dry flow channel image.

[0026] Adaptive binarization was then used to obtain a binarized image, allowing for clear extraction of liquid water within the flow channel. Next, image segmentation was performed: the original 1320×1059 image was divided into nine identical sub-images, each measuring 440×353 pixels. Finally, the water content was calculated, defining water cover ratio (WCR) as a parameter, calculated using the following formula: ; The basic structure of the constructed CNN network includes an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. During model training, the Adam optimizer was used, with an i7-12700H CPU and a Geforce 3070Ti GPU. The learning rate was set to 0.001, and the ratio of training, validation, and test sets was 6:2:2.

[0027] CNN hyperparameter optimization is based on the Sobol sensitivity analysis method. The Sobol method analysis yields the following optimization order for the three hyperparameters: Learning rate, Dropout rate, and Batch size. The optimized model is then used to predict the water content within the cathode channel and to provide corresponding early warnings for potential flooding.

[0028] Example 2 This embodiment provides a method for predicting the water content of the cathode flow channel in a visualized cathode fuel cell device, including: Please see Figure 1The diagram shows an exploded view of the structure of a cathode fuel cell device. The device includes an anode unit, a cathode unit, and a MEA (Mechanical and Electrical Assembly). The anode unit consists of an anode end plate, an anode current collector plate, and an anode graphite flow field plate. Heating elements are attached to the anode end plate. The graphite flow field plate uses a serpentine flow channel with a width of 0.9 mm, a rib width of 0.9 mm, and a depth of 1 mm. The cathode unit consists of a cathode side pressure plate, a sealing ring, a cathode transparent end plate, and a metal perforated flow field plate. The metal flow field plate is made of gold-plated copper and features a serpentine flow channel with attached heating elements. The cathode transparent end plate has a hollow cavity inside, through which circulating hot water is introduced to eliminate temperature differences on both sides of the end plate and prevent fogging of the viewing window. The anode unit, MEA, and cathode unit are stacked and secured using fastening nuts and bolts.

[0029] Please see Figure 2 and Figure 3 In this embodiment, the device is tested under different relative humidity and different operating temperature conditions, and the prediction model is used to monitor the water content in the cathode channel in real time.

[0030] During the test, images of the cathode flow channel with water and the dry flow channel were acquired through the transparent endplate of the cathode. First, the ROI region was cropped, retaining a 1320×1059 pixel region of interest; a mask image was created to mask the metal flow channel portion; the difference between the water-containing and dry images was calculated to obtain a difference image containing only liquid water; adaptive binarization was used to obtain a binarized image; the original image was divided into nine 440×353 pixel sub-images, and the water coverage ratio (WCR) of each sub-image was calculated using the following formula: .

[0031] Please see Figure 4 . Figure 4 This is a schematic diagram of the image processing flow in this embodiment. Wherein, Figure 4 (a) is the acquired image of the drying channel; Figure 4 (b) A mask image created to mask the metal flow channel portion, retaining only the flow channel area; Figure 4 (c) is an image of the water-bearing channel obtained from the acquisition; Figure 4 (d) is the binarized water-bearing image obtained by calculating the difference between the water-bearing image and the dry image, followed by adaptive binarization processing. Through... Figure 4 (d) The distribution pattern of liquid water in the flow channel can be clearly extracted.

[0032] Please see Figure 5 . Figure 5This is a schematic diagram of image segmentation in this embodiment. The original image with a size of 1320×1059 pixels and its corresponding mask image are each segmented into 9 sub-images of the same size, each sub-image having a size of 440×353 pixels. After segmentation, the water coverage ratio (WCR) of each sub-image is calculated to analyze the water content distribution characteristics in different regions within the flow channel.

[0033] A convolutional neural network (CNN) was constructed, consisting of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The Adam optimizer was used, with an initial learning rate of 0.001. The training, validation, and test sets were partitioned in a 6:2:2 ratio, and training was accelerated using a GeForce 3070 Ti GPU. The CNN hyperparameters were optimized using the Sobol sensitivity analysis method, with the optimization order being: learning rate, dropout rate, and batch size. The optimized model was then used to predict the water content in the cathode channel and issue an early warning before flooding occurs.

[0034] like Figure 2 As shown, tests were conducted under different relative humidity conditions (10%, 60%, 80%, and 100%). Other boundary conditions were: counter-current flow, operating temperature 70℃, anode inlet pressure 150 kPa, anode excess coefficient 1.5, cathode inlet pressure 150 kPa, and cathode excess coefficient 2. During the tests, the prediction model was used to output real-time curves of the cathode channel water coverage at each relative humidity, and these curves were compared with the measured polarization curves to verify the model's prediction accuracy.

[0035] like Figure 3 As shown, tests were conducted under different operating temperatures (40℃, 50℃, 60℃, 70℃, and 80℃), with the following boundary conditions: counter-current flow, anode inlet pressure of 150 kPa, anode excess coefficient of 1.5, cathode inlet pressure of 150 kPa, cathode excess coefficient of 2, and relative humidity of anode and cathode gases of 80%. During the tests, the prediction model accurately identified the generation and discharge patterns of liquid water in the flow channel at different temperatures and automatically issued a flooding warning when the water coverage exceeded a preset threshold.

[0036] The test results above show that, based on the visualized cathode fuel cell device and the CNN-Sobol optimized prediction model, it is possible to achieve quantitative prediction and dynamic early warning of the water content in the cathode flow channel, providing a reliable technical means for fuel cell water management.

[0037] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0038] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0039] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the water content of the cathode flow channel in a visualized cathode fuel cell device, characterized in that, include: Based on the water-containing and dry flow channel images of the cathode channel acquired from the transparent endplate of the visualized cathode fuel cell device, a difference image is obtained. Based on the difference image, a binarized image is obtained; based on the binarized image and the mask image, the water coverage is obtained; based on the water coverage, Construct a training dataset for a convolutional neural network; train the convolutional neural network based on the training dataset; The water content of the cathode channel is predicted based on the trained convolutional neural network.

2. The method according to claim 1, characterized in that, The visualized cathode fuel cell device includes an anode unit, a cathode unit, and an MEA; wherein, the cathode unit includes a cathode side pressure plate, a sealing ring, a cathode transparent end plate, and a metal perforated flow field plate; the anode unit includes an anode end plate, an anode current collector plate, and an anode graphite plate flow field plate.

3. The method according to claim 1, characterized in that, Before obtaining the difference image based on the cathode flow channel water-containing image and dry flow channel image acquired from the cathode transparent end plate, the method further includes: cropping the region of interest based on the original acquired image to obtain a cropped image; and creating a mask image based on the cropped image, wherein the mask image is used to mask the metal flow channel portion.

4. The method according to claim 1, characterized in that, Based on the cathode flow channel water-bearing image and the dry flow channel image acquired by the cathode transparent end plate, a difference image is obtained, specifically including: calculating the difference between the pixel value of each pixel in the water-bearing image and the pixel value of the corresponding pixel in the dry flow channel image, and obtaining the difference image; Obtaining a binarized image based on the difference image specifically includes: performing adaptive binarization processing on the difference image to obtain the binarized image.

5. The method according to claim 1, characterized in that, The water coverage rate is obtained based on the binarized image and the mask image, specifically by dividing the number of white pixels in the binarized image by the number of white pixels in the mask image to obtain the water coverage rate.

6. The method according to claim 1, characterized in that, Before constructing the training dataset for the convolutional neural network based on the water coverage rate, the method further includes: segmenting the original image into multiple sub-images; and obtaining the water coverage rate of each sub-image based on the binarized image and mask image corresponding to each sub-image.

7. The method according to claim 1, characterized in that, Based on the training dataset, a convolutional neural network is trained, specifically including: taking the water coverage rate as input, processing it through the input layer, convolutional layer, pooling layer, fully connected layer and output layer of the convolutional neural network, and training the model using the Adam optimizer.

8. The method according to claim 1, characterized in that, Before training the convolutional neural network, the process also includes: optimizing the hyperparameters of the convolutional neural network based on the Sobol sensitivity analysis method; and determining the optimization order of the hyperparameters based on the optimization results as learning rate, dropout rate, and batch size.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-8.

Images