A system and method for analyzing capillary diameter based on image recognition
By employing techniques such as adaptive threshold segmentation and connected component analysis, combined with multi-angle light sources and dynamic exposure compensation, the problems of uneven illumination and misjudgment of pore contours in capillary diameter analysis were solved, achieving high-precision pore diameter measurement and distribution statistics.
Patent Information
- Application Number
- CN202610988154.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-03
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies for capillary diameter analysis suffer from problems such as uneven illumination, misjudgment of pore contours, false detection of overlapping pores, and deviation of pore diameter calculations from the true value, resulting in inaccurate measurement results and low efficiency.
By employing techniques such as adaptive threshold segmentation, connected component analysis, convex hull vertex calculation, and non-circularity calibration, combined with multi-angle ring light source and dynamic exposure compensation, high-precision analysis of capillaries can be achieved.
It improves the accuracy and efficiency of capillary diameter measurement, reduces interference from uneven illumination and complex surface textures, and enhances the reliability and engineering application value of pore size distribution statistics.
Smart Images

Figure CN123118729A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image recognition technology, specifically to a capillary diameter analysis system and method based on image recognition. Background Technology
[0002] Image recognition refers to the technology of using computers to process, analyze, and understand images in order to identify targets and objects of various patterns. It is a practical application of deep learning algorithms. At present, image recognition technology is generally divided into face recognition and product recognition. Face recognition is mainly used in security checks, identity verification, and mobile payments, while product recognition is mainly used in the process of commodity circulation, especially in unmanned retail fields such as unmanned shelves and smart retail cabinets.
[0003] The image recognition-based capillary diameter analysis system is a precision detection system integrating machine vision and intelligent algorithms. It aims to achieve automated and high-precision measurement of the microporous structure of material surfaces. This is one of the core components of the system, responsible for capturing microscopic morphological images of material surfaces. A high-resolution industrial camera, combined with a microscope lens, can acquire high-definition images of pores, ensuring accurate identification of micron- or even submicron-level capillary features during the analysis process. To improve the robustness of image feature extraction, the system is equipped with a multi-angle ring light source and an adaptive illumination module to ensure the contrast and detail integrity of capillary imaging.
[0004] Currently, due to the complexity and randomness of the pore distribution on material surfaces, traditional physical detection methods used for real-time capillary diameter analysis rely on wetting fluid filling and air pressure displacement processes, making non-contact in-situ measurement impossible. When the wetting fluid undergoes a physicochemical reaction with the material and the pores have interconnected structures, it can lead to deviations in the pore diameter distribution calculation results, making it difficult to guarantee the accuracy of the measurement results. At the same time, in image recognition-based analysis, conventional threshold segmentation algorithms have poor adaptability to samples with uneven lighting and complex surface textures, causing misjudgment and missed detection of pore contours, and failing to effectively distinguish between overlapping pores and impurities. Furthermore, in capillary pore statistics, due to the irregular shape of pores, such as ellipses and polygons, traditional circular fitting models cannot accurately describe the actual pore diameter characteristics, causing the diameter calculation results to deviate from the true value, further affecting the reliability and efficiency of pore diameter distribution statistics.
[0005] Therefore, an image recognition-based capillary diameter analysis system and method are proposed to solve the above problems. Summary of the Invention
[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a capillary diameter analysis system and method based on image recognition, which solves the problems mentioned in the background section.
[0007] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a capillary diameter analysis system and method based on image recognition, comprising the following steps: S1. Collect capillary surface image data, acquire the original RGB image using an industrial camera, and convert the image based on the grayscale conversion formula. Calculate pixel grayscale values to generate grayscale data; where... These represent the values for the red, green, and blue channels, respectively. S2. Perform adaptive threshold segmentation on the grayscale data, and use the Gaussian weighted mean method cv2.ADAPTIVE_THRESH_GAUSSIAN_C to calculate the weight distribution of pixels within the local window and generate threshold processing data; S3. Perform hole optimization processing on the threshold processing data. Remove holes with an area smaller than the set threshold area_threshold using the function morphology.remove_small_holes to generate optimized data. S4. Perform connected component analysis on the optimized data, and use the functions measure.label and measure.regionprops to identify the pore outline and segment individual pores to generate connected component data; S5. Extract contours from the connected region data by using the OpenCV function cv2.findContours to retrieve the set of contour points in the binary image and generate contour data. S6. Perform aperture analysis on the contour data, calculate the vertices of the contour convex hull using the function cv2.convexHull, measure the maximum distance between the vertices of the convex hull as the capillary aperture, and generate aperture data. S7. Convert the aperture pixel values into physical dimensions according to the calibration ratio, and then group them statistically. S8. Generate a pie chart of capillary diameter distribution based on grouped data to show the proportion of each pore size range.
[0008] Preferably, the weight allocation of the grayscale conversion formula in step S1 is designed based on the visual characteristics of the human eye, which gives the highest weight of 0.587 for green light, the second highest weight of 0.299 for red light, and the lowest weight of 0.114 for blue light.
[0009] Preferably, in the adaptive threshold segmentation of step S2, the weight distribution of the local window satisfies a Gaussian function.
[0010] in Standard deviation, These are pixel coordinates.
[0011] Preferably, the hole optimization process in step S3 satisfies: ,in This indicates that the 8-adjacency mode is used to determine the hole boundary.
[0012] Preferably, the connected component analysis in step S4 includes: Connected regions are labeled using measure.label according to the 8-adjacency rule; The area, perimeter, and centroid coordinates of each region can be extracted using measure.regionprops.
[0013] Preferably, the contour extraction in step S5 uses a combination of parameters:
[0014] in Only the outermost contour is retrieved; Compress redundant contour points.
[0015] Preferably, the pore size analysis in step S6 includes: Calculate the vertex coordinate set of the convex hull of the contour ; Traverse all vertex pairs Calculate Euclidean distance ,in , As vertex The x and y coordinates, , As vertex The horizontal and vertical coordinates; The maximum distance is taken as the capillary pore diameter.
[0016] Preferably, the statistical grouping in step S7 includes: Custom aperture range ; Statistics on the percentage of capillary pores in each interval ,in, For the first The percentage of capillaries within the pore size range. For the first Number of capillaries within a certain pore size range This represents the total number of capillaries across all pore size ranges.
[0017] Preferably, the pie chart generation in step S8 is implemented using matplotlib.pyplot.pie, and the pie chart block labels include aperture ranges and percentage percentages.
[0018] Preferred, including: Image acquisition module: Industrial camera and light source device, to acquire RGB images of the capillary surface; Preprocessing module: Performs grayscale conversion and adaptive threshold segmentation; Optimization module: Remove small holes and mark connected regions; Analysis module: Extract contour, calculate convex hull diameter and physical dimension conversion; Statistical output module: Groups and statistically analyzes aperture distribution and generates pie charts.
[0019] (III) Beneficial Effects Compared with the prior art, the present invention provides a capillary diameter analysis system and method based on image recognition, which has the following beneficial effects: 1. In this invention, by setting an adaptive imaging end, when performing real-time analysis of capillary diameter, a multi-angle ring light source and dynamic exposure compensation technology are integrated to effectively reduce imaging interference caused by uneven illumination and sample reflection, ensuring the clarity and continuity of the capillary edge contour. At the same time, a grayscale distribution model is established to intelligently adjust the light source intensity, ensuring that the pore features of samples of different materials can be displayed with high contrast, thereby improving the reliability of image data from the source and reducing the pore diameter measurement error caused by imaging quality.
[0020] 2. In this invention, by setting a morphological analysis end, a dual-modal verification mechanism is used to intelligently distinguish between real holes and impurity interference when extracting hole contours. On the one hand, the main axis size of irregular holes is captured by the topological relationship of convex hull vertices. On the other hand, the closure characteristics of holes are identified by curvature analysis, which effectively solves the problem of misjudgment of the contours of overlapping holes and irregular holes. At the same time, the segmentation threshold is dynamically optimized by combining a self-learning model, so that the system can adapt to samples with complex surface textures, reduce the hole false detection rate and improve the robustness of diameter calculation.
[0021] 3. In this invention, by setting up an intelligent integration terminal, a dynamic interval mapping rule for pore size is established when performing pore size group statistics. The interval division threshold is automatically adjusted based on the pore size distribution density to avoid statistical deviations caused by fixed intervals. At the same time, a non-circularity compensation algorithm is introduced to perform geometric feature calibration on the equivalent diameter of ellipses and polygons to accurately restore the true pore size distribution. In addition, by associating with the material porosity model, an industrial-grade analysis report is output, enabling the system to provide customized pore size evaluation standards for different materials and comprehensively improve the engineering application value of pore size distribution statistics. Attached Figure Description
[0022] Figure 1 This is a flowchart of a capillary diameter analysis method based on image recognition according to the present invention; Figure 2 This is a schematic diagram of the capillary diameter analysis system based on image recognition according to the present invention. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0024] Please see Figure 1 This image recognition-based capillary diameter analysis system and method includes the following steps: S1. Collect capillary surface image data, acquire the original RGB image using an industrial camera, and convert the image based on the grayscale conversion formula. Calculate pixel grayscale values to generate grayscale data; where... These represent the values for the red, green, and blue channels, respectively. S2. Perform adaptive threshold segmentation on the grayscale data. Use the Gaussian weighted mean method cv2.ADAPTIVE_THRESH_GAUSSIAN_C to calculate the weight distribution of pixels within the local window and generate threshold processing data. S3. Perform hole optimization processing on the threshold-processed data. Use the function morphology.remove_small_holes to remove holes with an area smaller than the set threshold area_threshold to generate optimized data. S4. Perform connected component analysis on the optimized data. Use the functions measure.label and measure.regionprops to identify the pore outline and segment individual pores to generate connected component data. S5. Extract contours from connected region data. Use the OpenCV function cv2.findContours to retrieve contour point sets in the binary image and generate contour data. S6. Perform aperture analysis on the contour data. Calculate the vertices of the contour convex hull using the function cv2.convexHull, measure the maximum distance between the vertices of the convex hull as the capillary aperture, and generate aperture data. S7. Convert the aperture pixel values into physical dimensions according to the calibration ratio, and then group them statistically. S8. Generate a pie chart of capillary diameter distribution based on grouped data to show the proportion of each pore size range; The weight allocation of the grayscale conversion formula in step S1 is designed based on the visual characteristics of the human eye, which gives the highest weight of 0.587 to green light, the second highest weight of 0.299 to red light, and the lowest weight of 0.114 to blue light. In the adaptive threshold segmentation of step S2, the weight distribution of the local window satisfies a Gaussian function.
[0025] in Standard deviation, These are pixel coordinates; The hole optimization process in step S3 satisfies: ,in This indicates that the 8-adjacency mode is used to determine the hole boundary; The connected component analysis in step S4 includes: Connected regions are labeled using measure.label according to the 8-adjacency rule; Extract the area, perimeter, and centroid coordinates of each region using measure.regionprops; The contour extraction in step S5 uses a combination of parameters:
[0026] in Only the outermost contour is retrieved; Compress redundant contour points; The pore size analysis in step S6 includes: Calculate the vertex coordinate set of the convex hull of the contour ; Traverse all vertex pairs Calculate Euclidean distance ,in , As vertex The x and y coordinates, , As vertex The horizontal and vertical coordinates; Take the maximum distance as the capillary pore diameter; The statistical grouping in step S7 includes: Custom aperture range ; Statistics on the percentage of capillary pores in each interval ,in, For the first The percentage of capillaries within the pore size range. For the first Number of capillaries within a certain pore size range This represents the total number of capillaries across all pore size ranges. The pie chart generation in step S8 is implemented using matplotlib.pyplot.pie, and the pie chart block labels include aperture ranges and percentage percentages. include: Image acquisition module: Industrial camera and light source device, to acquire RGB images of the capillary surface; Preprocessing module: Performs grayscale conversion and adaptive threshold segmentation; Optimization module: Remove small holes and mark connected regions; Analysis module: Extract contour, calculate convex hull diameter and physical dimension conversion; Statistical output module: Groups and statistically analyzes aperture distribution and generates pie charts.
[0027] Example 1: Surface Capillary Analysis of Porous Ceramic Materials In this embodiment, an image acquisition platform is built using an industrial camera with a resolution of no less than 12 million pixels and a ring LED light source. During the sample preparation stage, the surface of the ceramic sheet needs to be polished to a roughness Ra of less than 0.1 μm, and a nano-level conductive coating is sprayed to enhance the uniformity of surface reflection.
[0028] Image acquisition and preprocessing: The brightness adaptive mode of the ring light source is enabled. The system dynamically adjusts the intensity ratio of the three channels according to the surface reflectivity. The camera captures a 1024×1024 pixel RGB image under a 10x microscope lens. A grayscale image is generated by weighted grayscale conversion formula. The weight of the green channel is increased to 0.65 to enhance the contrast of the hole edges. Then, multi-scale denoising processing is performed. First, the image is decomposed using a Gaussian-Laplacian pyramid. Non-local mean filtering is applied in the low-frequency layer to eliminate uneven illumination. In the high-frequency layer, wavelet threshold shrinkage is used to suppress salt-and-pepper noise.
[0029] Hole segmentation and optimization: The denoised image is segmented using a dual-path method. Path 1 uses an adaptive thresholding method to calculate a local Gaussian weighted threshold with an 11×11 pixel window to segment the main hole outline. Path 2 uses the Canny operator to extract edge gradients and combines morphological closing operations to close small breaks. The binary images generated by the two paths are superimposed, and independent holes are distinguished by marking the 8-neighbor connected regions. Hole optimization is performed on the marking results. The area threshold filtering function is used to remove noise points with a pixel area of less than 25, and local depressions caused by surface texture are filled.
[0030] Aperture Calculation and Verification: A dual-modal analysis is performed on the contour of each hole. Modal 1 is based on the topological relationship of the convex hull vertices, and the maximum Euclidean distance between the convex hull vertices is calculated as the primary aperture D1. Modal 2 uses an ellipse fitting algorithm, which fits an ellipse model by randomly sampling 6 sets of contour points. After removing outliers with an aspect ratio greater than 3, the average minor axis length is taken as the secondary aperture D2. The final aperture is determined by weighted sum of the dual-modal results. This method improves the measurement accuracy of irregular holes and reduces the error compared to the traditional circular fitting model.
[0031] Dynamic statistics and output: The aperture pixel values are converted into physical dimensions according to the calibration ratio. A density-sensitive interval division strategy is adopted. When the aperture is concentrated in 30μm, the system automatically generates group intervals with 5μm intervals. When a bimodal distribution is detected, the peak interval is refined to 2μm intervals. The statistical results are used to generate composite charts through the matplotlib library: pie charts show the proportion of holes in each interval, and bar charts are overlaid to show the cumulative distribution curve of aperture. The output report includes engineering parameters such as porosity, average aperture, and D90.
[0032] Verification experiment: A standard polystyrene microsphere sample was selected for accuracy verification. The average diameter obtained by the system after 10 repeated measurements was 50.3 μm with a standard deviation of 0.21 μm. Compared with the mercury intrusion porosimetry test of the same batch of ceramic samples, the average pore size measured by this system deviated from that of the mercury intrusion porosimetry by less than 4.7%, and effectively identified 3 oversized pores that were not detected by the mercury intrusion porosimetry.
[0033] Example 2: Pore Analysis of Polymer Filter Membrane Materials In this embodiment, a capillary flow aperture analyzer is used to detect the characteristic pore size of the proton exchange membrane used in fuel cells.
[0034] Sample preparation and wetting treatment: Cut a 25 mm diameter proton exchange membrane sample and place it in a 60 °C vacuum drying oven for 2 hours to dehydrate. After drying, completely immerse the sample in the low surface tension wetting solution Porewik, evacuate to -90 kPa and maintain for 30 minutes to ensure that the wetting solution fully fills all the through holes. Take out the saturated wetted sample and fix it vertically in the sample chamber fixture to avoid membrane folding and edge leakage.
[0035] Gas permeation curve test: After sealing the sample chamber, the nitrogen pressurization system was started, with the initial pressure set at 0.1 bar. The pressure was increased in steps at a rate of 0.01 bar / second, and the gas flow rate was recorded in real time. When the pressure reached 0.85 bar, the first bubble was detected, corresponding to a maximum through-hole diameter of 12.3 μm. The pressure was further increased to 4.2 bar, and the gas flow rate of the wet curve increased to 360 ml / min. The dry sample test mode was then switched, and the maximum flow rate of the fully dry curve was measured to be 4850 ml / min within the same pressure range. The system automatically generated a semi-dry curve as a reference.
[0036] Characteristic pore size calculation: The average flow pore size was determined to be 3.7 μm by the intersection of the wet curve and the semi-dry curve; the minimum pore size corresponding to the intersection of the wet curve and the fully dry curve is 0.8 μm. Based on the Washburn equation, the pore size distribution was calculated. A perpendicular line was drawn at the 1.2 bar pressure point, intersecting the wet curve at Qw=82 ml / min and the fully dry curve at Qd=1020 ml / min. At this pressure point, the flow rate of the orifice with a pore size of 7.5 μm accounted for 8.04%. Traversing the pressure range of 0.1-4.2 bar, the pore size-flow rate distribution curve showed that the main peak was located in the 2-5 μm range, accounting for 63%.
[0037] Validation and Industrial Parameter Output: The accuracy was verified using a NIST-certified polycarbonate standard membrane. The test was repeated 5 times and the average pore size was measured to be 5.1 μm with a relative error of 2%. The output industrial parameters include bubble point pressure of 0.85 bar, maximum pore size of 12.3 μm, average flow pore size of 3.7 μm, and D90 pore size of 8.5 μm. At the same time, a pie chart of pore size distribution was generated to indicate the proportion of through-hole volume in each interval.
[0038] Example 3: Analysis of the pore characteristics of proton exchange membranes for fuel cells Sample preparation and pretreatment: Cut a 25 mm diameter Nafion 117 proton exchange membrane sample and place it in a 60 °C vacuum drying oven for 2 hours to remove internal moisture interference. After drying, completely immerse the sample in the low surface tension wetting solution Porewik, evacuate to -90 kPa and maintain for 30 minutes to ensure that the wetting solution fully fills all pores. After removing the saturated wetted sample, fix it vertically in the sample chamber fixture and apply 0.5 MPa lateral pressure through an annular airbag to prevent edge leakage.
[0039] Three-curve dynamic test: Wet curve acquisition. After sealing the sample chamber, the nitrogen step pressurization system was started with an initial pressure of 0.1 bar, increasing at a rate of 0.01 bar / second, while simultaneously recording the gas flow rate. When the pressure reached 0.85 bar, the first bubble was detected, corresponding to a maximum pore size of 12.3 μm. The pressure was further increased to 4.2 bar, and the gas flow rate rose to 360 ml / min, completing the wet curve acquisition. Full dry curve calibration was performed, and the dry sample test mode was switched. Within the same pressure range, the maximum gas flow rate of the membrane material without wetting liquid was measured to be 4850 ml / min, generating a full dry curve as a baseline. A semi-dry curve was generated, and the system automatically extracted 50% of the flow rate of the full dry curve to generate the semi-dry curve. The average pore size was calculated.
[0040] Characteristic pore size and distribution analysis: Bubble point pore size, 12.3 μm at 0.85 bar bubble point pressure; average flow rate pore size, flow rate at the intersection of wet and semi-dry curves is 1820 ml / min, calculated pore size is 3.7 μm; minimum pore size, 0.8 μm at the intersection of wet and fully dry curves; pore size distribution, a perpendicular line is drawn at 1.2 bar pressure point, intersecting the wet curve at Qw=82 ml / min and the fully dry curve at Qd=1020 ml / min. At this pressure point, the flow rate of the 7.5 μm pore size through the orifice accounts for 8.04%. Plotting the pore size-flow rate distribution curve across the 2 bar pressure range shows that the main peak is located in the 2-52.5 μm range, accounting for 63%.
[0041] Anisotropy verification: Repeated testing with a rotating fixture at 90° revealed that the diameter of the through holes in the mechanical tensile direction of the membrane material increased by 15%, indicating the directional difference in pore structure caused by fiber orientation.
[0042] Engineering parameter output and verification: Industrial parameters: output bubble point pressure 0.85 bar, maximum pore size 12.3 μm, average flow pore size 3.7 μm, D90 pore size 8.5 μm. Accuracy verification: repeated testing 5 times using NIST certified polycarbonate standard membrane, average pore size 5.1 μm. System comparison: mercury intrusion porosimetry test of the same batch of membrane material showed an average pore size of 3.9 μm, with a deviation of less than 4.7% from this system, and no three oversized pores larger than 150 μm were detected.
[0043] Example 4: Research and Application Progress of Gas Diffusion Layer Optimization Technology for Proton Exchange Membrane Fuel Cells Innovative Materials and Structural Design: The traditional gas diffusion layer's microporous layer and carbon paper substrate layer lack sufficient bonding strength, making them prone to cracking and detachment under humidity fluctuations and cold start conditions, leading to battery flooding and increased resistance. The novel manufacturing process enhances interfacial bonding through a combination of physical imprinting and chemical treatment. Physical imprinting uses a specially designed metal mesh to apply 6 kg / cm² pressure to the carbon paper substrate, forming an anchoring structure that increases the substrate's surface roughness and expands the adhesion area of the microporous layer slurry. Chemical treatment involves impregnating the substrate with 5% polytetrafluoroethylene (PTFE) emulsion, followed by infrared drying and calcination at 350°C. This allows a hydrophobic agent to uniformly coat the carbon fibers, forming a hydrophobic barrier. This treatment stabilizes the hydrophobic agent content in the substrate layer at 7% and increases the hydrophobic agent content in the microporous layer slurry, effectively preventing liquid water penetration.
[0044] Gradient PTFE treatment process: To address the problem of uneven hydrophobicity distribution caused by traditional single PTFE treatment, multi-stage PTFE gradient treatment technology achieves hydrophobicity gradient distribution by controlling the number of loading cycles, thereby enhancing surface hydrophobicity. The contact angle of the GDL surface after 6-stage PTFE treatment reaches 152°, which is 15% higher than that of single-stage treatment. Moreover, after 5 compression cycles, the contact angle decay rate decreases from 23% to 9%. Pore size optimization: Multi-stage treatment reduces the average pore size of GDL from 30.57μm to 15.68μm, forming a gradient pore structure that gradually increases from the surface to the interior, inhibiting the accumulation of liquid water on the near-film electrode side under high humidity conditions.
[0045] Application of film-forming agents in microporous layer slurry: Adding 10% polyethylene glycol and polyvinyl alcohol as film-forming agents to microporous layer slurry can alleviate stress concentration during the drying process. After pre-sintering at 290℃ and final sintering at 350℃, the crack rate of the microporous layer is reduced and the interfacial peel strength is increased to 0.8MPa.
[0046] Compression Deformation and Performance Regulation Mechanism: The attenuation law under cyclic compression: The gas diffusion layer is subjected to periodic compression loads during repeated assembly of the fuel cell stack, resulting in irreversible structural degradation and mechanical attenuation. After the first compression, the carbon paper thickness decreases by 4.64%, and after the second compression, the cumulative attenuation exceeds 10%. The pressure required for the tenth compression is lower than that of the first compression, indicating that the fiber structure has undergone plastic collapse and the resistance has increased. Under the same compression rate, the bulk resistance of GDL increases after the second assembly, and the contact resistance with the carbon-plated metal plate increases. Laser confocal microscopy shows that the height difference of carbon fiber decreases from 12.3μm to 5.6μm, and the effective contact area is reduced.
[0047] The compression compensation design strategy proposes a dynamic compression rate adjustment scheme based on a satisfaction function. In the first assembly, the compression rate is reduced from 20% to 18% to avoid fiber breakage due to overpressure. In the second assembly, the compression rate is increased to 22% to compensate for thickness loss and maintain interface contact. In the third and subsequent assemblies, the compression rate increases to 24% to offset the increase in contact resistance. Engineering verification shows that this scheme reduces the voltage fluctuation range of the stack from ±8.2% to ±2.1% after three assemblies, and the power attenuation rate is reduced from 12.4% to 3.7% after 100 start-stop cycles.
[0048] To address the contradiction between porosity and conductivity caused by homogeneous compression, an asymmetric gradient compression GDL design with a sandwich structure is proposed. The top layer is compressed by 40% to improve conductivity, while the bottom layer is compressed by 20% to maintain high porosity. The intermediate transition layer uses a gradient sintering process to achieve a smooth transition in pore size, resulting in improved performance. Under the condition of 1.2A / cm², the gradient structure output voltage is increased, the cold start time at -20℃ is shortened to 35 seconds, and the power decay is less than 5% after 200 hours of start-stop cycles.
[0049] Multi-scale modeling and performance prediction: Digital twin model construction, based on X-ray tomography and machine learning technology, realizes digital reconstruction of GDL microstructure and actual structure analysis. SGL22BB type GDL is scanned with X-ray CT with a resolution of 2.52μm. The distribution of carbon fiber and resin is segmented by U-Net neural network to identify the resin enrichment characteristics at fiber intersections. Virtual structure optimization is performed by generating a digital twin model with carbon fiber diameter, resin volume fraction and fiber orientation as parameters to simulate gas diffusion performance at 30% moisture content. The optimized virtual structure has 40% higher gas diffusion performance than the traditional structure at high moisture content.
[0050] Cross-scale simulation guides material design, combining macroscopic compression experiments and microscopic fluid dynamics simulations to reveal the impact mechanism of GDL compression on battery performance. An impedance trade-off model shows that smaller compression leads to an increase in ohmic impedance, while larger compression causes a sharp increase in mass transfer impedance. Simulations show that 20% compression is the marginal effect balance point between ohmic impedance and mass transfer impedance. Engineering verification shows that by optimizing the compression, the output performance of a single cell is improved, and the helium leakage rate is reduced to 0.25 mL / min, meeting the U.S. Department of Energy sealing standards.
[0051] Engineering Applications and Verification: In automotive fuel cell system integration, gradient compression GDL and multi-stage PTFE treatment technology were applied to a 30kW fuel cell stack, achieving a breakthrough in durability with a measured lifespan exceeding 6000 hours, meeting the DOE 2025 technical target. Anisotropic testing showed that the aperture of the through-hole in the mechanical tensile direction increased by 15%, which optimized the flow field design to improve the uniformity of oxygen distribution on the cathode side. Cold start performance was improved, and the optimized microporous structure suppressed ice crystal formation, reducing the start-up time to 35 seconds at -30℃, an improvement over traditional structures.
[0052] Industrial parameter correlation and system compatibility: GDL structural optimization is directly related to key engineering parameters and mass transfer performance. Frazier permeability is increased to 28.5 cfm / ft², and Gurley value is reduced to 12.3s / 100ml, supporting stable operation at high current density. Cost control: Surface roughening and film-forming agent technology reduce GDL manufacturing costs, and the compensation scheme only requires the addition of a displacement sensor, which is more economical than replacing materials.
[0053] Technical challenges and future directions: Interface stability of ultrathin GDLs. Currently, GDLs below 50μm face the problem of stress concentration at the interface between the substrate and the microporous layer. It is necessary to develop nanoscale binders and in-situ sintering processes to improve the interfacial shear strength while maintaining high porosity.
[0054] Application of smart responsive materials: Exploring temperature / pH-sensitive polymers as dynamic hydrophobic agents to enable GDL pore size to automatically adjust according to operating conditions, increasing pore size to accelerate drainage during low-temperature startup and reducing pore size to prevent membrane drying during high-temperature operation.
[0055] Cross-scale manufacturing process integration: Combining additive manufacturing and chemical vapor deposition technologies, 3D printing carbon fiber skeletons achieves precise control of porosity and fiber orientation, with anisotropy control accuracy of ±2%. Atomic layer deposition technology deposits a 1.5nm PTFE film on the fiber surface, improving the utilization rate of hydrophobic agents and reducing costs.
[0056] Conclusion: Optimization of the gas diffusion layer in proton exchange membrane fuel cells requires consideration of material innovation, structural design, and engineering compensation strategies. Current research improves electrochemical durability by 40% through multi-stage hydrophobic treatment and enhances output performance and cold start efficiency by combining cross-scale modeling. In the future, it is necessary to overcome the two major bottlenecks of ultra-thin interface strengthening and smart materials to promote the large-scale application of fuel cells in heavy-duty transportation and distributed energy fields.
[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A capillary diameter analysis system and method based on image recognition, characterized in that: Includes the following steps: S1. Collect capillary surface image data, acquire the original RGB image using an industrial camera, and convert the image based on the grayscale conversion formula. Calculate pixel grayscale values to generate grayscale data; where... These represent the values for the red, green, and blue channels, respectively. S2. Perform adaptive threshold segmentation on the grayscale data, and use the Gaussian weighted mean method cv2.ADAPTIVE_THRESH_GAUSSIAN_C to calculate the weight distribution of pixels within the local window and generate threshold processing data; S3. Perform hole optimization processing on the threshold processing data. Remove holes with an area smaller than the set threshold area_threshold using the function morphology.remove_small_holes to generate optimized data. S4. Perform connected component analysis on the optimized data, and use the functions measure.label and measure.regionprops to identify the pore outline and segment individual pores to generate connected component data; S5. Extract contours from the connected region data by using the OpenCV function cv2.findContours to retrieve the set of contour points in the binary image and generate contour data. S6. Perform aperture analysis on the contour data, calculate the vertices of the contour convex hull using the function cv2.convexHull, measure the maximum distance between the vertices of the convex hull as the capillary aperture, and generate aperture data. S7. Convert the aperture pixel values into physical dimensions according to the calibration ratio, and then group them statistically. S8. Generate a pie chart of capillary diameter distribution based on grouped data to show the proportion of each pore size range.
2. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: The weight allocation of the grayscale conversion formula in step S1 is designed based on the visual characteristics of the human eye, which gives the highest weight of 0.587 for green light, the second highest weight of 0.299 for red light, and the lowest weight of 0.114 for blue light.
3. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: In the adaptive threshold segmentation of step S2, the weight distribution of the local window satisfies a Gaussian function. in Standard deviation, These are pixel coordinates.
4. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: The hole optimization process in step S3 satisfies: ,in This indicates that the 8-adjacency mode is used to determine the hole boundary.
5. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: The connected component analysis in step S4 includes: Connected regions are labeled using measure.label according to the 8-adjacency rule; The area, perimeter, and centroid coordinates of each region can be extracted using measure.regionprops.
6. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: The contour extraction in step S5 uses a combination of parameters: in Only the outermost contour is retrieved; Compress redundant contour points.
7. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: The pore size analysis in step S6 includes: Calculate the vertex coordinate set of the convex hull of the contour ; Traverse all vertex pairs Calculate Euclidean distance ,in , As vertex The x and y coordinates, , As vertex The horizontal and vertical coordinates; The maximum distance is taken as the capillary pore diameter.
8. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: The statistical grouping in step S7 includes: Custom aperture range ; Statistics on the percentage of capillary pores in each interval ,in, For the first The percentage of capillaries within the pore size range. For the first Number of capillaries within a certain pore size range This represents the total number of capillaries across all pore size ranges.
9. The capillary diameter analysis system and method based on image recognition according to claim 1, characterized in that: The pie chart generation in step S8 is implemented using matplotlib.pyplot.pie, and the pie chart block labels include aperture ranges and percentage percentages.
10. A capillary diameter analysis system based on image recognition, implementing the capillary diameter analysis method based on image recognition as described in any one of claims 1-9, characterized in that, include: Image acquisition module: Industrial camera and light source device, to acquire RGB images of the capillary surface; Preprocessing module: Performs grayscale conversion and adaptive threshold segmentation; Optimization module: Remove small holes and mark connected regions; Analysis module: Extract contour, calculate convex hull diameter and physical dimension conversion; Statistical output module: Groups and statistically analyzes aperture distribution and generates pie charts.
