Detection system and method for diversion bipolar plate of hydrogen fuel cell

Abstract

The invention relates to the technical field of bipolar plate detection, and discloses a detection system and method for a diversion bipolar plate of a hydrogen fuel cell, and the method comprises the steps: carrying out the defect recognition of collected image data based on a multi-scale feature fusion network, and carrying out the airtightness detection of the bipolar plate based on a multi-station airtightness testing device. And determining a comprehensive quality score of each bipolar plate according to the defect data and the air tightness data in combination with an attention mechanism and fuzzy logic. And determining a defect severity grade according to a relationship between the comprehensive quality score and a configured preset comprehensive quality score. And based on a Monte Carlo Dropout method, evaluating the confidence coefficient of bipolar plate grading after defect severity grading, and generating an output report. Through the technical means of multi-scale feature fusion, air tightness testing, intelligent dynamic adjustment, uncertainty evaluation and the like, the precision, stability and reliability of bipolar plate defect detection are improved, and precise recognition and comprehensive quality evaluation of complex defects of the bipolar plate are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of bipolar plate detection, and more particularly, to a detection system and method for a flow guiding bipolar plate of a hydrogen fuel cell. Background Art

[0002] As an important technology for clean energy, the flow guiding bipolar plate of a hydrogen fuel cell plays an important role in separating gases, conducting current, and supporting the structure in the fuel cell system. The manufacturing precision and quality of the bipolar plate directly affect the performance, lifespan, and safety of the fuel cell. Therefore, it is crucial to detect the bipolar plate with high precision and efficiency.

[0003] Currently, traditional detection methods mainly rely on single optical imaging or airtightness testing. However, when faced with complex defects (such as hidden cracks and micro depressions) and comprehensive evaluation of multiple factors, there are problems such as limited recognition ability, high false detection rate, and difficulty in accurately quantifying the impact of defects on the overall quality, which cannot meet the requirements of the fuel cell industry for high-precision detection. Moreover, most detection methods use single-modal visual detection or independent physical testing methods, making it difficult to fully utilize multi-source data to improve detection accuracy. For example, although optical detection can identify surface scratches and depressions, its detection ability for micro cracks or internal defects is limited; while airtightness testing can judge leakage conditions, it is difficult to directly locate the defect position. In addition, traditional defect evaluation relies on fixed rules and lacks an intelligent dynamic adjustment mechanism, resulting in insufficient reliability of detection results and difficulty in adapting to detection requirements under different working conditions. At the same time, existing methods lack an evaluation of the uncertainty of detection results, making it difficult for low-confidence classification results to provide an effective decision-making basis, affecting the reliability and stability of detection.

[0004] Therefore, there is an urgent need to invent a detection technology for bipolar plates to solve the problems of limited recognition ability of existing detection methods when faced with complex defects, difficulty in comprehensively evaluating multi-source data accurately, lack of intelligent dynamic adjustment and uncertainty quantification mechanisms, resulting in low detection accuracy and reliability for bipolar plates. Summary of the Invention

[0005] In view of this, the present invention proposes a detection system and method for a flow guiding bipolar plate of a hydrogen fuel cell, aiming to solve the problems of limited recognition ability when faced with complex defects in the current technology, difficulty in comprehensively evaluating multi-source data accurately, lack of intelligent dynamic adjustment and uncertainty quantification mechanisms, and thus resulting in low detection accuracy and reliability for bipolar plates.

[0006] The present invention proposes a detection method for a flow guiding bipolar plate of a hydrogen fuel cell, including:

[0007] Performing defect recognition on the collected image data based on a multi-scale feature fusion network, and performing airtightness detection on the bipolar plate based on a multi-station airtightness testing device;

[0008] Based on the defect data and airtightness data, combined with the attention mechanism and fuzzy logic, determine the comprehensive quality score of each bipolar plate;

[0009] According to the relationship between the comprehensive quality score and the preset comprehensive quality score configured, determine the severity level classification of the defects;

[0010] Based on the Monte Carlo Dropout method, evaluate the confidence level of the bipolar plate classification after the severity level classification of the defects, and generate an output report.

[0011] Furthermore, when performing defect recognition on the collected image data based on the multi-scale feature fusion network, it includes:

[0012] Preprocess the images collected under different lighting conditions to enhance the contrast and edge information of the images;

[0013] Based on the multi-scale feature extraction technology, extract the local detail information and global structure information of different receptive fields in the preprocessed images, and perform weighted fusion according to the extracted multi-scale features to obtain the comprehensive feature of the processed images;

[0014] According to the comprehensive feature of the fused images, perform a threshold comparison on the feature responses of each region in the images to determine whether there are defects in the bipolar plates.

[0015] Furthermore, when preprocessing the images collected under different lighting conditions, it includes:

[0016] Obtain the image contrast of each region in the collected images, and obtain the average value of the image contrast of each region. Compare according to the average value of the image contrast and the image contrasts to obtain the images of each region with an image contrast lower than the average value of the image contrast;

[0017] For the images of each region with an image contrast lower than the average value of the image contrast, perform brightness normalization processing based on histogram equalization;

[0018] Obtain the image information after brightness normalization processing, and perform denoising processing on the images based on non-local means filtering.

[0019] Furthermore, when obtaining the comprehensive feature of the processed images, it includes:

[0020] For the preprocessed images, adopt the multi-scale feature extraction technology, and based on a number of preset receptive fields of different sizes, extract the local detail information and global structure information in the images simultaneously;

[0021] According to the extracted multi-scale features, perform weighted fusion based on the preset weights to generate the comprehensive image feature, where:

[0022] Set the weights of the preset scale features according to historical data, match the extracted scale features with each preset scale feature, obtain the weights of the extracted scale features, and perform weighted stitching on the local details and global structure information according to the weights to generate comprehensive features.

[0023] Further, when comparing the feature responses of each region in the image with a threshold according to the fused image comprehensive features to determine whether there are defects in the bipolar plates, it includes:

[0024] Based on historical data and experimental samples, establish a correlation formula between the detection index and the defect response;

[0025] Obtain the distance metric between each correlation formula, and perform iterative clustering on each correlation formula according to the distance metric;

[0026] Based on the isolation forest tree algorithm, obtain the outliers of each clustering correlation formula in the clustering result, and determine the normal features and defect features in the same region according to the relationship between the outliers and the preset outliers;

[0027] Determine the defect judgment threshold according to the feature vector between the normal feature and the defect feature;

[0028] Obtain the feature vector of the feature response of the region in the image, and determine whether there are defects in the region in the image according to the relationship between the feature vector of the feature response and the defect judgment threshold:

[0029] When the feature vector of the feature response is lower than the defect judgment threshold, it is determined that there are no defects in the region in the image;

[0030] When the feature vector of the feature response is higher than or equal to the defect judgment threshold, it is determined that there are defects in the region in the image.

[0031] Further, when determining the comprehensive quality score of each bipolar plate according to the defect data and airtightness data, combining the attention mechanism and fuzzy logic, it includes:

[0032] Extract features from different modalities of the defect data and airtightness data of the bipolar plate;

[0033] Based on the attention mechanism, perform weighted analysis on the defect data and airtightness data to determine the key feature information;

[0034] Based on fuzzy logic, dynamically allocate the weights of the defect data and airtightness data, and determine the contribution ratio of each data in the quality assessment according to the confidence of the data;

[0035] Based on the weighted fusion of the defect score, airtightness score and morphology feature score, determine the comprehensive quality score of the bipolar plate.

[0036] Further, when obtaining the defect score, airtightness score, and morphology feature score of the bipolar plate, it includes:

[0037] Collect defect data of the bipolar plate, perform defect recognition based on computer vision algorithms, extract the size, shape, and distribution characteristics of the defects, and generate a defect score;

[0038] Obtain airtightness data, and determine the airtightness score based on the leakage rate and air pressure change parameters;

[0039] Obtain the surface morphology information of the bipolar plate based on three-dimensional imaging technology, and determine the morphology feature score based on the surface flatness and deformation degree.

[0040] Further, when determining the defect severity level according to the relationship between the comprehensive quality score and the preset comprehensive quality score configured, it includes:

[0041] Determine the defect severity level according to the relationship between the comprehensive quality score and the first preset comprehensive quality score and the second preset comprehensive quality score configured:

[0042] When the comprehensive quality score is lower than the first preset comprehensive quality score, it is determined that the defect severity level is a high level;

[0043] When the comprehensive quality score is higher than or equal to the first preset comprehensive quality score and lower than the second preset comprehensive quality score, it is determined that the defect severity level is a medium level;

[0044] When the comprehensive quality score is higher than or equal to the second preset comprehensive quality score, it is determined that the defect severity level is a low level;

[0045] Among them, the first preset comprehensive quality score is less than the second preset comprehensive quality score, and the defect severity levels from high to low are high level, medium level, and low level in sequence.

[0046] Further, when evaluating the confidence level of the bipolar plate classification after defect severity level classification based on the Monte Carlo Dropout method and generating an output report, it includes:

[0047] Perform multiple forward propagations in the inference stage based on the Monte Carlo Dropout method to obtain multiple prediction results of the defect classification;

[0048] Obtain the probability distribution characteristics of the defect classification prediction results, and determine the classification confidence level based on uncertainty measurement methods such as entropy value or variance;

[0049] Determine the output result according to the relationship between the confidence level and the preset confidence level configured:

[0050] When the confidence level is lower than the preset confidence level, it is determined that a secondary evaluation of the comprehensive quality score of the bipolar plate is to be carried out;

[0051] When the confidence level is higher than or equal to the preset confidence level, it is determined that the output is to be carried out according to the classification of the defect severity of the bipolar plate.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows: Through multi-scale feature fusion, the system can extract key defect information from different image data, so as to identify complex defects including hidden cracks and micro depressions. In addition, the combination of airtightness testing and image detection makes the detection more comprehensive, which can not only identify surface defects, but also accurately evaluate the integrity of the internal structure through airtightness testing. The synergistic effect of this multi-modal data significantly improves the accuracy and comprehensiveness of the detection, avoiding the problems of missed detection and misdetection that may occur in traditional single detection methods. In addition, by combining the attention mechanism and fuzzy logic, defect data and airtightness data can be processed more intelligently, dynamically adjusting the weights of different data sources to generate a more accurate comprehensive quality score. This mechanism can effectively cope with changes under different working conditions and improve the adaptive ability of the detection system. At the same time, evaluating the confidence level of the defect severity based on the Monte Carlo Dropout method can quantify the uncertainty of the detection results, provide a reliable basis for the final decision-making, and ensure the accuracy of the classification results. Finally, the output quality report provides more objective and scientific data support for production manufacturing and quality control, optimizes the decision-making in the detection process, and improves the overall quality assurance ability of the bipolar plate.

[0053] On the other hand, the present application also provides a detection system for a hydrogen fuel cell flow guiding bipolar plate, including:

[0054] A composite optical imaging unit, configured with an annular LED array light source, a near-infrared imaging module, and a structured light module, and the composite optical imaging unit is configured to perform defect identification on the collected image data based on a multi-scale feature fusion network;

[0055] A multi-station airtightness testing device, configured to perform airtightness detection on the bipolar plate;

[0056] An evaluation module, configured to determine the comprehensive quality score of each bipolar plate according to the defect data and the airtightness data, in combination with the attention mechanism and fuzzy logic; the evaluation module is also configured to determine the defect severity classification according to the relationship between the comprehensive quality score and the preset comprehensive quality score configured;

[0057] An output module, configured to evaluate the confidence level of the bipolar plate classification after the defect severity classification based on the Monte Carlo Dropout method and generate an output report.

[0058] It is understandable that the detection system and method for the flow guiding bipolar plate of a hydrogen fuel cell in each of the above embodiments of the present invention have the same beneficial effects and will not be elaborated herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered as limiting the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0060] Figure 1 is a flowchart of a detection method for a flow guiding bipolar plate of a hydrogen fuel cell provided by an embodiment of the present invention;

[0061] Figure 2 is a functional block diagram of a detection system for a flow guiding bipolar plate of a hydrogen fuel cell provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0063] As Figure 1 shown, in some embodiments of the present application, the present embodiment provides a detection method for a flow guiding bipolar plate of a hydrogen fuel cell, including:

[0064] Step S100, performing defect recognition on the collected image data based on a multi-scale feature fusion network, and performing airtightness detection on the bipolar plate based on a multi-station airtightness testing device.

[0065] Specifically, when performing defect recognition on the collected image data based on a multi-scale feature fusion network, it includes: preprocessing the images collected under different lighting conditions to enhance the contrast and edge information of the images. Based on the multi-scale feature extraction technology, extracting the local detail information and global structure information of different receptive fields in the preprocessed images, and performing weighted fusion according to the extracted multi-scale features to obtain the comprehensive feature of the processed image. According to the comprehensive feature of the fused image, comparing the feature responses of each region in the image with a threshold to determine whether there are defects on the bipolar plate.

[0066] Specifically, when preprocessing the images collected under different lighting conditions, it includes: obtaining the image contrast of each region in the collected images, obtaining the average value of the image contrast of each region, comparing based on the average value of the image contrast and each image contrast, and obtaining the images of each region with an image contrast lower than the average value of the image contrast. For the images of each region with an image contrast lower than the average value of the image contrast, perform brightness normalization processing based on histogram equalization. Obtain the image information after brightness normalization processing, and perform denoising processing on the image based on non-local means filtering.

[0067] Specifically, when obtaining the comprehensive features of the processed images, it includes: adopting a multi-scale feature extraction technique for the preprocessed images, and based on a number of preset receptive fields with different sizes, simultaneously extracting the local detail information and global structure information in the images. Perform weighted fusion on the extracted multi-scale features based on preset weights to generate comprehensive image features, where: set the weights of the preset scale features according to historical data, match between the extracted scale features and each preset scale feature, obtain the weights of the extracted scale features, and perform weighted stitching on the local detail and global structure information according to the weights to generate comprehensive features.

[0068] Specifically, when comparing the feature responses of each region in the image with a threshold according to the fused comprehensive image features to determine whether there are defects in the bipolar plate, it includes: establishing a correlation formula between the detection index and the defect response based on historical data and experimental samples. Obtain the distance metric between each correlation formula, and perform iterative clustering on each correlation formula according to the distance metric. Obtain the outliers of each clustering correlation formula in the clustering result based on the isolation forest tree algorithm, and determine the normal features and defect features in the same region according to the relationship between the outliers and the preset outliers. Determine the defect determination threshold according to the feature vectors between the normal features and the defect features. Obtain the feature vector of the feature response of the region in the image, and determine whether there are defects in the region in the image according to the relationship between the feature vector of the feature response and the defect determination threshold: when the feature vector of the feature response is lower than the defect determination threshold, it is determined that there are no defects in the region in the image. When the feature vector of the feature response is higher than or equal to the defect determination threshold, it is determined that there are defects in the region in the image.

[0069] It can be seen that by analyzing the contrast of the image, calculating the average contrast of each region, and finding the regions below the average, the image information in these regions will be enhanced in subsequent processing. Brightness normalization processing (such as histogram equalization) and denoising (such as non-local means filtering) ensure that the noise in the image is effectively removed, making subsequent defect detection more accurate and eliminating the influence of uneven illumination or noise interference. Next, the image is processed by multi-scale feature extraction technology. In this process, the image is divided into multiple different receptive fields for feature extraction, and each receptive field can extract different-scale local detail information and global structure information. The multi-scale feature extraction technology can capture the details and macroscopic structures in the image, which is crucial for identifying tiny defects or cracks with complex shapes. By setting several preset receptive field scales, feature information at different scales is extracted, providing rich feature support for subsequent defect judgment. After multi-scale feature extraction, the extracted features at each scale are weighted and fused according to the set weights. Historical data is used to set the weights of the preset scale features, and these weights can be trained by a machine learning model based on existing labeled data to ensure that important features obtain higher weights while secondary features obtain lower weights. After weighted fusion, the generated comprehensive image features include fine information of local details and global structures, enabling more accurate defect identification and classification in subsequent stages. Based on the comprehensive image features after feature fusion, the feature responses of each region in the image are compared with a threshold. By establishing a correlation formula based on historical data and experimental samples, the similarity of each feature response can be measured, the distance metric between each feature can be calculated, and this information is used for clustering analysis. Through clustering analysis, normal and abnormal regions can be effectively distinguished, and the defective regions can be identified. Finally, through the isolation forest tree algorithm, the outliers in the clustering results can be identified. By comparing these outliers with the preset outliers, the characteristic differences between defects and normal regions can be determined, and the defect judgment threshold can be set based on this. Ultimately, by comparing the feature vector with the threshold, it is determined whether there are defects in the regions of the image. If the feature response of a region is lower than the judgment threshold, it is considered that there is no defect in this region; if the feature response is higher than or equal to the judgment threshold, it is considered that there is a defect in this region.

[0070] It can be understood that in the image preprocessing stage, it enhances the contrast and processes the edge information of the images collected under different lighting conditions, improving the image quality. Since the lighting conditions during the collection process may be uneven, directly affecting the image clarity and details, the image preprocessing eliminates these lighting problems through contrast analysis and brightness normalization processing (such as histogram equalization), thus ensuring the accuracy and stability of the image in subsequent analysis. The denoising process further improves the image quality, making the defect detection process more reliable and reducing the interference of noise on image feature extraction. Secondly, through the multi-scale feature extraction technology, it is able to extract both local detail information and global structure information in the image simultaneously. The settings of different receptive fields enable the detection to capture defect features of different sizes and complexities from multiple scales. This technical method enhances the recognition ability of microcracks, internal defects, and other complex-shaped defects, making the detection more accurate, especially providing effective recognition support when facing tiny and hidden defects. By weighted fusion of multi-scale features, it can reasonably allocate the contributions of each scale feature according to historical data and preset weights. The process of weighted fusion improves the integration ability of different image features, effectively combines the detail information and global structure information, and generates comprehensive image features. This not only enhances the accuracy of defect recognition but also enables the detection strategy to be adaptively adjusted according to different image features, improving the overall detection effect. In addition, the combination of threshold comparison and clustering analysis enables the establishment of a correlation formula based on historical data and experimental samples, thus providing stronger theoretical support for defect determination. Through distance measurement and clustering analysis, it can effectively distinguish the normal area and the defect area, improving the recognition accuracy. The outliers in the clustering results can be effectively identified, providing a reliable basis for defect judgment, and further strengthening the accuracy through comparison with the preset outliers, avoiding false detection or missed detection. Finally, based on the isolation forest tree algorithm, it can optimize the results of clustering analysis, identify outliers, and perform effective feature determination. By determining the defect determination threshold, it can conduct a detailed detection of each area in the image and accurately judge whether there are defects in the area according to the relationship between the feature response and the threshold. The effectiveness of this process ensures that the entire defect detection process has high reliability and can provide accurate feedback for the quality assessment of bipolar plates. This not only improves the accuracy of defect recognition but also provides a scientific basis for subsequent quality assessment, ensuring the comprehensiveness and efficiency of the detection process.

[0071] Step S200: According to the defect data and airtightness data, combined with the attention mechanism and fuzzy logic, determine the comprehensive quality score of each bipolar plate.

[0072] Specifically, when determining the comprehensive quality score of each bipolar plate according to the defect data and airtightness data, combining the attention mechanism and fuzzy logic, it includes: extracting features from the defect data and airtightness data of the bipolar plate in different modalities. Conducting weighted analysis on the defect data and airtightness data based on the attention mechanism to determine the key feature information. Dynamically allocating the weights of the defect data and airtightness data based on fuzzy logic, and determining the contribution ratio of each data in the quality assessment according to the confidence level of the data. Determining the comprehensive quality score of the bipolar plate based on the weighted fusion of the defect score, airtightness score, and morphology feature score.

[0073] Specifically, when obtaining the defect score, airtightness score, and morphology feature score of the bipolar plate, it includes: collecting the defect data of the bipolar plate, performing defect recognition based on computer vision algorithms, extracting the size, shape, and distribution characteristics of the defects, and generating the defect score; obtaining the airtightness data and determining the airtightness score based on the leakage rate and air pressure change parameters. Obtaining the surface morphology information of the bipolar plate based on three-dimensional imaging technology, and determining the morphology feature score based on the surface flatness and deformation degree.

[0074] It can be seen that the extraction of defect data is achieved through computer vision algorithms, which can automatically identify defects in images and extract relevant size, morphology and distribution characteristics. This process can effectively obtain detailed information about defects from images and provide a basis for subsequent defect scoring. The acquisition of airtightness data is achieved by analyzing the leakage rate and air pressure change parameters, ensuring the comprehensiveness of the airtightness test and reflecting the airtightness performance of the bipolar plate in actual use. The morphological feature scoring obtains the surface morphology data of the bipolar plate through three-dimensional imaging technology, and ensures that the geometric quality of the bipolar plate is fully quantified based on the evaluation of surface flatness and deformation degree. Secondly, the defect data and airtightness data are weightedly analyzed through the attention mechanism. The attention mechanism can highlight key features and automatically adjust the influence of each data item on the final score by assigning different importance weights to different data items. This technology can identify the most meaningful features from a variety of information, focus on those parameters that are most relevant to quality assessment, and optimize data utilization efficiency. Combined with fuzzy logic, this process further enhances the ability to handle fuzzy and uncertain data. Fuzzy logic can dynamically assign weights to defect and airtightness data, and adjust them in each evaluation link based on the confidence of the data, so that the final quality score is more in line with the uncertainty and variability in the actual detection process. Finally, the comprehensive scoring method based on weighted fusion can integrate data from different modalities (such as defect scores, airtightness scores, and morphological feature scores). Each score is calculated independently based on different physical parameters, and after weighted fusion, a comprehensive quality score for the bipolar plate is generated. This process ensures that the actual contribution of each data in the quality assessment of the bipolar plate can be fully reflected in the final score, making the quality score more accurate and comprehensive. Through this multi-dimensional comprehensive analysis, the test can comprehensively evaluate the quality of the bipolar plate and ensure its reliability in various performance aspects.

[0075] It is understandable that by combining defect data and airtightness data, and integrating various technical means such as computer vision and three-dimensional imaging technology, it is possible to comprehensively evaluate the defect situation, airtightness performance, and morphological characteristics of bipolar plates. Traditional detection methods often focus on a single indicator, while the comprehensive use of multiple data sources can more comprehensively and accurately reflect the overall quality of bipolar plates, avoiding the one-sidedness and deficiencies of traditional methods and improving the reliability and accuracy of detection. Secondly, the use of attention mechanism and fuzzy logic further enhances the intelligent level of data processing. The attention mechanism can perform weighted analysis on data of different modalities, automatically identify key feature information, so that important defect features or airtightness parameters get greater weights in the quality score, thus effectively improving the pertinence and effectiveness of the detection results. Fuzzy logic can dynamically adjust the weights according to the uncertainty and confidence of the data, enhancing the adaptability of the system in the face of different working conditions and detection environments, ensuring that the evaluation results are more in line with the actual situation, and improving the robustness and flexibility of the detection process. Finally, the application of the comprehensive scoring mechanism brings a multi-dimensional comprehensive evaluation effect. By weighted fusion of defect scores, airtightness scores, and morphological feature scores, the advantages and information of each item of data can be maximally integrated together to obtain a more comprehensive and accurate quality score. This method can not only evaluate each performance separately, but also consider the relationship between them during the comprehensive evaluation, further improving the comprehensiveness and reliability of the quality detection results, and helping to provide data support for the subsequent optimization and quality control of the bipolar plates for hydrogen fuel cells.

[0076] Step S300: Determine the defect severity classification according to the relationship between the comprehensive quality score and the preset comprehensive quality score configured.

[0077] Specifically, when determining the defect severity classification according to the relationship between the comprehensive quality score and the preset comprehensive quality score configured, it includes: determining the defect severity classification according to the relationship between the comprehensive quality score and the first preset comprehensive quality score and the second preset comprehensive quality score configured: when the comprehensive quality score is lower than the first preset comprehensive quality score, it is determined that the defect severity classification is a high level. When the comprehensive quality score is higher than or equal to the first preset comprehensive quality score and lower than the second preset comprehensive quality score, it is determined that the defect severity classification is a medium level. When the comprehensive quality score is higher than or equal to the second preset comprehensive quality score, it is determined that the defect severity classification is a low level. Among them, the first preset comprehensive quality score is less than the second preset comprehensive quality score, and the defect severity classification from high to low is high level, medium level, and low level in turn.

[0078] Step S400: Evaluate the confidence of the bipolar plate classification after the defect severity classification based on the Monte Carlo Dropout method and generate an output report.

[0079] Specifically, when evaluating the confidence of the bipolar plate grading after defect severity grading based on the Monte Carlo Dropout method and generating an output report, it includes: performing multiple forward propagations during the inference stage based on the Monte Carlo Dropout method to obtain multiple prediction results of the defect grading. Obtaining the probability distribution characteristics of the defect grading prediction results, and determining the grading confidence based on uncertainty measurement methods such as entropy value or variance. According to the relationship between the confidence and the preset confidence configured, determining the output result: when the confidence is lower than the preset confidence, it is determined that a secondary evaluation is to be performed on the comprehensive quality score of the bipolar plate. When the confidence is higher than or equal to the preset confidence, it is determined that the output is to be performed according to the defect severity grading of the bipolar plate.

[0080] It can be seen that by using the Monte Carlo Dropout method to handle the prediction uncertainty of the model and evaluating the confidence of the defect severity based on the probability distribution characteristics obtained from multiple forward propagations. In deep learning models, Dropout is a common regularization technique, usually used to prevent overfitting. During the inference stage, by introducing Dropout, some nodes in the neural network are randomly "discarded" during each forward propagation, so that different model outputs can be obtained, thereby simulating the uncertainty of the model. By performing multiple forward propagations, multiple prediction results can be obtained, thus generating a probability distribution for each defect grading result. Secondly, by calculating the probability distribution characteristics of multiple prediction results, an uncertainty measurement can be obtained. These measurement methods, such as entropy value or variance, can quantify the prediction uncertainty. For example, a higher entropy value indicates a greater uncertainty of the model in defect grading, indicating that there are significant differences in the model's determination of a specific defect; while the variance measures the degree of dispersion between different prediction results. Through these measurement methods, a confidence can be assigned to each defect grading result, reflecting the reliability and stability of the model when making defect judgments. Finally, based on the calculated confidence, it is compared with the preset confidence threshold to decide whether a secondary evaluation of the bipolar plate is needed. Specifically, when the confidence is lower than the preset threshold, it indicates that the uncertainty of the model in a certain grading is relatively large, and at this time, further evaluation or adjustment may be required to ensure the reliability of the detection results. When the confidence is higher than or equal to the preset threshold, it means that the prediction of the model has a high certainty and can be directly output. This method helps to improve the decision-making accuracy of the detection results by introducing the evaluation of confidence, especially when dealing with defect grading tasks with high uncertainty.

[0081] It is understandable that the uncertainty of the model is effectively handled by the Monte Carlo Dropout method. In the task of defect severity grading, due to the complexity and diversity of the data, traditional models often produce varying degrees of uncertainty. Through multiple forward propagations, Dropout can provide multiple samples for each prediction result, thus forming a probability distribution. This method enables the model to not only give prediction results but also quantify the uncertainty of these results, thereby enhancing the reliability and stability of defect grading. Through this technology, a more accurate risk assessment of the model prediction results can be carried out, avoiding misjudgments caused by single predictions. Secondly, based on uncertainty measurement methods such as entropy value or variance, the decision-making ability of the model in dealing with complex scenarios can be further enhanced. Entropy value and variance provide effective ways to quantify the uncertainty of model predictions. When there are significant differences in the model's judgment of certain defects, a higher entropy value or a larger variance can alert the decision-making system to conduct further analysis. This method makes the evaluation results not only rely on single predictions but also flexibly handle detection tasks of different complexities through uncertainty measurement methods, ensuring a more credible decision-making process. Finally, by comparing with a preset confidence threshold, the model can adaptively adjust whether a secondary evaluation is needed. This mechanism improves the intelligence and flexibility of the system, can trigger further verification when the confidence is low, and effectively avoids wrong decisions caused by low-confidence predictions. At the same time, when the confidence is higher than the threshold, the system can directly output according to the existing grading results, improving the overall detection efficiency. This mechanism optimizes the detection process, improves the overall accuracy and efficiency of bipolar plate detection, and can provide a more accurate basis for the final quality assessment, especially when facing complex defects or data.

[0082] In the above embodiments, through multi-scale feature fusion, the system can extract key defect information from different image data, so as to identify complex defects including hidden cracks, minute depressions, etc. In addition, the combination of airtightness testing and image detection makes the detection more comprehensive. It can not only identify surface defects, but also accurately evaluate the integrity of the internal structure through airtightness testing. The synergistic effect of such multi-modal data significantly improves the accuracy and comprehensiveness of detection, and avoids the problems of missed detection and misdetection that may occur in traditional single detection methods. In addition, by combining the attention mechanism and fuzzy logic, defect data and airtightness data can be processed more intelligently, dynamically adjusting the weights of different data sources to generate a more accurate comprehensive quality score. This mechanism can effectively cope with changes under different working conditions and improve the adaptive ability of the detection system. At the same time, based on the Monte Carlo Dropout method to evaluate the confidence of the defect severity, the uncertainty of the detection results can be quantified, providing a reliable basis for the final decision-making and ensuring the accuracy of the grading results. Finally, the output quality report provides more objective and scientific data support for production manufacturing and quality control, optimizes the decision-making in the detection process, and improves the overall quality assurance ability of the bipolar plate.

[0083] In another preferred embodiment based on the above embodiments, as Figure 2 shown, this embodiment provides a detection system for a hydrogen fuel cell flow guiding bipolar plate, including:

[0084] A composite optical imaging unit, configured with an annular LED array light source, a near-infrared imaging module, and a structured light module. The composite optical imaging unit is configured to perform defect identification on the collected image data based on a multi-scale feature fusion network.

[0085] A multi-station airtightness testing device, configured to perform airtightness detection on the bipolar plate.

[0086] An evaluation module, configured to determine the comprehensive quality score of each bipolar plate according to the defect data and the airtightness data, in combination with the attention mechanism and fuzzy logic. The evaluation module is also configured to determine the defect severity grading according to the relationship between the comprehensive quality score and the preset comprehensive quality score configured.

[0087] An output module, configured to evaluate the confidence of the bipolar plate grading after defect severity grading based on the Monte Carlo Dropout method and generate an output report.

[0088] It can be understood that the detection system and method for a hydrogen fuel cell flow guiding bipolar plate in the above embodiments of the present invention have the same beneficial effects and will not be elaborated here.

[0089] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0090] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0091] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: still modifications or equivalent replacements can be made to the specific embodiments of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. A method for detecting a flow guide bipolar plate of a hydrogen fuel cell, characterized in that: include: Defect recognition is performed on the collected image data based on a multi-scale feature fusion network, and airtightness testing of bipolar plates is performed based on a multi-station airtightness testing device; Based on the defect data and air tightness data, the comprehensive quality score of each bipolar plate is determined by combining the attention mechanism and fuzzy logic; Determine the defect severity classification based on the relationship between the comprehensive quality score and the configured preset comprehensive quality score; The confidence of bipolar plate classification after defect severity classification is evaluated based on the Monte Carlo Dropout method, and an output report is generated.

2. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 1, characterized in that: When defect recognition is performed on the collected image data based on the multi-scale feature fusion network, it includes: Preprocess images collected under different lighting conditions to enhance image contrast and edge information; Based on the multi-scale feature extraction technology, the local detail information and global structure information of different receptive fields in the pre-processed image are extracted, and weighted fusion is performed according to the eliminated multi-scale features to obtain the comprehensive features of the processed image; According to the comprehensive features of the fused image, the characteristic responses of each area in the image are compared by threshold to determine whether the bipolar plate has defects.

3. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 2, characterized in that: When preprocessing images collected under different lighting conditions, it includes: Obtaining the image contrast of each area in the collected image, and obtaining the image contrast mean of each area, and comparing the image contrast mean with the contrast of each image to obtain images of each area whose contrast is lower than the image contrast mean; For each area image with contrast lower than the mean value of the image, brightness normalization is performed based on histogram equalization; The image information after brightness normalization is obtained, and the image is denoised based on non-local mean filtering.

4. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 3, characterized in that: When obtaining the comprehensive features of the processed image, it includes: Multi-scale feature extraction technology is used on the preprocessed image to simultaneously extract local detail information and global structure information in the image based on a number of preset receptive fields of different sizes; The extracted multi-scale features are weighted and fused based on the preset weights to generate comprehensive image features, where: The weights of preset scale features are set according to historical data, and the extracted scale features are matched with the preset scale features to obtain the weights of the extracted scale features. The local details and global structure information are weightedly spliced ​​according to the weights to generate comprehensive features.

5. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 4, characterized in that: According to the comprehensive features of the fused image, the threshold value comparison is performed on the feature response of each area in the image to determine whether the bipolar plate has defects, including: Based on historical data and experimental samples, establish the correlation between detection indicators and defect responses; Obtain the distance metric between each correlation formula, and iteratively cluster each correlation formula according to the distance metric; Based on the lonely forest tree algorithm, the outliers of each cluster association in the clustering result are obtained, and the normal features and defect features in the same area are determined according to the relationship between the outliers and the preset outliers; According to the feature vector between the normal feature and the defect feature, a defect determination threshold is determined; Obtain the feature vector of the feature response of the region in the image, and determine whether there is a defect in the region in the image based on the relationship between the feature vector of the feature response and the defect judgment threshold: When the characteristic vector of the characteristic response is lower than the defect judgment threshold, it is determined that there is no defect in the area in the image; When the feature vector of the feature response is higher than or equal to the defect judgment threshold, it is determined that there is a defect in the area in the image.

6. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 1, characterized in that: Based on the defect data and air tightness data, combined with the attention mechanism and fuzzy logic, the comprehensive quality score of each bipolar plate is determined, including: For the defect data and air tightness data of the bipolar plate, feature extraction is performed on the data of different modes; Perform weighted analysis on defect data and airtightness data based on the attention mechanism to determine key feature information; Dynamically allocate the weights of defect data and airtightness data based on fuzzy logic, and determine the contribution ratio of each data in quality assessment according to the confidence level of the data; The comprehensive quality score of the bipolar plate is determined based on the weighted fusion defect score, air tightness score and morphology feature score.

7. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 6, characterized in that: When obtaining the defect score, air tightness score and morphological feature score of the bipolar plate, it includes: Collect defect data of bipolar plates, identify defects based on computer vision algorithms, extract defect size, shape, and distribution characteristics, and generate defect scores; Obtain air tightness data and determine the air tightness score based on leakage rate and air pressure change parameters; The surface morphology information of the bipolar plate is obtained based on three-dimensional imaging technology, and the morphology feature score is determined based on the surface flatness and deformation degree.

8. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 7, characterized in that: Determine the defect severity level based on the relationship between the comprehensive quality score and the configured preset comprehensive quality score, including: The defect severity level is determined based on the relationship between the comprehensive quality score and the configured first preset comprehensive quality score and second preset comprehensive quality score: When the comprehensive quality score is lower than the first preset comprehensive quality score, the defect severity level is determined to be a high level; When the comprehensive quality score is higher than or equal to the first preset comprehensive quality score, and the comprehensive quality score is lower than the second preset comprehensive quality score, the defect severity level is determined to be medium; When the comprehensive quality score is higher than or equal to the second preset comprehensive quality score, the defect severity level is determined to be a low level; Among them, the first preset comprehensive quality score is less than the second preset comprehensive quality score, and the defect severity levels are graded from high to low as high level, medium level and low level.

9. The detection method for a hydrogen fuel cell flow guide bipolar plate according to claim 1, characterized in that: The confidence level of the bipolar plate classification after defect severity classification is evaluated based on the Monte Carlo Dropout method, and the output report is generated, including: Based on the Monte Carlo Dropout method, multiple forward propagations are performed in the inference stage to obtain multiple prediction results of defect classification; Obtain the probability distribution characteristics of the defect classification prediction results, and determine the classification confidence based on uncertainty measurement methods such as entropy or variance; The output result is determined based on the relationship between the confidence level and the configured preset confidence level: When the confidence level is lower than the preset confidence level, it is determined that a secondary evaluation is performed on the comprehensive quality score of the bipolar plate; When the confidence level is higher than or equal to the preset confidence level, it is determined that the output is graded according to the severity of the defects of the bipolar plate.

10. A detection system for a hydrogen fuel cell flow guide bipolar plate, applicable to a detection method for a hydrogen fuel cell flow guide bipolar plate according to any one of claims 1 to 9, characterized in that: include: A composite optical imaging unit, configured with a ring-shaped LED array light source, a near-infrared imaging module and a structured light module, and the composite optical imaging unit is configured to perform defect recognition on the collected image data based on a multi-scale feature fusion network; A multi-station airtightness testing device configured to perform airtightness testing on bipolar plates; An evaluation module is configured to determine a comprehensive quality score of each bipolar plate based on the defect data and the air tightness data in combination with an attention mechanism and fuzzy logic; the evaluation module is also configured to determine a defect severity classification based on a relationship between the comprehensive quality score and a configured preset comprehensive quality score; The output module is configured to evaluate the confidence of the bipolar plate classification after defect severity classification based on the Monte Carlo Dropout method and generate an output report.

Images