A water distribution real-time imaging detection method for fuel stack tilting test

Abstract

The present application relates to the field of fuel cell, specifically to a water distribution real-time imaging detection method for fuel cell stack tilt test, which comprises the following steps: S1, debugging the six-degree-of-freedom motion platform; deploying a high-precision synchronization system; implementing air tightness guarantee measures; S2, operating a high-speed imaging system; performing real-time image processing to realize sub-pixel alignment; taking quality control measures; S3, performing feature extraction and modeling, outputting a binary image, calculating water film thickness, and calculating the correlation coefficient of voltage sudden drop and water accumulation area; S4, performing system verification and calibration, calculating comprehensive uncertainty, water film thickness ± 5 mu m (k = 2), migration speed ± 0.2 mm / s (k = 2). The present application can clearly show the spatio-temporal evolution rule of water film thickness when the tilt angle continuously changes, provide the correlation mechanism of millisecond water migration process and voltage fluctuation, and provide the quantitative criterion required for water management optimization.

Description

Technical Field

[0001] This invention relates to the field of fuel cell technology, and in particular to a real-time imaging detection method for water distribution in fuel cell stack tilting tests. Background Technology

[0002] In practical applications such as automotive and aviation, proton exchange membrane fuel cell stacks (PEMFCs) are frequently in a dynamic tilting condition (0°-90°). During aircraft takeoff (a 30° tilt lasting 90 seconds), the water content of the membrane electrode assembly can change by up to 40%, and current technologies cannot capture this transient process. This dynamic water migration behavior directly affects gas transport efficiency and membrane electrode wetting status, but there is currently no publicly available solution that can simultaneously acquire water distribution images and stack impedance data at a sampling frequency >5Hz.

[0003] Chinese patent CN118090783A discloses an X-ray spectroscopy testing system suitable for fuel cell operating conditions. This system achieves simultaneous testing of temperature, humidity, voltage, and absorption spectra. It utilizes a self-developed additive to ensure stability under high current conditions. A data acquisition and processing system enables rapid data processing and feedback, allowing for timely monitoring of catalyst structural evolution under actual fuel cell operating conditions. This testing system is not limited to fuel cell operating condition measurements but can be extended to XAFS measurements of other electrochemical conversion processes (including electrocatalytic carbon dioxide reduction, oxygen reduction, and methanol oxidation). Furthermore, as a general method, this approach can be extended to synchrotron radiation characterization, including diffraction and scattering.

[0004] However, the above-mentioned publicly available solutions have the following shortcomings: they use a fixed-angle sample stage, which cannot simulate dynamic tilting conditions; they rely on static X-ray microscopy (<1fps), which cannot capture the water flow process; and they only provide structural morphology data without establishing a correlation with performance parameters. Summary of the Invention

[0005] The purpose of this invention is to address the problems in the background technology that static testing cannot simulate dynamic tilting engineering, have insufficient time resolution, and have a single data dimension, and to propose a real-time imaging detection method for water distribution in fuel cell stack tilting tests.

[0006] The technical solution of the present invention: a real-time imaging detection method for water distribution in fuel cell stack tilting tests, comprising the following steps:

[0007] S1. Debug the six-degree-of-freedom motion platform, calibrate the six-axis position using a laser tracker, and verify the range of motion of each joint; deploy a high-precision synchronization system, use an oscilloscope to test the equipment trigger delay, and verify dynamic accuracy using a standard tilting plate; implement airtightness assurance measures.

[0008] S2. Operate the high-speed imaging system, use an aluminum stepped wedge for energy spectrum calibration to obtain the optimal contrast ratio, and photograph a rotating resolution target to verify the 5μm line pair resolution capability at 100fps; perform real-time image processing to achieve sub-pixel level alignment; periodically perform point spread function detection and monitor the signal-to-noise ratio in real time.

[0009] S3. Perform feature extraction and modeling, use a deep learning network to segment the flow channel region and output a binary image, calculate the water film thickness; calculate the water front movement velocity, establish the relationship curve between tilt angle and velocity; calculate the correlation coefficient between voltage drop and water accumulation area, where the coefficient of determination R² > 0.9;

[0010] S4. Perform system verification and calibration. Use a standard water film of known thickness for static calibration to establish a grayscale-thickness lookup table. Track the movement trajectory of water droplets under tilted conditions and compare it with the results of high-speed camera. Calculate the overall uncertainty.

[0011] Preferably, in S1, the laser tracker is placed in a suitable position so that it can accurately measure the position information of each axis of the six-degree-of-freedom motion platform. The laser tracker emits a laser beam to obtain the position coordinates of specific target points on each axis. After multiple measurements and data processing, high-precision calibration of the six-axis position is achieved.

[0012] Preferably, in S1, the airtightness protection measures are implemented by using a magnetic fluid rotary sealing interface. During installation, the concentricity of the magnetic fluid rotary sealing interface and the rotating component is ensured. A flexible graphite composite sealing ring is configured. During installation, a flexible graphite composite sealing ring of appropriate specifications is selected according to the size and shape of the sealing part, and it is ensured that it is installed tightly without wrinkles or gaps.

[0013] Preferably, in S2, the resolution capability of 5μm line pairs at 100fps is verified by shooting a rotating resolution target. The rotating resolution target is mounted on a high-speed rotating device and rotated at a certain speed within the field of view of the imaging system. The imaging system shoots the rotating resolution target at a frame rate of 100fps to acquire a series of images. By analyzing the images, it is determined whether the imaging system can clearly distinguish 5μm line pairs, thereby verifying the dynamic resolution of the imaging system.

[0014] Preferably, in S3, a deep learning network is used to segment the flow channel region. First, a deep learning network model suitable for fuel cell stack image segmentation is constructed. A large amount of image data containing fuel cell stack flow channels and water distribution is collected and labeled. The flow channel region and non-flow channel region are classified. The labeled data is used to train the deep learning network. By continuously adjusting the network parameters, the network can accurately segment the flow channel region and output a binarized image, requiring a segmentation accuracy of 99.2%.

[0015] Preferably, the water film thickness is calculated based on Beer-Lambert's law. Given the attenuation coefficient of water, the water film thickness is calculated by measuring the intensity change of X-rays before and after passing through the water film using a formula. In the actual calculation process, the grayscale value of the acquired X-ray image is analyzed, and the grayscale value is converted into an X-ray intensity value. Combined with the known water attenuation coefficient, the water film thickness is calculated.

[0016] Preferably, the overall uncertainty of the water film thickness is ±5 μm, which includes a factor k=2, representing the uncertainty range of the water film thickness measurement result at a certain confidence level. The overall uncertainty of the water film thickness is obtained by analyzing and synthesizing various error sources. The overall uncertainty of the migration velocity is ±0.2 mm / s, which includes a factor k=2, and is obtained by analyzing and synthesizing various error sources that affect the migration velocity measurement.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects:

[0018] 1. It can clearly define the spatiotemporal evolution law of water film thickness when the tilt angle changes continuously;

[0019] 2. A correlation mechanism between millisecond-level water migration processes and voltage fluctuations is provided;

[0020] 3. It provides quantitative criteria for optimizing water management. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of a structure according to an embodiment of the present invention. Detailed Implementation

[0022] Example 1, as Figure 1 As shown, the present invention proposes a real-time imaging detection method for water distribution in fuel cell stack tilting tests, comprising the following steps:

[0023] S1. Debug the six-degree-of-freedom motion platform, calibrate the six-axis position using a laser tracker, and verify the range of motion of each joint; deploy a high-precision synchronization system, integrate an inertial measurement unit to acquire tilt angles in real time, configure a μs-level synchronization clock to coordinate the working sequence of each device, establish a time synchronization mechanism for the data acquisition system, use an oscilloscope to detect the device trigger delay, and perform dynamic accuracy verification using a standard tilting plate; implement airtightness assurance measures.

[0024] S2. Operate the high-speed imaging system, use an aluminum stepped wedge for energy spectrum calibration, optimize kV / mA parameters to obtain the best contrast ratio, and photograph a rotating resolution target to verify the 5μm line pair resolution capability at 100fps; perform real-time image processing, load pre-calibrated correction parameters, calculate the transformation matrix based on the real-time tilt angle, apply perspective transformation to eliminate geometric distortion, and achieve sub-pixel level alignment; take quality control measures, periodically perform point spread function detection, and monitor the signal-to-noise ratio in real time;

[0025] S3. Perform feature extraction and modeling, use a deep learning network to segment the flow channel region and output a binarized image, calculate the water film thickness; construct a four-dimensional data matrix (X,Y,Z,Time) for spatiotemporal data fusion, set the voxel size to 50×50×50μm³, calculate the water front movement velocity, and establish the relationship curve between tilt angle and velocity; perform multi-parameter synchronization: X-ray image (100Hz, hardware trigger), voltage (10kHz, clock synchronization), impedance (1kHz, interpolation alignment), set a warning threshold for local water accumulation rate >35%, calculate the correlation coefficient between voltage drop and water accumulation area, and the determination coefficient R² of the correlation coefficient is >0.9;

[0026] S4. Perform system verification and calibration. Use a standard water film of known thickness for static calibration to establish a grayscale-thickness lookup table. Track the water droplet motion trajectory under tilt and compare it with the results of the high-speed camera. Identify error sources: angle measurement error ±0.1°, time synchronization error ±5μs, thickness calculation error ±3μm. Calculate the comprehensive uncertainty: water film thickness ±5μm, which includes a factor k=2, migration speed ±0.2mm / s, which includes a factor k=2.

[0027] Example 2, as Figure 1 As shown, the present invention proposes a real-time imaging detection method for water distribution in a fuel stack tilting test. Compared with Embodiment 1, this embodiment provides a detailed description of S1.

[0028] 1. Platform mechanical calibration:

[0029] A laser tracker is used to precisely calibrate the six-axis position. Specifically, the laser tracker is placed in a suitable position to accurately measure the position information of each axis of the six-degree-of-freedom motion platform. By emitting laser beams from the laser tracker, the position coordinates of specific target points on each axis are acquired. Through multiple measurements and data processing, high-precision calibration of the six-axis position is achieved.

[0030] The range of motion of each joint was verified, requiring a pitch angle of ±90° and a roll angle of ±30°. During the verification process, the operation of the joint motors on the control platform was used to move the joints to their limit positions in both positive and negative directions. Angle sensors were used to monitor the rotation angle of the joints in real time to ensure that they met the specified range of motion requirements.

[0031] 2. Deployment of a high-precision synchronization system:

[0032] Multi-sensor networking:

[0033] An integrated inertial measurement unit (IMU) acquires tilt angle information in real time through its internal sensors such as accelerometers and gyroscopes. The IMU is connected to the motion platform control system, transmitting the acquired tilt angle data to the control system at high speed, providing crucial data support for coordinating the timing of subsequent equipment operations.

[0034] A μs-level synchronization clock is configured, which serves as the time reference for the entire system, ensuring that all devices operate on the same time scale. The synchronization clock signal is distributed to each data acquisition device and imaging system via wired or wireless communication, enabling triggering, data acquisition, and other operations of each device to be performed under precise time synchronization.

[0035] A time synchronization mechanism is established for the data acquisition system. By precisely calibrating and synchronizing the timestamps of each device, the consistency of data collected by different devices in terms of time is ensured. For example, before data acquisition, the time of all devices is initialized to keep the time error with the synchronization clock within a minimal range. During the acquisition process, the deviation of the timestamps of each device is monitored in real time and dynamically corrected using software algorithms.

[0036] Timing verification:

[0037] Use an oscilloscope to test the device's trigger delay, which must be less than 10μs. Specifically, connect the oscilloscope to both the device's trigger signal output and receiver. When the device receives the trigger signal, the oscilloscope records the time interval between the signal transmission and reception. By taking the average of multiple measurements, the device's trigger delay is ensured to meet the requirement.

[0038] Dynamic accuracy verification was performed using a standard tilting plate. The standard tilting plate was mounted on a six-degree-of-freedom motion platform, which moved according to a preset tilt angle and speed. Simultaneously, high-precision measuring equipment (such as a laser interferometer) was used to measure the actual tilt angle of the standard tilting plate. The measurement results were compared with the tilt angle set by the motion platform control system to verify whether the platform's accuracy during dynamic motion met the requirements.

[0039] 3. Implementation of airtightness assurance:

[0040] Fixture sealing test:

[0041] Helium mass spectrometry is used to detect leak rates, which must be less than 0.5%. The specific procedure involves filling the sealed space of the fixture with helium gas and using a helium mass spectrometer leak detector to measure the concentration of helium in the surrounding environment. If the leak rate exceeds the specified value, the leak point is located by applying helium mass spectrometry leak detection fluid, and then sealed and repaired.

[0042] A 200 kPa pressure holding test is required, with a pressure drop of less than 3% over 30 minutes. During the test, the pressure is increased to 200 kPa using a pressure pump, then the pressure source is shut off, and a high-precision pressure sensor is used to monitor pressure changes in real time. If the pressure drop exceeds the specified range, check the installation of the seals, the flatness of the sealing surface, etc., and take appropriate measures to improve the situation.

[0043] Dynamic sealing solution:

[0044] A magnetic fluid rotary seal interface is adopted, which utilizes the sealing properties of magnetic fluid under the action of a magnetic field to achieve dynamic sealing of rotating components. During installation, ensure the concentricity of the magnetic fluid rotary seal interface and the rotating components, and optimize the sealing effect by adjusting parameters such as magnetic field strength.

[0045] A flexible graphite composite sealing ring is used, which has good flexibility and sealing performance. During installation, select the appropriate size of the flexible graphite composite sealing ring according to the size and shape of the sealing part, and ensure that it is installed tightly without wrinkles or gaps.

[0046] Example 3, as Figure 1 As shown, the present invention proposes a real-time imaging detection method for water distribution in a fuel stack tilting test. Compared with Embodiment 1, this embodiment provides a detailed description of S2.

[0047] The operation procedure for the high-speed imaging system is as follows:

[0048] 1. Imaging parameter optimization:

[0049] Energy spectral calibration:

[0050] KV / mA parameter optimization was performed using an aluminum stepped wedge. The aluminum stepped wedge was placed between the X-ray source and the imaging detector. By changing the tube voltage (kV) and tube current (mA) of the X-ray source, images of the aluminum stepped wedge under different parameters were acquired. The grayscale value changes of different thickness regions of the aluminum stepped wedge in the images were analyzed, and the kV / mA parameter combination that achieved the optimal contrast ratio between water and material (water / material > 15%) was selected.

[0051] By analyzing images of aluminum stepped wedges of different thicknesses, the attenuation characteristics of the material under X-ray irradiation of different energies were determined, thereby optimizing imaging parameters and improving the contrast between the target object (water) and the background material in the image.

[0052] Dynamic resolution test:

[0053] The dynamic resolution capability of the imaging system was verified by photographing a rotating resolution target at 100fps. The rotating resolution target was mounted on a high-speed rotating device and rotated at a certain speed within the imaging system's field of view. The imaging system captured a series of images of the rotating resolution target at 100fps. Analysis of the images determined whether the imaging system could clearly distinguish 5μm line pairs, thus verifying the dynamic resolution of the imaging system.

[0054] 2. Real-time image processing:

[0055] Distortion correction:

[0056] Pre-calibrated correction parameters are loaded; these parameters are obtained through extensive calibration experiments on the imaging system. In these experiments, standard checkerboard and other calibration objects are used to acquire images at different positions and angles. Image processing algorithms are then used to calculate the imaging system's distortion parameters, such as radial distortion and tangential distortion.

[0057] The transformation matrix is ​​calculated based on the real-time tilt angle. Using the tilt angle information acquired in real time by the IMU, combined with the intrinsic and extrinsic parameters of the imaging system, the transformation matrix used to correct image distortion is calculated using coordinate transformation formulas.

[0058] Perspective transformation is applied to eliminate geometric distortion. The acquired original image is subjected to perspective transformation using a calculated transformation matrix, restoring the objects in the image to their correct geometric shapes and eliminating geometric distortion caused by changes in the relative positions of the imaging system and the objects.

[0059] Motion artifact removal:

[0060] Perspective transformation is used to eliminate geometric distortion, based on the same principle as above. By eliminating geometric distortion, image blurring and deformation caused by object motion and changes in the relative position of the imaging system are reduced.

[0061] Achieve sub-pixel-level alignment (accuracy of 0.1px). Utilize image matching algorithms, such as feature-point-based matching algorithms (SIFT, SURF, etc.) or region-based matching algorithms, to accurately match adjacent frames in an image sequence after geometric distortion removal. Through sub-pixel-level interpolation algorithms, the matching accuracy is improved to 0.1px, thereby effectively eliminating motion artifacts.

[0062] 3. Quality control measures:

[0063] Regular point spread function (PSF) testing is performed. Images of the imaging system are acquired using a point light source or a small object as the test target, and the PSF is calculated using image processing algorithms. By analyzing the shape and parameters of the PSF, the imaging quality of the imaging system, such as resolution and contrast transfer function, is evaluated.

[0064] Real-time monitoring of signal-to-noise ratio (SNR>20dB). During imaging, the acquired image data is analyzed to calculate the image's SNR. When the SNR falls below 20dB, imaging parameters are automatically adjusted, such as increasing the X-ray source tube current and optimizing the detector gain, to improve the image's SNR.

[0065] Example 4, as Figure 1 As shown, the present invention proposes a real-time imaging detection method for water distribution in a fuel stack tilting test. Compared with Embodiment 1, this embodiment provides a detailed description of S3.

[0066] The detailed steps of feature extraction and modeling are as follows:

[0067] 1. Water film thickness calculation process:

[0068] Image segmentation:

[0069] Deep learning networks are used to segment the flow channel region. First, a deep learning network model suitable for fuel cell stack image segmentation, such as the U-Net model based on convolutional neural networks (CNN), is constructed. A large amount of image data containing fuel cell stack flow channels and water distribution is collected and labeled, classifying flow channel regions and non-flow channel regions. The labeled data is used to train the deep learning network. By continuously adjusting the network parameters, the network can accurately segment the flow channel region and output a binarized image, requiring a segmentation accuracy of 99.2%.

[0070] Thickness inversion:

[0071] The thickness of the water film is calculated based on Beer-Lambert's law. According to Beer-Lambert's law, the attenuation of X-rays by a substance is related to the thickness of the substance and the attenuation coefficient. Given the attenuation coefficient of water, the thickness of the water film is calculated by measuring the intensity change of X-rays before and after passing through the water film, using the formula I = I0e^(-μd) (where I is the intensity of the X-rays after passing through the substance, I0 is the initial X-ray intensity, μ is the attenuation coefficient of the substance, and d is the thickness of the substance).

[0072] In the actual calculation process, grayscale analysis is performed on the acquired X-ray images, and the grayscale values ​​are converted into X-ray intensity values. Combined with the known water attenuation coefficient, the water film thickness is calculated.

[0073] 2. Dynamic modeling implementation:

[0074] Spatiotemporal data fusion:

[0075] A four-dimensional data matrix (X, Y, Z, Time) is constructed, where X, Y, and Z represent spatial coordinates, and Time represents time. During the detection of water distribution in a fuel cell stack, water distribution images at different times and spatial locations are acquired using an imaging system. This information is organized according to spatial coordinates and time order to construct the four-dimensional data matrix. The voxel size is set to 50×50×50μm³ to ensure a balance between spatial resolution and data processing volume.

[0076] Migration pattern analysis:

[0077] Calculate the velocity (mm / s) of the water front. Identify the position of the water front by analyzing water distribution images at different times. Calculate the velocity of the water front using the change in its position and the time interval between adjacent times. For example, if the water front's position is (x1, y1, z1) at time t1 and (x2, y2, z2) at time t2, then the velocity of the water front is v = √[(x2-x1)² + (y2-y1)² + (z2-z1)²] / (t2-t1).

[0078] Establish an angle-velocity curve. Measure the velocity of the water front at different tilt angles of the fuel cell stack. Plot the angle-velocity curve with the tilt angle on the x-axis and the water front velocity on the y-axis to analyze the influence of the tilt angle on the water migration velocity.

[0079] 3. Performance Correlation Analysis:

[0080] Multi-parameter synchronization:

[0081] X-ray images are acquired at a frequency of 100Hz, and a hardware triggering method is used to ensure the accuracy and synchronization of image acquisition. Through a hardware triggering circuit connected to a synchronization clock, the imaging system acquires one frame of X-ray image in each trigger cycle.

[0082] Voltage data is acquired at a frequency of 10kHz, and clock synchronization ensures time consistency with other parameter acquisitions. The voltage acquisition device is connected to a synchronization clock to keep the voltage acquisition time reference consistent with the imaging system and other parameter acquisition systems.

[0083] Impedance data was acquired at a frequency of 1 kHz and synchronized with other parameters using an interpolation alignment method. Because the impedance data acquisition frequency differs from other parameters, an interpolation algorithm was used to interpolate the impedance data along the time axis based on timestamps, aligning it temporally with the X-ray images and voltage data.

[0084] Critical value determination:

[0085] A local water accumulation rate exceeding 35% is set as the warning threshold. The water accumulation rate in local areas within the fuel cell stack flow channels is calculated by analyzing water distribution images. When the local water accumulation rate exceeds 35%, the system issues a warning signal, indicating a potential water accumulation problem.

[0086] Calculate the correlation coefficient (R² > 0.9) between voltage sag and water accumulation area. Monitor voltage changes during fuel cell stack operation. When a voltage sag occurs, analyze the corresponding water accumulation area. Calculate the correlation coefficient between voltage sag and water accumulation area using statistical analysis methods, requiring a correlation coefficient R² greater than 0.9 to establish a quantitative relationship between the two.

[0087] Example 5, such as Figure 1 As shown, the present invention proposes a real-time imaging detection method for water distribution in fuel stack tilting tests. Compared with Embodiment 1, this embodiment provides a detailed description of S4.

[0088] The system verification and calibration are as follows:

[0089] 1. Calibration test:

[0090] Static calibration:

[0091] Using standard water films of known thickness (ranging from 10 to 200 μm), water film standards of different thicknesses were placed within the field of view of the imaging system, and X-ray images were acquired. The grayscale values ​​of water films of different thicknesses in the images were analyzed to establish a grayscale-thickness lookup table. This lookup table allows for the rapid determination of the water film thickness based on the acquired image grayscale values ​​during subsequent actual testing.

[0092] Dynamic verification:

[0093] Tracking the trajectory of a water droplet under tilt. A water droplet is placed at a specific location on a fuel cell stack model. The fuel cell stack is tilted using a six-degree-of-freedom motion platform, and the trajectory of the water droplet during the tilting process is tracked using a high-speed imaging system.

[0094] The tracking results are compared with those from a high-speed camera, with an error requirement of less than 8%. By comparing the two measurement results, the accuracy of this detection system in measuring the trajectory of water droplets under dynamic tilt is verified. If the error exceeds 8%, the system's parameter settings and image processing algorithms are optimized and adjusted.

[0095] 2. Uncertainty Analysis:

[0096] Error source identification:

[0097] The angle measurement error is ±0.1°, mainly due to the accuracy limitations of angle measurement equipment such as IMUs. The angle measurement error can be minimized by calibrating the IMU and performing error compensation.

[0098] The time synchronization error is ±5μs, caused by factors such as the accuracy of the synchronization clock and signal transmission delay. The time synchronization error can be reduced by optimizing the synchronization clock system and minimizing signal transmission line losses.

[0099] The thickness calculation error is ±3μm, mainly caused by approximations in the Beer-Lambert law calculation process and errors in image grayscale measurement. This error can be reduced by improving the accuracy of the imaging system and optimizing the thickness calculation algorithm.

[0100] Overall uncertainty:

[0101] The overall uncertainty of the water film thickness is ±5 μm, which includes a factor k=2, representing the range of uncertainty in the water film thickness measurement at a certain confidence level. The overall uncertainty of the water film thickness is obtained by analyzing and synthesizing various error sources.

[0102] The overall uncertainty of the migration speed is ±0.2 mm / s, which includes a factor k=2. This uncertainty is obtained by analyzing and synthesizing various error sources affecting the migration speed measurement. In practical applications, the reliability of the measurement results is assessed based on the overall uncertainty.

[0103] This technical solution, through the above standardized implementation steps, ensures that the repeatability error of dynamic tilt testing is less than 3%, providing reliable data support for fuel cell water management research.

[0104] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A real-time imaging detection method for water distribution in fuel cell stack tilting tests, characterized in that, Includes the following steps: S1. Debug the six-degree-of-freedom motion platform, calibrate the six-axis position using a laser tracker, and verify the range of motion of each joint; deploy a high-precision synchronization system, use an oscilloscope to test the equipment trigger delay, and verify dynamic accuracy using a standard tilting plate; implement airtightness assurance measures. S2. Operate the high-speed imaging system, use an aluminum stepped wedge for energy spectrum calibration, obtain the optimal contrast ratio, and photograph a rotating resolution target to verify the 5μm line pair resolution capability at 100fps; perform real-time image processing to achieve sub-pixel level alignment. Regularly perform point spread function testing and monitor the signal-to-noise ratio in real time; S3. Perform feature extraction and modeling, use a deep learning network to segment the flow channel region and output a binary image, calculate the water film thickness; calculate the water front movement velocity, establish the relationship curve between tilt angle and velocity; calculate the correlation coefficient between voltage drop and water accumulation area, where the coefficient of determination R² > 0.9; S4. Perform system verification and calibration. Use a standard water film of known thickness for static calibration to establish a grayscale-thickness lookup table. Track the movement trajectory of water droplets under tilted conditions and compare it with the results of high-speed camera. Calculate the overall uncertainty.

2. The real-time imaging detection method for water distribution in fuel cell stack tilting tests according to claim 1, characterized in that, In S1, the laser tracker is placed in a suitable position so that it can accurately measure the position information of each axis of the six-degree-of-freedom motion platform. The laser tracker emits a laser beam to obtain the position coordinates of specific target points on each axis. After multiple measurements and data processing, high-precision calibration of the six-axis position is achieved.

3. The real-time imaging detection method for water distribution in fuel cell stack tilting tests according to claim 1, characterized in that, In S1, the airtightness guarantee measures are implemented by using a magnetic fluid rotary sealing interface. During installation, the concentricity of the magnetic fluid rotary sealing interface and the rotating component is ensured. A flexible graphite composite sealing ring is configured. During installation, a flexible graphite composite sealing ring of appropriate specifications is selected according to the size and shape of the sealing part, and it is ensured that it is installed tightly without wrinkles or gaps.

4. The real-time imaging detection method for water distribution in fuel cell stack tilting tests according to claim 1, characterized in that, In S2, the resolution capability of 5μm line pairs at 100fps is verified by shooting a rotating resolution target. The rotating resolution target is mounted on a high-speed rotating device and rotated at a certain speed within the field of view of the imaging system. The imaging system shoots the rotating resolution target at a frame rate of 100fps to acquire a series of images. By analyzing the images, it is determined whether the imaging system can clearly distinguish 5μm line pairs, thereby verifying the dynamic resolution of the imaging system.

5. The real-time imaging detection method for water distribution in fuel cell stack tilting tests according to claim 1, characterized in that, In S3, a deep learning network is used to segment the flow channel region. First, a deep learning network model suitable for fuel cell stack image segmentation is constructed. A large amount of image data containing fuel cell stack flow channels and water distribution is collected and labeled. Flow channel regions and non-flow channel regions are classified. The labeled data is used to train the deep learning network. By continuously adjusting the network parameters, the network can accurately segment the flow channel region and output a binarized image, requiring a segmentation accuracy of 99.2%.

6. The real-time imaging detection method for water distribution in fuel cell stack tilting tests according to claim 5, characterized in that, The water film thickness is calculated based on Beer-Lambert's law. Given the water attenuation coefficient, the water film thickness is calculated by measuring the intensity change of X-rays before and after passing through the water film. In the actual calculation process, the grayscale values ​​of the acquired X-ray images are analyzed and converted into X-ray intensity values. Combined with the known water attenuation coefficient, the water film thickness is calculated.

7. The real-time imaging detection method for water distribution in fuel cell stack tilting tests according to claim 1, characterized in that, The overall uncertainty of the water film thickness is ±5 μm, which includes a factor k=2, representing the uncertainty range of the water film thickness measurement result at a certain confidence level. The overall uncertainty of the water film thickness is obtained by analyzing and synthesizing various error sources. The overall uncertainty of the migration velocity is ±0.2 mm / s, which includes a factor k=2. It is obtained by analyzing and synthesizing various error sources that affect the migration velocity measurement.

Images