Fuel cell electro-deposition temperature measuring method based on fusion of infrared light and visible light

Abstract

The invention discloses a fuel cell electro-deposition temperature measuring method based on infrared light and visible light fusion. The method comprises the following steps: firstly, obtaining a space conversion relation between visible light and an infrared camera through joint calibration; then synchronously acquiring a dual-spectrum video stream and extracting key frame image pairs with synchronous time; performing image fusion based on the spatial conversion relation and an image matching algorithm to generate a fused image; then under a plurality of preset time scales, carrying out feature temperature value extraction and statistical analysis on a target area in the sequence fusion image to obtain temperature distribution feature data; and finally, synthesizing the whole temperature field of the electric pile, analyzing the characteristic of the temperature field along with time change, and associating to different operation stages of the electric pile. According to the invention, non-contact, full-view-field and millisecond-level dynamic monitoring of the surface temperature of the fuel cell stack is realized, the limitation of traditional single-point contact type temperature measurement is overcome, and accurate data support is provided for thermal management optimization and state monitoring of the stack.

Description

Technical Field

[0001] This invention relates to the field of fuel cell monitoring technology, specifically a method for achieving high-precision reconstruction of the stack temperature field, multi-timescale analysis, and operational status diagnosis through the fusion of infrared and visible light images. Background Technology

[0002] Proton exchange membrane fuel cell systems are complex integrated devices, typically composed of a fuel stack, fuel supply, oxygen supply, thermal management, water management, and power control, with an overall energy conversion efficiency exceeding 50%. The fuel stack is the core component of the fuel cell system and has high requirements for thermal management. Traditional thermocouple temperature measurement methods have the following drawbacks: single-point measurement cannot capture the spatial distribution of the temperature field in the fuel stack; contact measurement interferes with the thermal field distribution; and it is difficult to achieve millisecond-level dynamic response.

[0003] Chinese patent application CN119984518A discloses an integrated detection method and device for tobacco shred speed and drying machine cavity temperature, including acquiring visible light and infrared images of tobacco shreds; preprocessing the visible light and infrared images; fusing the preprocessed visible light and infrared images using a spectral fusion algorithm to generate a fused image; extracting tobacco shred speed information from the fused image using an optical flow method; and obtaining tobacco shred temperature and cavity temperature information from the fused image using a spectral radiation method. However, this prior art is used to detect the temperature of tobacco shreds and is not suitable for measuring the stack temperature of fuel cells.

[0004] The invention application with publication number CN117197197A discloses a method, device, electronic device and readable storage medium for detecting the temperature of a target object. It acquires visible light images and infrared images of the target object, performs target detection on the visible light images, and maps the target detection results to the infrared images to achieve the purpose of accurately distinguishing the target object from the environment in the infrared images. Finally, the temperature of the target object is obtained based on the target detection results in the infrared registration images.

[0005] Existing infrared temperature measurement technology is not well adapted to fuel cell stacks and has not been specifically optimized for the spatial temperature distribution characteristics under long-term operating conditions. This invention utilizes the advantages of non-contact measurement with infrared devices to achieve non-contact measurement of the stack surface temperature, which is of positive significance for monitoring the stack's operating status. Summary of the Invention

[0006] This solution addresses the aforementioned issues through dual-spectrum fusion, innovatively combining infrared temperature sensitivity with visible light spatial details to achieve non-contact, full-field-of-view temperature monitoring.

[0007] The specific technical solution of the present invention is as follows: This invention provides a method for measuring the stack temperature of a fuel cell based on the fusion of infrared and visible light, comprising the following steps: S1. Perform joint calibration of the visible light camera and the infrared camera to obtain the spatial transformation relationship between the two cameras; S2. Based on the spatial transformation relationship and the image registration algorithm based on maximizing mutual information, the keyframe image pairs are fused to generate a fused image; S3. Based on the spatial transformation relationship and image feature matching algorithm, the keyframe image pairs are fused to generate a fused image; S4. At multiple preset time scales, feature temperature values ​​are extracted and statistically analyzed from the target region in the sequence fusion image to generate temperature distribution feature data of the target region at different time scales. S5. Based on the temperature distribution characteristics data of all target areas, synthesize the overall temperature field of the fuel cell stack, analyze its characteristics as a function of time, and correlate it with different operating stages of the fuel cell stack.

[0008] Furthermore, the joint calibration of the visible light camera and the infrared camera in step S1 includes: S11. Use a black and white checkerboard calibration plate with a high emissivity coating or specific temperature control conditions. S12. Simultaneously acquire image sequences of the calibration plate in different poses from the visible light camera and the infrared camera; S13. Use Zhang Zhengyou's calibration algorithm to calculate the internal and external parameters of each camera respectively; S14. Based on the calibration results, obtain the spatial transformation relationship between the two cameras.

[0009] Furthermore, in step S11, the black and white checkerboard calibration board is a customized calibration board, wherein the checkerboard is composed of alternating distributions of high-reflectivity aluminum foil tape and low-emissivity graphite sprayed squares. The width of the checkerboard on the customized calibration board According to the camera's effective focal length Target object feature dimensions and working distance Dynamic adjustment, the adjustment formula is:

[0010] Among them, the width of the chessboard grid The width of the chessboard grid is in mm. This is the camera's effective focal length, in mm. The feature dimensions of the target object are in mm. Working distance, in mm; This is a proportionality coefficient, with a value ranging from 0.8 to 1.2. In step S13, the application of Zhang Zhengyou's calibration algorithm to calculate the internal and external parameters of each camera includes: Construct the camera projection equation: , in, As a scale factor, Two-dimensional pixel coordinates a homogeneous dimensionless vector For focal length With the main point intrinsic parameter matrix ,in , It is the focal length, in mm. Is it pixels? Dimensions in the direction, in mm / pixel. , Is it pixels? Dimensions in the direction, in mm / pixel. These are the pixel coordinates of the image center. Let the external parameter rotation and translation matrix be the one from the world coordinate system to the camera coordinate system. It is a 3×3 matrix, dimensionless. It is a 3×1 vector, with units of mm, that is, a 3×3 matrix extended to the right by a 3×1 matrix; Point in the world coordinate system A homogeneous vector, with coordinates in mm; In step S14, the radial distortion coefficient of the camera is solved using a nonlinear optimization algorithm. and tangential distortion coefficient ; Based on the calibration results of the visible light camera and the infrared camera, The extrinsic parameter matrix of the infrared camera is denoted as The extrinsic parameter matrix of the visible light camera is recorded as follows: The spatial matrix for transforming from the infrared camera coordinate system to the visible light coordinate system Calculated using the following formula:

[0011] in, This represents the rotation matrix from the visible light camera coordinate system to the infrared camera coordinate system. This represents the translation vector from the visible light camera coordinate system to the infrared camera coordinate system.

[0012] in, It is an abbreviation for infraRed. It is an abbreviation for visible light.

[0013] Further, step S2 includes: S21. Fix the visible light camera and the infrared camera in place, and make their optical axes parallel or maintain a stable relative positional relationship. S22. Simultaneously start the two cameras and continuously acquire visible light video stream and infrared video stream at the set frame rate; S23. From the visible light video stream and infrared video stream acquired in step S22, frame-level time synchronization is achieved based on timestamps and using interpolation algorithms. The visible light image and infrared temperature image corresponding to the same timestamp or time point are extracted as a paired keyframe dataset.

[0014] Furthermore, the interpolation algorithm includes linear interpolation or cubic spline interpolation; The acquisition process employs layered synchronous control, with the software layer calibrating the timestamp. When the frame time deviation... When the time is greater than 2ms, the temperature field is reconstructed using motion compensation, and the expression is:

[0015] in, This is the reconstructed infrared temperature field, measured in Kelvin. This is the infrared temperature field of the previous frame, in Kelvin. It is the time difference between frames, in milliseconds, gradient. The time gradient of the temperature field is derived from the optical flow field of the adjacent frame and is expressed in K / ms.

[0016] Further, step S3 includes: S31. Perform contrast enhancement preprocessing on the visible light image and noise reduction preprocessing on the infrared temperature image; S32. Based on the spatial transformation relationship obtained in step S1, perform preliminary spatial alignment on the preprocessed visible light image and infrared temperature image; The spatial alignment includes affine transformation or perspective transformation; S33. Using a cross-modal image registration algorithm based on maximizing mutual information, calculate the mutual information or normalized mutual information of the initially aligned visible light image and infrared temperature image, and use it as a similarity evaluation index for image registration. S34. Based on the mutual information or normalized mutual information, the geometric transformation parameters are iteratively updated through a numerical optimization algorithm to maximize the mutual information or normalized mutual information, thereby obtaining the optimal spatial transformation matrix between the two images. The geometric transformation parameters include translation, rotation, and / or scaling parameters; S35. Perform precise geometric transformation on the infrared temperature image using the optimal spatial transformation matrix; S36. Employ pixel-level or region-level fusion strategies to fuse the registered infrared temperature information with the visible light image to generate a fused image.

[0017] Furthermore, in step S31, the contrast enhancement preprocessing of the visible light image employs a contrast-limited adaptive histogram equalization algorithm; the noise reduction preprocessing of the infrared temperature image employs a Gaussian blur algorithm with a standard deviation of [missing information]. The value is 1.5; In step S32, Two-dimensional matrix based on spatial transformation relationships obtained through joint calibration The initial spatial alignment of the infrared temperature image is performed using the following coordinate transformation formula:

[0018] in, This is a two-dimensional matrix representing the spatial transformation relationship between the two cameras obtained through joint calibration. pixel coordinates of a visible light image , a homogeneous dimensionless vector pixel coordinates in an infrared image a homogeneous dimensionless vector; Based on the calibration results of the visible light camera and the infrared camera, the extrinsic parameter matrix of the infrared camera is denoted as follows: The extrinsic parameter matrix of a visible light camera is denoted as... Two-dimensional matrix The calculation formula is as follows:

[0019] in It is the normal vector of the target plane. Let be the distance from the origin of the camera coordinate system to the normal vector direction of the target plane in the reference camera coordinate system. This represents the rotation matrix from the visible light camera coordinate system to the infrared camera coordinate system. This represents the translation vector from the visible light camera coordinate system to the infrared camera coordinate system. and These are the intrinsic parameter matrices for the visible light camera and the infrared camera, respectively. In step S33, a cross-modal image similarity measurement method based on maximizing mutual information is adopted. By constructing a joint histogram of the visible light image and the infrared image, the normalized mutual information between the two images is calculated. The normalized mutual information calculation formula is as follows:

[0020] in, and The edge entropy of infrared and visible light images, respectively. Let the joint entropy be ; let the visible light image be . Infrared image is Its 8-bit grayscale value is divided into 0~128 intervals, and a joint distribution histogram is constructed. ,in: It is an image Gray quantized index, It is an image Quantization index, This means that both images fall into the same interval at the same pixel position. The number of times is normalized to obtain the joint probability distribution:

[0021] Edge probability distribution of visible light images:

[0022] Edge distribution of infrared images:

[0023] Edge entropy of visible light images:

[0024] Edge entropy of infrared images:

[0025] Joint entropy:

[0026] In step S34, a numerical optimization algorithm, including but not limited to the Powell method in the open-source software package minimize, is used to iteratively update the geometric transformation parameters, including the translation vector. Rotation angle With scaling factor The mutual information or normalized mutual information is maximized to obtain the optimal spatial transformation matrix between the two images. If the mutual information value decreases during the iteration process, the search is reversed to the parameters of the previous step and continues to avoid local optima. The two-dimensional image transformation matrix is ​​updated according to the registration parameters.

[0027] in, and These represent the translation vectors along the X and Y axes, respectively. In step S36, bilinear interpolation is used to map the infrared image to the visible light coordinate system. The mapping formula is as follows:

[0028] in, This represents the temperature value at the target coordinate point, in Kelvin (K). It is the original infrared image. Line 1 The pixel temperature value of the column, in K. These are the pixel coordinates of the target location. These are the pixel network temperature values ​​along the X and Y axes of the original image, respectively.

[0029] Further, step S4 includes: S41. Under multiple preset time scales, select a series of key frame images within the corresponding time period in the fused image sequence; The multiple time scales include second-level scales, minute-level scales, and / or the entire test process scale; S42. For each keyframe image, calculate the specified characteristic temperature value of the target region within the fuel cell stack; S43. Perform statistical analysis on a series of characteristic temperature values ​​calculated for each time scale and target area, and calculate at least one of the following: average temperature, maximum temperature, minimum temperature, temperature standard deviation, and temperature change rate, or generate a time-temperature distribution map. S44. Integrate the temperature distribution feature data of each target area at multiple time scales to form a temperature distribution feature frame dataset.

[0030] Furthermore, the boundary regions of individual fuel cell stacks are first identified, specifically the outer contour of the cuboid fuel cell stack. The characteristic temperature values ​​of the target region are extracted frame by frame. These characteristic temperature values ​​include spatial uniformity, transient response, and cumulative heat load, as shown in the following formula: Spatial uniformity:

[0031] Transient response:

[0032] Cumulative heat load:

[0033] in, It is the average temperature of space, measured in Kelvin (K). This is a manually specified reference temperature, in Kelvin (K). It is the number of statistical points, spatial uniformity. The unit is T, transient response The unit is K / s, cumulative heat load. Statistics from time Start to Time The accumulated temperature, measured in K·s, The unit is K, and it represents the temperature at time t.

[0034] Further, step S5 includes: S51. Collect temperature distribution characteristic data of all target areas in step S4; S52. Calculate at least one global characteristic temperature parameter that characterizes the overall thermal state of the fuel cell stack. The global characteristic temperature parameter includes the global maximum temperature, the global average temperature, the maximum temperature difference, the temperature uniformity index, or the global temperature distribution pseudo-color spectrum. S53. Analyze the sequence values ​​of the global characteristic temperature parameters at different time scales to generate the overall temperature time-varying characteristic curve or feature vector of the fuel cell stack. S54. Based on the preset stack operation stage division rules and / or synchronously monitored external ambient temperature data; The stack operation phase is characterized by the feature division of the time-varying characteristic curve obtained from the time, load, operation mode, or step S53.

[0035] In measurement applications, the overall temperature time-varying characteristics of the fuel cell stack are correlated with the cold start stage, activation stage, stable operation stage, or shutdown cooling stage. That is, the surface temperature of the fuel cell stack is monitored under different operating conditions (operation stages).

[0036] Furthermore, based on the output temperature distribution feature frame dataset, the overall thermal state model of the fuel cell stack is constructed according to the following process: a pseudo-color mapping model, adjusting the HSV mapping function based on the real-time temperature range:

[0037]

[0038]

[0039] The gradient is calculated using the Sobel operator: , and These represent the temperature gradients in the x and y directions obtained by convolving the temperature image using the Sobel operator, respectively. , They are defined as follows:

[0040]

[0041] , , These represent the hue, saturation, and brightness of pixels in the image, respectively. The maximum temperature in the image. The lowest temperature in the image. The target temperature to be highlighted is in Kelvin (K). This is the standard deviation parameter for temperature distribution, in units of... . It is the temperature field gradient, and the unit is K / pixel.

[0042] The beneficial effects of this invention are: This invention utilizes infrared and visible light dual-spectrum fusion technology to achieve non-contact, full-field-of-view, millisecond-level dynamic monitoring of the surface temperature field of fuel cell stacks. It effectively overcomes the shortcomings of traditional thermocouple single-point measurement and contact interference, and can accurately capture the spatial temperature distribution and transient change patterns of the stack under key operating conditions such as cold start and steady-state operation. This provides key data support for stack thermal management optimization, fault diagnosis, and performance improvement, significantly enhancing the reliability and efficiency of fuel cell system operation. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the checkerboard calibration board used in this application.

[0044] Figure 2 This is a false-color image taken using an infrared camera.

[0045] Figure 3 A color image taken using a visible light lens.

[0046] Figure 4 Grayscale visible light image of the fuel cell stack before registration.

[0047] Figure 5 The image shows the registered fuel cell stack in visible light.

[0048] Figure 6 This is a grayscale display of the infrared image of the fuel cell stack.

[0049] Figure 7 To display a stacked effect of infrared and visible light images using pseudo-color images.

[0050] Figure 8 This is a graph showing the change in surface temperature of the fuel cell stack over time. Detailed Implementation

[0051] 1. Dual-camera joint calibration and data acquisition To achieve pixel-level alignment between visible light and infrared images, a precise mapping relationship between the coordinate systems of the two cameras needs to be established.

[0052] First, a custom checkerboard calibration board with alternating aluminum foil tape and graphite-sprayed squares is used, as shown in the pattern. Figure 1 As shown, Figure 1 The black frame is coated with graphite, while the white frame is covered with aluminum foil tape. The grid width is dynamically adjusted according to the camera's focal length and working distance. Among them, the width of the chessboard square The width of the chessboard grid is in mm. This is the camera's effective focal length, in mm. The feature dimensions of the target object are in mm. Working distance, in mm; This is a proportionality coefficient, with a value ranging from 0.8 to 1.2.

[0053] Secondly, the camera's intrinsic and extrinsic parameters are solved simultaneously using Zhang Zhengyou's calibration algorithm: the projection equation is constructed. ,in, As a scale factor, Two-dimensional pixel coordinates A homogeneous dimensionless vector, where K is the focal length. With the main point intrinsic parameter matrix ,in , It is the focal length, in mm. Is it pixels? Dimensions in the direction, in mm / pixel. , Is it pixels? Dimensions in the direction, in mm / pixel. These are the pixel coordinates of the image center. Let the external parameter rotation and translation matrix be the one from the world coordinate system to the camera coordinate system. It is a 3×3 matrix, dimensionless. It is a 3×1 vector, with units of mm. Point in the world coordinate system The homogeneous coordinates are in mm.

[0054] The radial distortion coefficient of the camera is solved using a nonlinear optimization algorithm. and tangential distortion coefficient Based on the calibration results of the visible light camera and the infrared camera, the extrinsic parameter matrix of the infrared camera is denoted as follows: The extrinsic parameter matrix of a visible light camera is denoted as... The three-dimensional spatial matrix transformed from the infrared camera coordinate system to the visible light coordinate system. Calculated using the following formula:

[0055] in, This represents the rotation matrix from the visible light camera coordinate system to the infrared camera coordinate system. This represents the translation vector from the visible light camera coordinate system to the infrared camera coordinate system.

[0056] The acquisition process employs layered synchronous control, with the software layer calibrating the timestamp. When the frame time deviation... Temperature field reconstruction using motion compensation for >2ms: , in, This is the reconstructed infrared temperature field, measured in Kelvin. This is the infrared temperature field of the previous frame, in Kelvin. It is the time difference between frames, in milliseconds, gradient. The time gradient of the temperature field is derived from the optical flow field of the adjacent frame and is expressed in K / ms.

[0057] 2. Fusion of Infrared and Visible Light Images Image fusion includes three core stages: preprocessing, feature matching, and temperature field fusion.

[0058] In the preprocessing stage, contrast-limited adaptive histogram equalization (CLAHE) is applied to the visible light image to enhance texture details, and Gaussian filtering is applied to the infrared image. Suppress thermal noise.

[0059] During spatial registration, the two-dimensional spatial transformation matrix obtained through joint calibration is used. The infrared image is pre-aligned by performing an affine transformation to eliminate large-scale displacements caused by differences in camera pose. , in, This is a two-dimensional matrix representing the spatial transformation relationship between the two cameras obtained through joint calibration. pixel coordinates of a visible light image , a homogeneous dimensionless vector pixel coordinates in an infrared image a homogeneous dimensionless vector;

[0060] in It is the normal vector of the target plane. Let be the distance from the origin of the camera coordinate system to the normal vector direction of the target plane in the reference camera coordinate system. This represents the rotation matrix from the visible light camera coordinate system to the infrared camera coordinate system. This represents the translation vector from the visible light camera coordinate system to the infrared camera coordinate system. and These are the intrinsic parameter matrices for the visible light camera and the infrared camera, respectively. First, the infrared image is subjected to an affine transformation using the spatial transformation matrix obtained from joint calibration to achieve preliminary spatial alignment:

[0061] To further achieve precise registration, this invention employs a cross-modal image registration algorithm based on maximizing mutual information. On the pre-aligned image, instead of relying on any feature point extraction, the overall image alignment is directly optimized through the following steps: Construct a joint histogram of the visible light image and the infrared image, and calculate the normalized mutual information:

[0062] in and The edge entropy of infrared and visible light images, respectively. Let be the joint entropy.

[0063] Let the visible light image be Infrared image is Its 8-bit grayscale value is divided into 0~128 intervals, and a joint distribution histogram is constructed. ,in: It is an image Gray quantized index, It is an image Quantization index, This means that both images fall into the same interval at the same pixel position. The number of times is normalized to obtain the joint probability distribution:

[0064] Edge probability distribution of visible light images:

[0065] Edge distribution of infrared images:

[0066] Edge entropy of visible light images:

[0067] Edge entropy of infrared images:

[0068] Joint entropy:

[0069] With translation vector Rotation angle With scaling factor To optimize the variables, construct a two-dimensional affine transformation matrix:

[0070] in, and These represent the translation vectors along the X and Y axes, respectively. The above uses numerical optimization algorithms, including but not limited to the Powell method in the minimize package, to continuously update the geometric transformation parameters to maximize the normalized mutual information value, thereby obtaining the optimal spatial transformation matrix between the two images. If the mutual information value decreases during the iteration process, the algorithm will backtrack to the parameters of the previous step and continue searching to avoid local optima.

[0071] During the temperature field fusion process, bilinear interpolation is used to map the infrared image to the visible light coordinate system:

[0072] in, This represents the temperature value at the target coordinate point, in Kelvin (K). The value is the pixel temperature of the original infrared image, in Kelvin (K). (x, y) are the pixel coordinates of the target location. These are the pixel network temperature values ​​along the X and Y axes of the original image, respectively.

[0073] 3. Multi-scale temperature feature extraction The time-varying characteristics of the fuel cell stack temperature field are analyzed on the fused images. First, the boundary regions of individual fuel cell stack units are identified, and temperature statistics for the target region (a single unit or module) are extracted frame by frame. Spatial uniformity:

[0074] Transient response:

[0075] Cumulative heat load:

[0076] in It is the average temperature of space. These are manually specified reference temperatures, all in Kelvin (K). It is the number of statistical points, spatial uniformity. The unit is T, transient response The unit is K / s, cumulative heat load. Statistics from time Start to Time The accumulated temperature, measured in K·s, The unit is K, and it represents the temperature at time t.

[0077] Construct temperature distribution feature frames at multiple time scales (second-level / minute-level / full-scale): capture rapid dynamics (such as cold start shock) at the second-level, analyze steady-state fluctuations (such as load changes) at the minute-level, and evaluate the temperature distribution status at the full-scale.

[0078] 4. Overall temperature characteristics assessment Based on the output temperature distribution feature frame dataset, the overall thermal state model of the fuel cell stack is constructed according to the following procedure: Pseudo-color mapping model Adjust the HSV mapping function based on the real-time temperature range:

[0079]

[0080]

[0081] The gradient is calculated using the Sobel operator: , and These represent the temperature gradients in the x and y directions obtained by convolving the temperature image using the Sobel operator, respectively. , They are defined as follows:

[0082]

[0083] , , These represent the hue, saturation, and brightness of pixels in the image, respectively. The maximum temperature in the image. The lowest temperature in the image. The target temperature to be highlighted is in Kelvin (K). This is the standard deviation parameter for temperature distribution, in units of... . It is the temperature field gradient, and the unit is K / pixel.

[0084] Example 1 Create as Figure 1 The calibration board shown uses aluminum foil tape to fill the white areas and graphite to fill the black areas.

[0085] Taking the top surface of the fuel cell stack as an example, the temperature distribution at various points on the stack surface is monitored, and temperature feature frames are constructed using short-duration monitoring data. Pseudo-color images captured by an infrared camera are shown below. Figure 2 As shown, color images captured using a visible light lens are as follows: Figure 3 As shown, the false-color image captured by the infrared camera uses low color temperature red to represent the highest temperature and high color temperature blue to represent the lowest temperature. In the visible light image, the overall structure of the fuel cell stack is clearly visible. There is no significant difference in viewing angle between the infrared camera image and the visible light image.

[0086] The cross-modal image registration method based on maximizing mutual information proposed in this invention is used to perform a two-dimensional affine transformation on the visible light image to precisely align it with the infrared image. In this embodiment, the translation vector is adjusted... Rotation angle With scaling factor Numerical optimization was performed to maximize the normalized mutual information. The optimal registration parameters obtained from the optimization calculation are as follows: Translation amount: =-4.98 pixels = -3.98 pixels, Rotation angle: =0.88°, Scaling factor: =1.26, A two-dimensional affine transformation matrix is ​​constructed based on the above optimal parameters, and the visible light image is then subjected to the final spatial transformation. The visible light grayscale image before registration is shown below. Figure 4 As shown, the registered visible light image is as follows: Figure 5 As shown, the registered image outline closely matches the infrared image.

[0087] The temperature field information of the three-dimensional surface of the fuel cell stack can be obtained from the monitoring data, and the corresponding grayscale representation of the infrared image is as follows: Figure 6 As shown in the diagram, in the temperature fusion demonstration, the infrared image and the visible light image are mapped to the R and G channels of the RGB image, respectively, and then superimposed on the same fused image to visually display the registration effect, as shown below. Figure 7 As shown.

[0088] Example 2 By capturing surface temperature data of the fuel cell stack during long-term operation, the average operating temperature of the stack surface can be obtained through long-term monitoring at the level of hundreds of seconds, avoiding inaccurate temperature measurements caused by short-term airflow disturbances and current fluctuations.

[0089] Taking the surface temperature of the fuel cell stack during normal operation as an example, the temperature at a fixed point is photographed every 4 seconds to obtain the surface temperature curve of the monitoring point, as shown in the figure. Figure 8 As shown.

[0090] In this embodiment, a total of 120 seconds of monitoring data was collected, resulting in 30 valid temperature frames. The statistical results based on the monitoring data are as follows: average temperature 42.50℃, mean 42.265725℃, variance 0.001568, standard deviation 0.039597, range 0.130035℃, maximum 42.310028℃, and minimum 42.179993℃.

[0091] Furthermore, after identifying the outer boundary region of the fuel cell stack, the temperature feature values ​​of the target region are extracted frame by frame, and the spatial uniformity, transient response, and cumulative heat load are calculated according to the formula, yielding the following results: ① Spatial uniformity

[0092] In the same frame image Taking the temperature of 5000 valid pixels as an example, the average value measured is: =42.27℃, According to the formula:

[0093] The spatial uniformity of this frame is obtained: =0.041 K, This indicates that the temperature field on the surface of the fuel cell stack is spatially uniform, with no obvious local hot spots.

[0094] ② Transient response

[0095] Taking Δt = 4 s (sampling interval) as an example, the following is used:

[0096] Select data around time t=40 s: =42.240℃, =42.284℃, The transient response is calculated as follows: ≈(42.284-42.240) / 8=0.0055 K / s.

[0097] The results show that the stack temperature changes slowly during this stage, which is consistent with a stable operating state.

[0098] ③ Cumulative heat load

[0099] The reference temperature is set to the initial operating temperature. =42.20℃, exist =0 s、 Calculations during a period of 120 seconds:

[0100] Obtained by integrating discrete data: Average deviation ≈ 0.065 K, duration = 120 s. ≈0.065×120=7.8 K·s.

[0101] The low cumulative heat load indicates that the temperature of the fuel cell stack remained stable during long-term operation.

Claims

1. A method for measuring the stack temperature of a fuel cell based on the fusion of infrared and visible light, characterized in that, Includes the following steps: S1. Perform joint calibration of the visible light camera and the infrared camera to obtain the spatial transformation relationship between the two cameras; S2. Synchronously acquire visible light video stream and infrared video stream, and extract time-synchronized visible light image and infrared temperature image as keyframe image pairs; S3. Based on the spatial transformation relationship and the image registration algorithm based on maximizing mutual information, the keyframe image pairs are fused to generate a fused image; S4. At multiple preset time scales, feature temperature values ​​are extracted and statistically analyzed from the target region in the sequence fusion image to generate temperature distribution feature data of the target region at different time scales. S5. Based on the temperature distribution characteristics data of all target areas, synthesize the overall temperature field of the fuel cell stack, analyze its characteristics as a function of time, and correlate it with different operating stages of the fuel cell stack.

2. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 1, characterized in that, The joint calibration of the visible light camera and the infrared camera in step S1 includes: S11. Use a black and white checkerboard calibration plate with a high emissivity coating or specific temperature control conditions. S12. Simultaneously acquire image sequences of the calibration plate in different poses from the visible light camera and the infrared camera; S13. Use Zhang Zhengyou's calibration algorithm to calculate the internal and external parameters of each camera respectively; S14. Based on the calibration results, obtain the spatial transformation relationship between the two cameras.

3. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 2, characterized in that, In step S11, the black and white checkerboard calibration board is a customized calibration board, wherein the checkerboard is composed of aluminum foil tape and graphite sprayed squares arranged alternately. The width of the chessboard grid on the customized calibration board According to the camera's effective focal length Target object feature dimensions and working distance Dynamic adjustment, the adjustment formula is: Among them, the width of the chessboard grid The width of the chessboard grid is in mm. This is the camera's effective focal length, in mm. The feature dimensions of the target object are in mm. Working distance, in mm; This is a proportionality coefficient, with a value ranging from 0.8 to 1.

2. In step S13, the application of Zhang Zhengyou's calibration algorithm to calculate the internal and external parameters of each camera includes: Construct the camera projection equation: , in, As a scale factor, Two-dimensional pixel coordinates a homogeneous dimensionless vector For focal length With the main point intrinsic parameter matrix ,in , It is the focal length, in mm. Is it pixels? Dimensions in the direction, in mm / pixel. , Is it pixels? Dimensions in the direction, in mm / pixel. These are the pixel coordinates of the image center. Let the external parameter rotation and translation matrix be the one from the world coordinate system to the camera coordinate system. It is a 3×3 matrix, dimensionless. It is a 3×1 vector, with units of mm. Point in the world coordinate system A homogeneous vector, with coordinates in mm; In step S14, the radial distortion coefficient of the camera is solved using a nonlinear optimization algorithm. and tangential distortion coefficient ; Based on the calibration results of the visible light camera and the infrared camera, the extrinsic parameter matrix of the infrared camera is denoted as follows: The extrinsic parameter matrix of a visible light camera is denoted as... The three-dimensional spatial matrix transformed from the infrared camera coordinate system to the visible light coordinate system. Calculated using the following formula: in, This represents the rotation matrix from the visible light camera coordinate system to the infrared camera coordinate system. This represents the translation vector from the visible light camera coordinate system to the infrared camera coordinate system.

4. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 1, characterized in that, Step S2 includes: S21. Fix the visible light camera and the infrared camera in place, and make their optical axes parallel or maintain a stable relative positional relationship. S22. Simultaneously start the two cameras and continuously acquire visible light video stream and infrared video stream at the set frame rate; S23. From the visible light video stream and infrared video stream acquired in step S22, frame-level time synchronization is achieved based on timestamps and using interpolation algorithms. The visible light image and infrared temperature image corresponding to the same timestamp or time point are extracted as a paired keyframe dataset.

5. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 4, characterized in that, The interpolation algorithm includes linear interpolation or cubic spline interpolation; The acquisition process employs layered synchronous control, with the software layer calibrating the timestamp. When the frame time deviation... When the time is greater than 2ms, the temperature field is reconstructed using motion compensation, and the expression is: in, This is the reconstructed infrared temperature field, measured in Kelvin. This is the infrared temperature field of the previous frame, in Kelvin. It is the time difference between frames, in milliseconds, where the gradient The time gradient of the temperature field is derived from the optical flow field of the adjacent frame and is expressed in K / ms.

6. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 1, characterized in that, Step S3 includes: S31. Perform contrast enhancement preprocessing on the visible light image and noise reduction preprocessing on the infrared temperature image; S32. Based on the spatial transformation relationship obtained in step S1, perform preliminary spatial alignment on the preprocessed visible light image and infrared temperature image; The spatial alignment includes affine transformation or perspective transformation; S33. Using a cross-modal image registration algorithm based on maximizing mutual information, calculate the mutual information or normalized mutual information of the initially aligned visible light image and infrared temperature image, and use it as a similarity evaluation index for image registration. S34. Based on the mutual information or normalized mutual information, the geometric transformation parameters are iteratively updated through a numerical optimization algorithm to maximize the mutual information or normalized mutual information, thereby obtaining the optimal spatial transformation matrix between the two images. The geometric transformation parameters include translation, rotation, and / or scaling parameters; S35. Perform precise geometric transformation on the infrared temperature image using the optimal spatial transformation matrix; S36. Employ pixel-level or region-level fusion strategies to fuse the registered infrared temperature information with the visible light image to generate a fused image.

7. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 6, characterized in that, In step S31, the contrast enhancement preprocessing of the visible light image employs a contrast-limited adaptive histogram equalization algorithm; the noise reduction preprocessing of the infrared temperature image employs a Gaussian blur algorithm, with a standard deviation of [missing information]. The value is 1.5; In step S32, the two-dimensional matrix of spatial transformation relationships obtained based on joint calibration is... The initial spatial alignment of the infrared temperature image is performed using the following coordinate transformation formula: in, This is a two-dimensional matrix representing the spatial transformation relationship between the two cameras obtained through joint calibration. pixel coordinates of a visible light image , a homogeneous dimensionless vector pixel coordinates in an infrared image a homogeneous dimensionless vector; Based on the calibration results of the visible light camera and the infrared camera, the extrinsic parameter matrix of the infrared camera is denoted as follows: The extrinsic parameter matrix of a visible light camera is denoted as... Two-dimensional matrix The calculation formula is as follows: in It is the normal vector of the target plane. Let be the distance from the origin of the camera coordinate system to the normal vector of the target plane in the reference camera coordinate system. This represents the rotation matrix from the visible light camera coordinate system to the infrared camera coordinate system. This represents the translation vector from the visible light camera coordinate system to the infrared camera coordinate system. and These are the intrinsic parameter matrices for the visible light camera and the infrared camera, respectively. In step S33, a cross-modal image similarity measurement method based on maximizing mutual information is adopted. By constructing a joint histogram of the visible light image and the infrared image, the normalized mutual information between the two images is calculated. The normalized mutual information calculation formula is as follows: in, and The edge entropy of infrared and visible light images, respectively. Let the joint entropy be ; let the visible light image be . Infrared image is Its 8-bit grayscale value is divided into 0~128 intervals, and a joint distribution histogram is constructed. ,in: It is an image Gray quantized index, It is an image Quantization index, This means that both images fall into the same interval at the same pixel position. The number of times is normalized to obtain the joint probability distribution: Edge probability distribution of visible light images: Edge distribution of infrared images: Edge entropy of visible light images: Edge entropy of infrared images: Joint entropy: In step S34, a numerical optimization algorithm is used to iteratively update the geometric transformation parameters, including the translation vector. Rotation angle With scaling factor The mutual information or normalized mutual information is maximized to obtain the optimal spatial transformation matrix between the two images. If the mutual information value decreases during the iteration process, the search is reversed to the parameters of the previous step and continues to avoid local optima. The two-dimensional image transformation matrix is ​​updated according to the registration parameters. in, and These represent the translation vectors along the X and Y axes, respectively. In step S36, bilinear interpolation is used to map the infrared image to the visible light coordinate system. The mapping formula is as follows: in, This represents the temperature value at the target coordinate point, in Kelvin (K). It is the original infrared image. Line number The column's pixel temperature values, in Kelvin (K). These are the pixel coordinates of the target location. These are the pixel network temperature values ​​along the X and Y axes of the original image, respectively.

8. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 1, characterized in that, Step S4 includes: S41. Under multiple preset time scales, select a series of key frame images within the corresponding time period in the fused image sequence; The multiple time scales include second-level scales, minute-level scales, and / or the entire test process scale; S42. For each keyframe image, calculate the specified characteristic temperature value of the target region within the fuel cell stack; S43. Perform statistical analysis on a series of characteristic temperature values ​​calculated for each time scale and target area, and calculate at least one of the following: average temperature, maximum temperature, minimum temperature, temperature standard deviation, and temperature change rate, or generate a time-temperature distribution map. S44. Integrate the temperature distribution feature data of each target area at multiple time scales to form a temperature distribution feature frame dataset.

9. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 8, characterized in that, First, the boundary regions of individual fuel cell stack units are identified, and the characteristic temperature values ​​of the target regions are extracted frame by frame. These characteristic temperature values ​​include spatial uniformity, transient response, and cumulative heat load, as shown in the following formula: Spatial uniformity: Transient response: Cumulative heat load: in, It is the space average temperature, measured in Kelvin (K). This is a manually specified reference temperature, in Kelvin (K). It is the number of statistical points, spatial uniformity. The unit is T, transient response The unit is K / s, cumulative heat load. Statistics from time Start to Time The accumulated temperature, measured in K·s, The unit is K, and it represents the temperature at time t.

10. The fuel cell stack temperature measurement method based on the fusion of infrared and visible light according to claim 1, characterized in that, Step S5 includes: S51. Collect temperature distribution characteristic data of all target areas in step S4; S52. Calculate at least one global characteristic temperature parameter that characterizes the overall thermal state of the fuel cell stack. The global characteristic temperature parameter includes the global maximum temperature, the global average temperature, the maximum temperature difference, the temperature uniformity index, or the global temperature distribution pseudo-color spectrum. S53. Analyze the sequence values ​​of the global characteristic temperature parameters at different time scales to generate the overall temperature time-varying characteristic curve or feature vector of the fuel cell stack. S54. Based on the preset rules for dividing the operation stages of the fuel cell stack and / or the external ambient temperature data monitored synchronously, the overall temperature time-varying characteristics of the fuel cell stack are correlated with the cold start stage, activation stage, stable operation stage or shutdown cooling stage. The stack operation phase is characterized by the feature division of the time-varying characteristic curve obtained from the time, load, operation mode, or step S53.

Images