Capacitor detection method and system

Abstract

The invention discloses a capacitor detection method and system, and the method comprises the steps: obtaining an original thermal image set of a capacitor in a power-on state, determining the temperature change rate of each point position of the capacitor based on the original thermal image set, and accurately describing the dynamic change condition of the temperature of the capacitor; performing abnormal point identification on the capacitor based on the temperature change rate distribution diagram to obtain a hot spot distribution diagram corresponding to the capacitor; based on the hot spot distribution diagram, heat flow transmission path analysis is carried out on the capacitor, the propagation track of the temperature change rate of the capacitor is determined, and an abnormal heat flow path diagram of the capacitor is obtained; according to the method, the abnormal heat flow and the internal hot spots are accurately identified, the key feature points are extracted from the abnormal heat flow path diagram, and the fault detection result of the capacitor is determined based on the difference between the key feature points and the preset template of the capacitor, so that the potential fault of the capacitor can be accurately and effectively detected, and the detection accuracy of the capacitor is improved.

Description

Technical Field

[0001] The present invention relates to the field of electronic information technology, and particularly discloses a method and system for detecting a capacitor. Background Art

[0002] A capacitor is a core component in an electronic device, and the stability and reliability of the performance of the capacitor have a crucial impact on the operation of the system. For example, a thin-film capacitor undertakes core functions such as energy storage and filtering in the fields of power electronics, industrial control, and new energy. Its failure may trigger major accidents such as equipment failures and system shutdowns. Therefore, the detection of capacitors is of great significance.

[0003] In the related art, an electrical parameter measurement method is usually adopted. The electrical parameter measurement method is mainly used in scenarios of off-line detection and regular maintenance. Specifically, devices such as a voltmeter and an ammeter are used to measure parameters such as the voltage and current of the capacitor; the measured value is compared with a preset safety threshold, and if it exceeds the threshold, it is determined as abnormal.

[0004] The related art simply relies on a fixed temperature threshold for detection, which is only limited to simple situations. Especially for the working process of a capacitor under complex working conditions, it is difficult to discover potential faults during the working process of the capacitor. Therefore, the accuracy of capacitor detection is relatively low. Summary of the Invention

[0005] The present invention provides a method and system for detecting a capacitor to improve the accuracy of capacitor detection.

[0006] On the one hand, a method for detecting a capacitor is provided, including: Obtaining an original thermal image set of the capacitor in a powered-on state, where the original thermal image set includes a plurality of original thermal images collected based on a target frequency, and each original thermal image includes temperature data corresponding to each pixel point in the original thermal image; Based on the original thermal image set, determining the temperature change rate of each point of the capacitor to obtain a temperature change rate distribution map corresponding to the capacitor, where one point of the capacitor corresponds to one pixel point of each original thermal image; Identifying abnormal points of the capacitor based on the temperature change rate distribution map to obtain a hot spot distribution map corresponding to the capacitor; Analyzing the heat flow transfer path of the capacitor based on the hot spot distribution map to determine the propagation trajectory of the temperature change rate of the capacitor and obtain an abnormal heat flow path map of the capacitor; Extracting key feature points from the abnormal heat flow path map, and determining the fault detection result of the capacitor based on the difference between the key feature points and a preset template of the capacitor.

[0007] In another aspect, a detection system for a capacitor is provided, including: An acquisition module, configured to acquire an original thermal image set of the capacitor in a powered-on state, where the original thermal image set includes a plurality of original thermal images collected based on a target frequency, and each original thermal image includes temperature data corresponding to each pixel point in the original thermal image; A first determination module, configured to determine a temperature change rate of each point of the capacitor based on the original thermal image set, and obtain a temperature change rate distribution map corresponding to the capacitor, where one point of the capacitor corresponds to one pixel point of each original thermal image; A second determination module, configured to perform abnormal point identification on the capacitor based on the temperature change rate distribution map, and obtain a hot spot distribution map corresponding to the capacitor; A third determination module, configured to perform a heat flow transfer path analysis on the capacitor based on the hot spot distribution map, determine a propagation trajectory of the temperature change rate of the capacitor, and obtain an abnormal heat flow path map of the capacitor; A detection module, configured to extract key feature points from the abnormal heat flow path map, and determine a fault detection result of the capacitor based on a difference between the key feature points and a preset template of the capacitor.

[0008] In another aspect, an electronic device is provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, where when the processor executes the computer program, the above-mentioned capacitor detection method is implemented.

[0009] In another aspect, a computer-readable storage medium is provided, where the computer-readable storage medium includes a stored computer program, and when the computer program runs, it controls a device where the computer-readable storage medium is located to implement the above-mentioned capacitor detection method.

[0010] The beneficial effects achieved by the present invention are: The present invention provides a method for detecting a capacitor. By obtaining a set of original thermal images of the capacitor in the energized state, determining the temperature change rate of each point of the capacitor based on the set of original thermal images, and obtaining a temperature change rate distribution map based on this, the dynamic thermal distribution of the capacitor can be detected in real time, and the transient characteristics of the capacitor can be captured in real time, so as to accurately describe the dynamic change of the capacitor temperature and lay a foundation for the subsequent dynamic analysis of the capacitor. Then, based on the temperature change rate distribution map, the abnormal points of the capacitor are identified to obtain the hot spot distribution map corresponding to the capacitor; and based on the hot spot distribution map, the heat flow transfer path of the capacitor is analyzed to determine the propagation trajectory of the temperature change rate of the capacitor and obtain the abnormal heat flow path map of the capacitor; based on the obtained abnormal heat flow path map, the abnormal conditions in the dynamic thermal process of the capacitor can be accurately captured, so as to accurately identify the abnormal heat flow and internal hot spots. Based on this, key feature points are extracted from the abnormal heat flow path map, and based on the difference between the key feature points and the preset template of the capacitor, the fault detection result of the capacitor is determined. Since the detection is based on the key feature points in the abnormal heat flow path map with high accuracy, especially for the working scenarios of capacitors under complex working conditions, the potential faults in the working process of the capacitor can be accurately and effectively detected, improving the accuracy of capacitor detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic flowchart of a method for detecting a capacitor provided by an embodiment of the present invention.

[0012] Figure 2 It is a schematic structural diagram of a detection system for a capacitor provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] Figure 1 It is a schematic flowchart of a method for detecting a capacitor provided by an embodiment of the present invention. As Figure 1 shown, the present invention provides a method for detecting a capacitor, and the method includes the following steps: Step S101: An electronic device obtains a set of original thermal images of the capacitor in the energized state. The set of original thermal images includes a plurality of original thermal images collected based on a target frequency, and each original thermal image includes temperature data corresponding to each pixel point in the original thermal image.

[0014] Among them, the capacitor may be a thin film capacitor. For example, a thin film capacitor that undertakes core functions such as energy storage and filtering in the fields of power electronics, industrial control, and new energy.

[0015] In this step, a high-speed thermal imaging device may be used to collect the original thermal images. For example, an infrared thermal imager may be used.

[0016] Specifically, when acquiring a sequence of continuous-frame thermal images of a thin-film capacitor in the energized state, first use a high-speed thermal imaging device to record the surface temperature change of the capacitor at a rate of 100 frames per second, ensuring that the time resolution reaches 10 milliseconds. The device uses an infrared thermal imager, which can accurately capture the minute temperature fluctuations of the capacitor during energization. Through the real-time image processing algorithm built into the device, the original thermal image data is converted into a temperature matrix, where each pixel point corresponds to a temperature value, forming a temperature dataset in the time dimension. Then, use a time-series-based filtering algorithm, such as the Kalman filter, to denoise the original thermal image set and eliminate the influence of ambient temperature fluctuations and device noise.

[0017] Specifically, an electronic device can use a high-speed thermal imaging device to detect a capacitor in the energized state and record the surface temperature of the capacitor as the original thermal image data. For example, set the recording rate to 100 frames per second to obtain continuous original thermal image data; set the time resolution to 10 milliseconds to more finely capture the change in the surface temperature of the capacitor over time. For example, the high-speed thermal imaging device can use an infrared thermal imager, and the thermal sensitivity of the infrared thermal imager is 0.05°C. Using an infrared thermal imager with precise sensitivity for data acquisition can help accurately capture the minute temperature fluctuations that occur in the capacitor during energization and improve the accuracy and continuity of capacitor detection.

[0018] For example, set the time resolution to 10 milliseconds, and use the thermal imager to continuously capture 1000 thermal images at a speed of 100 frames per second for 10 seconds. The resolution of each image is 640×480 pixels, and each pixel point records a temperature value.

[0019] The electronic device can convert the original thermal image data into a temperature matrix through the real-time image processing algorithm built into the device. Specifically, the electronic device can use the real-time image processing algorithm built into the infrared thermal imager to process the recorded original thermal image data. For each frame of original thermal image data, each pixel point in the image data is corresponded to a temperature value, thereby constructing a temperature matrix, which is the original thermal image corresponding to this frame of original thermal image data. Based on multiple frames of original thermal image data, multiple temperature matrices are obtained, thus forming a temperature dataset in the time dimension, providing basic data for subsequent analysis.

[0020] The electronic device can use a time-series-based filtering algorithm to denoise the original thermal image set. For example, the filtering algorithm can use an algorithm such as the Kalman filter algorithm. The electronic device uses the Kalman filter algorithm to filter the obtained original thermal image set to eliminate the influence of ambient temperature fluctuations and device noise on the thermal image, improve the quality and accuracy of the original thermal image, and make the subsequent analysis of the temperature change of the capacitor more accurate and reliable.

[0021] Step S102: Based on the original thermal image set, the electronic device determines the temperature change rate of each point of the capacitor, and obtains the temperature change rate distribution map corresponding to the capacitor.

[0022] Among them, one point of the capacitor corresponds to one pixel point of each original thermal image. That is to say, one point of the capacitor corresponds to one pixel point in each original thermal image, and the pixel value of one pixel point in each original thermal image represents the temperature of that point. Based on this, the original thermal image set can represent the temperature of each point changing with time.

[0023] Among them, the temperature change rate distribution map includes the temperature change rate of each point. The temperature change distribution map can characterize the trend of the temperature of each point of the capacitor changing with time.

[0024] In this step, for each point, the electronic device can extract the temperatures of that point changing with time from the original thermal image set, and according to the temperatures corresponding to that point and the time stamps of each temperature, use the linear regression method to fit the change rate of the temperature of each point with time, and obtain the temperature change rate distribution map including the change rates of multiple points.

[0025] It should be noted that the pixel value of one pixel point in the temperature change rate distribution map can include the temperature change rate of the corresponding point.

[0026] Specifically, taking the time span of the thermal image sequence as 10 seconds, the frame rate as 100 frames per second, and the total number of frames as 1000 frames as an example; for each point, extract the temperature values of that point in 1000 original thermal images and the acquisition time stamps of 1000 original thermal images, and use the least squares method for linear regression fitting according to the 1000 temperature values corresponding to that point and the acquisition time stamps to calculate the change rate of the temperature with time.

[0027] In specific implementation, taking the point (x, y) as an example, the temperature value sequence of the pixel points corresponding to it in 1000 original thermal images is expressed as T1, T2,..., T1000, and the time stamp sequence is t1, t2,..., t1000. Through the following formula: Among them, n = 1000. Based on this, the change rate of the temperature of each point can be calculated, and this change rate reflects the speed of the temperature change of that pixel point. Store the change rate value of each point in a 640×480 matrix to generate the temperature change rate distribution map. The temperature change rate situation of each point in this distribution map intuitively reflects the distribution of the temperature change rate in the image space.

[0028] In another possible implementation, the transient change rate of each point can also be calculated by using the differential thermal image sequence corresponding to the original thermal image set. Correspondingly, the specific process of step S102 may include the following steps S1021 to S1023: Step S1021: The electronic device determines the temperature change rate of each point of the capacitor based on the original thermal image set; Step S1022: The electronic device uses the inter-frame difference algorithm to extract the dynamic features of the original thermal image set, obtains the differential thermal image sequence, and obtains the transient change rate of the temperature of each point of the capacitor based on the differential thermal image sequence; the differential thermal image sequence characterizes the dynamic change features of the temperature distribution between adjacent images in the original thermal image set; Step S1023: Based on the temperature change rate and the transient change rate of each point of the capacitor, a temperature change rate distribution map corresponding to the capacitor is obtained.

[0029] In step S1022, the electronic device can use the inter-frame difference algorithm to subtract the pixel values of adjacent frames pixel by pixel to extract the dynamic change features of the temperature distribution. Specifically, for adjacent-frame thermal images, perform the following operation for each pixel point position: subtract the pixel values of two pixel points corresponding to the same position to obtain the differential thermal image corresponding to the adjacent-frame thermal image. Based on this, the differential thermal image sequence corresponding to the original thermal image set is obtained. Among them, the pixel value of the pixel point can be the temperature value, that is, calculate the temperature difference corresponding to two pixel points at the same position in adjacent-frame thermal images, and the pixel value of each pixel point in the differential thermal image is the corresponding temperature difference.

[0030] Specifically, taking the original thermal image sequence of 100 frames per second as an example, the electronic device can calculate the difference value between the nth frame of the original thermal image and the (n + 1)th frame of the original thermal image, and obtain a new image, that is, the differential thermal image, by subtracting the temperature values of each corresponding pixel point in these two frames of images.

[0031] Taking the resolution of each frame of thermal image as 640×480 pixels as an example, matrix operations are used when calculating the difference value. Subtraction operations are performed on the temperature values corresponding to each pixel point, and thus a differential matrix with the same size of 640×480 is obtained, and this matrix records the temperature change of each pixel point in adjacent two frames of images.

[0032] Further, the differential matrix can be corrected to improve its accuracy. Specifically, a threshold filtering method can be used to set the pixel points with differential values less than a preset threshold in the differential matrix to zero, eliminating the influence of tiny noises. For example, to make the differential effect more obvious and reduce the interference of tiny noises on the result, a preset threshold, that is, the temperature change threshold is 3°C, is adopted. For the pixel points in the differential matrix with differential values less than 3°C, their differential values are set to zero, which can highlight the areas with larger temperature changes and eliminate the temperature change information caused by tiny noises, so as to improve the accuracy of the differential matrix.

[0033] Further, the differential thermal image can also be smoothed by using the Gaussian smoothing algorithm. Among them, a 3×3 Gaussian convolution kernel matrix can be used as a tool for the smoothing operation, and the kernel parameter is set to σ = 0. By performing convolution operation on the differential thermal image in this way, the noises at the image edges can be reduced while the main temperature change areas are retained, making the image smoother and clearer.

[0034] For example, define a 3×3 Gaussian convolution kernel matrix G: ; Among them, when the kernel parameter σ = 0, it is equivalent to mean filtering, but retains the advantage of central weight. The above-mentioned Gaussian convolution kernel matrix G can be used to perform sliding window convolution processing on the differential matrix.

[0035] Based on this, the pixel value of a pixel point in the differential thermal image represents the temperature difference corresponding to this point in adjacent frame thermal images. The differential thermal image sequence can represent the temperature difference changing with time at each point.

[0036] Among them, for each point, the electronic device can extract the temperature difference changing with time at this point from the differential thermal image sequence, and calculate the transient change rate of the temperature difference changing with time at this point according to the respective temperature differences corresponding to this point and the time gradient of each temperature difference.

[0037] For example, the differential value of each pixel point in the differential thermal image sequence represents the temperature change amount between adjacent frames. If the time interval between adjacent thermal images in the original thermal image set is Δt, for example, the inter-frame time interval Δt can be 10 milliseconds; the pixel value D of the pixel point in each differential thermal image n = T n+1 - T n , that is, ΔT; where, T n+1 , T n represent the (n + 1)-th frame original thermal image and the n-th frame original thermal image respectively. Then, the instantaneous change rate of the point corresponding to this pixel point can be ΔT / Δt.

[0038] Based on this, in step S1023, the temperature change rate values and transient change rate values of each point can be stored in a 640×480 matrix to generate a temperature change rate distribution map.

[0039] Step S103: The electronic device identifies abnormal points of the capacitor based on the temperature change rate distribution map to obtain a hot spot distribution map corresponding to the capacitor.

[0040] The electronic device can identify abnormal points with abnormal temperature change rates among the pixels of the temperature change rate distribution map, perform clustering processing based on the abnormal points to obtain multiple clustering clusters, and use the clustering center corresponding to each clustering cluster as a hot spot. Based on this, a hot spot distribution map corresponding to the capacitor is obtained. For example, points with a temperature change rate reaching abnormal conditions can be used as abnormal points. For example, for each pixel of the temperature change rate distribution map, the average value of each temperature change rate in the temperature change rate matrix can be calculated, the difference between the temperature change rate of each pixel and the average value can be calculated, and points with a difference exceeding a preset temperature change rate difference threshold can be used as abnormal points.

[0041] Specifically, the abnormal region can be identified first based on the temperature change rate distribution map, and then clustering can be performed on the abnormal points in the abnormal region to obtain a hot spot distribution map. In a possible implementation manner, step S103 includes steps S1031 - S1033: Step S1031: Identify an abnormal heat flow candidate region set corresponding to the capacitor based on the temperature change rate distribution map.

[0042] In this step, the electronic device can perform normalization processing on the temperature change rate distribution map, that is, perform normalization processing on the temperature change rate matrix; based on each temperature change rate in the processed temperature change rate matrix, identify abnormal points with abnormal temperature change rates. Perform clustering processing on each abnormal point, and cluster to obtain each abnormal heat flow candidate region including the abnormal points. Based on this, an abnormal heat flow candidate region set is obtained.

[0043] In a possible implementation manner, specifically, a statistical-based outlier detection method can be used in the temperature change rate distribution map, with the threshold set to 10°C / s, and the Z-score algorithm is used to perform normalization processing on the temperature change rate matrix. For example, after obtaining the temperature change rate distribution map, the points with abnormal temperature change rates in the distribution map can be found. Specifically, a statistical-based method can be used to detect outliers, and a standard is set, for example, if the temperature change rate of a certain point exceeds the threshold of 10°C / s, then this point is an abnormal point.

[0044] Among them, in order to better judge which points are outliers, the Z-score algorithm can be used to process the data in the temperature change rate matrix.

[0045] For example, the following formula can be used to calculate the score of each point based on the temperature change rate of each point in the temperature change rate matrix: Z = (k - μ) / σ; Where k represents the temperature change rate of the point in the temperature change rate matrix. μ represents the global mean (i.e., the average value) of the temperature change rates of each point in the temperature change rate matrix. μ can be obtained by adding up the temperature change rates of each point in the temperature change rate matrix and taking the average. σ is the standard deviation, which reflects the gap between the temperature change rates of most points and the average value and can be used to measure the degree of dispersion of the data. For example, μ = 3 °C / s and σ = 2 °C, indicating that the change rates of most points are between 3 °C ± 2 °C (i.e., 1 - 5 °C / s).

[0046] Then, outlier identification is performed based on the scores of each point. For example, for each point, if the absolute value of the score of this point is greater than the target threshold, then this point is determined to be an outlier. For example, if the absolute value of the score (i.e., the Z-score value) of a certain point is greater than 3, it means that the temperature change rate of this point varies greatly from the overall average situation, and it can be determined as an outlier. If the absolute value of Z is greater than 3 (i.e., exceeding 3 times the standard deviation of the average value), it is determined as an outlier. For example, a point with Z = 3.5 is a "point of abnormal temperature increase", and a point with Z = -3.2 is a "point of abnormal temperature decrease".

[0047] In another possible implementation, the temperature change rate value and the transient change rate value of each point in the temperature change rate distribution map can also be combined for outlier identification. Among them, the difference between multiple transient change rate values of this point and the temperature change rate can be combined for identification. For example, if the absolute value of the score of a certain point is greater than 3, and the difference between the maximum value among the transient change rate values of this point and the temperature change rate value of this point exceeds a predetermined difference threshold, such as the difference exceeds 10; then this point can be identified as an outlier. Of course, for the specific combination method of how to combine the temperature change rate value and the transient change rate value for outlier identification, other combination methods can also be included. Here, only the above example is used for illustration, but the combination method is not limited. Another example is that if there are more than a predetermined number of abnormal transient change rate values among the multiple transient change rate values of this point, this point can also be identified as an outlier. For example, if the difference between the transient change rate value and the temperature change rate value of this point exceeds the predetermined difference threshold, then this transient change rate value is an abnormal transient change rate value; if the absolute value of the score of a certain point is greater than 3, and the abnormal transient change rate values of this point exceed 40% of the total number of transient change rate values, this point is an outlier.

[0048] Then, the abnormal points can be clustered into candidate regions. That is, after the points determined to be outliers have been found, these abnormal points can be clustered together to obtain the regions where the outliers are concentrated.

[0049] For example, the DBSCAN algorithm can be used for spatial clustering. This algorithm will cluster the abnormal points that are closer together into one class according to the positional relationship between each point, forming candidate regions one by one. For example, set the parameters of the DBSCAN algorithm, the neighborhood radius is 5 pixels, and the minimum number of samples is 10. That is, taking a certain point as the center, the points within the range of 5 pixels are the neighborhood of this point. If there are at least 10 abnormal points in a region, then this region is determined to be a valid clustering region. If the number of abnormal points in a region is less than 10, then this region is determined not to be a valid clustering region. For example, Region A has 15 abnormal points that are within 5 pixels of each other, so Region A is marked as a candidate region for abnormal heat flux; Region B has only 8 abnormal points, less than 10, so it is ignored. Based on this, a set of candidate regions for abnormal heat flux can be obtained.

[0050] Step S1032: Cluster the abnormal points in the set of candidate regions for abnormal heat flux to obtain the internal hot spot distribution map corresponding to the capacitor.

[0051] After the set of candidate regions for abnormal heat flux is available, the distribution of the abnormal points in these regions can be further analyzed.

[0052] In this step, for the abnormal points in each set of candidate regions for abnormal heat flux, the DBSCAN algorithm can be further used for spatial clustering analysis to better obtain which abnormal points are closer and can be clustered into one class. Among them, based on the continuity of the heat flux transfer path, the adjacent abnormal points are aggregated to obtain the internal hot spot distribution map. That is, if the heat flux transfer path between some abnormal points is continuous and the distance is close, these abnormal points can be aggregated together to form an internal hot spot distribution map. This distribution map can reflect which regions are the hot spot regions where the heat is concentrated. The data of each pixel point in the internal hot spot distribution map can include the temperature change rate as an abnormal point. In addition, the temperature data of this pixel point can also be included.

[0053] Specifically, the neighborhood radius of the DBSCAN algorithm can be set to 8 pixels, and the minimum number of samples is 15. That is, taking a certain abnormal point as the center, the other abnormal points within the range of 8 pixels belong to the neighborhood range of this abnormal point; if the number of abnormal points in a region is less than 15, then this region is determined not to be a valid clustering region. Based on this, the clustering result can accurately reflect the spatial distribution characteristics of heat transfer.

[0054] Among them, clustering can be performed by calculating the Euclidean distance between each abnormal point and other points within the neighborhood of the abnormal point. For example, for each abnormal point, calculate the Euclidean distance between the abnormal point and other abnormal points within the neighborhood of the abnormal point. If the Euclidean distance is less than 8 pixels and the number of abnormal points within the neighborhood of the abnormal point is not less than 15, then these points are classified into the same class, that is, the same clustering cluster.

[0055] For each clustering cluster after clustering, calculate the geometric center point of the clustering cluster as the position of the internal hot spot corresponding to the clustering cluster. That is, for each clustering cluster, the geometric center point of the clustering cluster represents the position of this hot spot area. For example, clustering cluster A contains 50 abnormal points, and the geometric center coordinates of clustering cluster A are (120, 80); clustering cluster B contains 30 abnormal points, and the geometric center coordinates of clustering cluster B are (150, 100); these geometric center coordinates can reflect the specific position of the internal hot spot in the thermal image.

[0056] Step S1033: Correct the internal hot spot distribution map according to the environmental temperature fluctuation data and the capacitor material characteristics, and obtain the hot spot distribution map of the capacitor based on the corrected internal hot spot distribution map.

[0057] Step S1033 may include the following steps 1 to step 5: Step 1: Use an infrared thermal imager to collect the infrared thermal image of the capacitor, and use a distributed temperature sensor network to record the environmental temperature fluctuation of the capacitor to obtain an environmental temperature change matrix.

[0058] An infrared thermal imager can be used to collect multiple infrared thermal images of the capacitor at a sampling frequency of 5 frames per second. Each infrared thermal image represents the temperature data of the capacitor; for example, each infrared thermal image includes multiple pixel points, and the pixel value of each pixel point is the temperature data corresponding to the pixel point.

[0059] Among them, the distributed temperature sensor network may include arranging multiple high-precision temperature sensors around the capacitor. For example, PT100 platinum resistors. Multiple high-precision temperature sensors can cover different orientations of the capacitor. For example, the upper, lower, left, right, front, back, etc. orientations of the capacitor to ensure all-round monitoring of the environmental temperature of the capacitor.

[0060] Among them, the environmental temperature can be recorded once every 10 seconds through the distributed temperature sensor network, with an accuracy of 1°C.

[0061] Among them, the temperature data of the temperature sensor can also be aligned with the timestamp of the infrared thermal imager to generate an environmental temperature change matrix T with a time resolution of 10 seconds env , T envEach row of the matrix contains fields: [timestamp, sensor ID, position coordinates, temperature value]. For example, at a certain moment in the ambient temperature matrix, sensor A records 25.3 °C and sensor B records 24.8 °C.

[0062] Based on this, an ambient temperature change matrix with a time resolution of 10 seconds can be established, which can record the ambient temperature changes within every 10 seconds for subsequent analysis.

[0063] Furthermore, in combination with the ambient temperature change matrix, the Kalman filtering algorithm can be used to correct the infrared thermal image to obtain smoothed temperature data with environmental noise removed, that is, the corrected infrared thermal image. For example, the process noise covariance can be set to 0.1 and the observation noise covariance can be set to 0.5. Iterative filtering is performed through the state equation and the observation method, that is, the temperature data collected by the infrared thermal imager is corrected in real time to suppress the environmental noise and device noise in the temperature data. For example, a certain pixel point is measured by the thermal imager at 85 °C, the predicted value is 83 °C due to ambient temperature fluctuations, and the output after filtering is 84 °C.

[0064] Step 2: Establish a target lookup table based on the material emissivity database. Based on this target lookup table and the infrared thermal image of the capacitor, obtain the standard emissivity of each pixel point in the internal hot spot distribution map.

[0065] Among them, the target lookup table can include the emissivities of various materials at different temperatures. That is, the target lookup table can be an "emissivity-temperature" lookup table. For example, the two-dimensional lookup table can be: LUT(material type, temperature) → emissivity.

[0066] In this step, the material emissivity database can be obtained in advance. The emissivity database records the emissivity information of various materials at different temperatures. Based on this emissivity database, an "emissivity-temperature" lookup table is established. For example, the emissivity of aluminum alloy is 0.12 at 80 °C, and the emissivity of stainless steel is 0.18 at 120 °C. Among them, the emissivity in the target lookup table can be called the standard emissivity.

[0067] Among them, for the electronic device, based on the material and temperature corresponding to each pixel point in the internal hot spot distribution map, the standard emissivity corresponding to this pixel point is looked up from the target lookup table. Among them, the temperature of each pixel point in the internal hot spot distribution map can be obtained according to each infrared thermal image of the capacitor. For example, according to the position of each pixel point in the internal hot spot distribution map, the pixel value of this position in each infrared thermal image is obtained, and this pixel value can include the temperature of this pixel point. In addition, the material of the pixel point can be obtained before step S103, or can also be obtained during step S103. For example, the temperature and material of each pixel point are obtained when collecting the infrared thermal image of the capacitor.

[0068] Step 3: spatially correct the standard emissivity of each pixel in the internal hotspot distribution map using a bilinear interpolation algorithm to obtain a corrected emissivity; and, based on the corrected emissivity, compensate the temperature data of each pixel in the internal hotspot distribution map to obtain compensated temperature data.

[0069] For step 1, a bilinear interpolation algorithm can be used to perform spatial correction on the standard emissivity of each pixel in the internal hotspot distribution map to obtain the corrected emissivity of the pixel. Specifically, for each pixel (x, y), the following steps can be performed to calculate the weight based on the material distribution of the pixel's neighboring pixels, and based on the weights and emissivity of the neighboring pixels, the corrected emissivity of the pixel can be obtained. The corrected emissivity can also be called the actual emissivity: (1) Determine the material type and temperature value of multiple reference points around the pixel point (x, y); the multiple reference points around the pixel point may be adjacent pixels of the pixel point; (2) Calculate the actual emissivity of the pixel (x, y) by weighting it inversely proportional to the distance using the following formula: ; in, Represents the corrected emissivity of the pixel (x, y). n represents the number of reference points surrounding the pixel (x, y). For example, if four reference points are used around the pixel (x, y), then n = 4.

[0070] in, , Represents the Euclidean distance from pixel (x, y) to reference point i.

[0071] In addition, for each pixel, the temperature data of the pixel can be compensated based on the actual emissivity and the standard emissivity of the pixel. For example, the compensated temperature data of the pixel can be calculated using the following formula: ; in, Indicates compensated temperature data, also known as corrected temperature data; Indicates the temperature data before correction, 、 , respectively represent the standard emissivity and actual emissivity. For example, if the original temperature of a pixel (that is, the temperature before correction) is 100°C, the actual emissivity is ε=0.15, and the standard emissivity is ε=0.12, then the corrected temperature of the pixel is = ≈96.3℃.

[0072] It should be noted that for each pixel point in the internal hot spot distribution map, the pixel value, that is, the temperature, of this pixel point in each infrared thermal image of the capacitor can be obtained; and the temperature of this pixel point in each infrared thermal image is compensated to obtain the compensated temperature. Based on this, the compensated temperatures corresponding to multiple infrared thermal images for each pixel point in the internal hot spot distribution map are obtained.

[0073] Since the materials, temperatures, and emissivities at different positions can be different, by performing spatial correction on the emissivity of each pixel point in the infrared thermal image in this step, based on this, the emissivity of the material represented by each pixel point in the infrared thermal image can be made more accurate.

[0074] Step 4: Based on the compensated temperature data corresponding to the internal hot spot distribution map and the finite element heat conduction model, perform coupled calculations to obtain the three-dimensional temperature field distribution inside the material of the capacitor.

[0075] Among them, the position coordinates, compensated temperature, and material parameters of each pixel point in the internal hot spot distribution map can be coupled with the finite element heat conduction model to obtain the three-dimensional temperature field distribution.

[0076] Among them, the finite element heat conduction model is a tool that simulates the law of heat transfer inside an object through mathematical methods. It is like a "digital microscope" that can decompose the calculation of the temperature distribution of a complex object into small pieces for processing and finally restore the entire heat change process of the object. In some scenarios, the shape of the real object is complex and it is difficult to directly calculate the overall heat transfer. For example, for a capacitor. Therefore, based on the finite element heat conduction model, the object can be divided into countless small elements (such as triangles, tetrahedrons), just like disassembling a jigsaw puzzle into small pieces, analyzing the heat change of each piece one by one, and then piecing back the overall result.

[0077] Specifically, in this step, the target area can be first divided into multiple unit meshes; for each unit mesh, based on the corresponding compensated temperature data and the environmental temperature change matrix, the conjugate gradient method is used to solve the heat conduction equation for calculation.

[0078] Among them, the iterative solution calculation can be performed according to the position coordinates, compensated temperature, material parameters of each pixel point in the internal hot spot distribution map, and the environmental temperature change matrix through the following formula: ; Among them, ρ represents the density of the material of the pixel point, which determines the thermal inertia. c p represents the specific heat capacity of the material of the pixel point, which reflects the heat storage capacity. For example, for aluminum, ρ = 2700 kg / m 3 . k represents the thermal conductivity, with the unit of W / (m·K), which is an index of the heat conduction ability of the material and determines the heat transfer rate. ∇ TDenotes the temperature gradient, representing the rate of change of temperature with respect to space (℃ / m), which drives heat flow from the high-temperature region to the low-temperature region. Denotes the rate of change of temperature with respect to time (℃ / s), reflecting the transient characteristics.

[0079] It should be noted that the left side of the equation represents the rate of heat accumulation per unit volume, and the right side represents the net heat flux entering or leaving this volume due to heat conduction. In the steady state ( = 0), the equation simplifies to the Laplace equation, which describes the equilibrium state of the temperature field.

[0080] Among them, a three-dimensional mesh can be first generated for the capacitor; for example, the capacitor can be discretized into a tetrahedral element mesh with an element size less than or equal to 1 mm to ensure spatial resolution. Or, the element size can be refined to 0.5 mm in key areas (such as near hot spots), and relaxed to 2 mm in non-critical areas.

[0081] During the solution process, spatial discretization can be first carried out, dividing the three-dimensional model of the capacitor into a finite number of elements. The temperature distribution within each element is approximated by a shape function (such as a linear shape function). The global stiffness matrix K and the load vector F are generated to describe the heat conduction relationship between elements. Then, time discretization can be carried out, using implicit time integration. For example, the time step = 0.1 s. The discretized equation is: ; where M is the mass matrix, is the temperature field at the nth step.

[0082] Based on this, regarding the application of boundary conditions: Among them, the Dirichlet boundary condition can adopt the direct amplitude of the surface temperature. For example, the corrected temperature data can include the compensated temperature data corresponding to each pixel point. Among them, the Neumann boundary condition can be set based on the ambient temperature change matrix, considering convective heat dissipation, that is, considering surface heat dissipation. The ambient temperature change matrix T env , that is, the boundary conditions are expressed as follows: ; Based on this, iterative solution is carried out. Specifically, the conjugate gradient method can be used to solve the heat conduction equation, setting the upper limit of the number of iterations to 200 times and the residual tolerance to 1e-6 to judge the accuracy of the calculation results. Based on this, the calculation is carried out to obtain the three-dimensional temperature field distribution inside the material, so as to obtain the temperature at different positions inside the material.

[0083] It should be noted that the finite element heat conduction model is verified by physical law constraints to ensure that the corrected hot spot distribution map conforms to the laws of thermodynamics. Its input integrates multi-source data (measured temperature, material properties, environmental parameters), and outputs a high-precision three-dimensional temperature field, providing a key basis for the reliability assessment of capacitors. Through multi-dimensional technical optimization (mesh, nonlinear processing, transient analysis), the model combines computational efficiency and accuracy, and is the core bridge connecting experimental measurement and engineering decision-making.

[0084] Step 5: Extract the isothermal surfaces from the three-dimensional temperature field distribution of the capacitor, obtain a three-dimensional hot spot distribution map based on the extracted isothermal surfaces, and use this three-dimensional hot spot distribution map as the hot spot distribution map corresponding to the capacitor.

[0085] In this step, for the three-dimensional temperature field distribution inside the material of the capacitor, the Marching Cubes algorithm is used to generate isothermal surfaces in order to convert the three-dimensional temperature field into a visualized hot spot distribution map. For example, the three-dimensional temperature field distribution can be divided into regular cubic meshes, that is, voxels. Each voxel is composed of 8 vertices, and each mesh vertex stores a temperature value. The temperature threshold can be set to 85 °C. According to the temperature values stored at each network vertex, the intersection points corresponding to the temperature threshold of 85 °C for each voxel are calculated to extract the regions equal to or higher than 85 °C. Based on the extracted regions, surfaces with equal temperature are generated, that is, isothermal surfaces. Among them, the isothermal surfaces can be smoothed by calculating the vertex normal vectors.

[0086] Based on the obtained isothermal surfaces equal to or higher than 85 °C, a three-dimensional hot spot distribution map is generated. Since the three-dimensional temperature field is obtained through processes such as temperature compensation and emissivity correction, considering the influence of the ambient temperature and the change of the material emissivity, this three-dimensional hot spot distribution map can also be called a three-dimensional hot spot distribution map including temperature compensation and emissivity correction. Among them, the high-temperature regions are represented by a red gradient, and the temperature range is marked as 85 - 120 °C to visually represent the temperature situation.

[0087] Step S104: The electronic device analyzes the heat flow transfer path of the capacitor based on this hot spot distribution map, detects the propagation trajectory of the temperature change rate of the capacitor from high to low, and obtains the abnormal heat flow path map of the capacitor.

[0088] This step may include the following steps 1 to 7: Step 1: Use the Sobel operator to calculate the temperature gradient field of this hot spot distribution map, and obtain the gradient amplitude and direction of each pixel point in this hot spot distribution map.

[0089] In this step, the hot spot distribution map can be a three-dimensional hot spot distribution map, and the hot spot distribution map includes temperature information at each position in the target area. Among them, each pixel point in the hot spot distribution map is a pixel point in three-dimensional space, and each pixel point can also be called a voxel.

[0090] The electronic device can use the Sobel operator to calculate the temperature gradient field of the hot spot distribution map. Specifically, the electronic device can use a 3×3 convolution kernel for horizontal gradient calculation and vertical gradient calculation. For example, use a 3×3 convolution kernel to scan each voxel of the hot spot distribution map horizontally to calculate the temperature difference between the left and right of the voxel, that is, "slide" the convolution kernel on the three-dimensional hot spot distribution map, and each time it slides, it can calculate the temperature data at the corresponding position to obtain the horizontal temperature change. Also, transpose the 3×3 convolution kernel, and use the transposed convolution kernel to scan each voxel of the hot spot distribution map vertically to calculate the temperature difference between the top and bottom of the voxel.

[0091] Among them, the 3×3 convolution kernel is: ; Based on this, the gradient magnitude and direction of each voxel (pixel point in three-dimensional space) can be obtained. Among them, the gradient magnitude can represent the magnitude of temperature change, and the gradient direction represents the direction in which the temperature changes fastest.

[0092] Step 2: Based on the gradient magnitude and direction of each pixel point in the hot spot distribution map, use the non-maximum suppression algorithm to screen out each key path point from each pixel point in the hot spot distribution map.

[0093] Among them, the key path point is the point used to form the heat flow path of the capacitor.

[0094] In this step, the non-maximum suppression algorithm can be used for calculation. The electronic device can screen each pixel point based on the gradient magnitude and direction of each pixel point, and use the non-maximum suppression algorithm to screen out the key path points from each pixel point. For example, according to the gradient magnitude and direction of each pixel point, it can be determined whether each pixel point is the point with the largest gradient among the surrounding pixel points. If so, retain the pixel point; if not, filter out the pixel point. Based on this, the points at the "peak" positions among each pixel point in the three-dimensional hot spot distribution map can be retained.

[0095] In addition, the electronic device can also filter out noise points by means of threshold filtering. Based on the gradient magnitude threshold and the gradient magnitudes of each pixel point, the pixel points with gradient magnitudes less than the gradient magnitude threshold are filtered out. Specifically, when the gradient magnitude of a pixel point is less than the gradient magnitude threshold, then this pixel point is filtered out; that is, only the pixel points with gradient magnitudes greater than or equal to the gradient magnitude threshold are retained. For example, a gradient magnitude threshold is set to 5 °C / mm, and only when the gradient magnitude of a pixel point is greater than 5 °C / mm, this pixel point is retained.

[0096] Moreover, the electronic device screens out the adjacent points with a gradient direction change rate exceeding the gradient direction threshold among the adjacent points of each pixel point based on the gradient direction change rate of the adjacent points of each pixel point, and takes the adjacent points with a gradient direction change rate exceeding the gradient direction threshold as key path points. Among them, the key path points with a gradient direction change rate exceeding the gradient direction threshold can be path inflection points.

[0097] For example, the gradient direction threshold can be 15°. Only if it does not exceed will it be retained as a heat flow path node. Based on this, the electronic device can screen out the pixel points with large temperature changes and obvious direction changes as important path points and as key nodes of the heat flow path.

[0098] Furthermore, the electronic device can also calculate the heat flow direction and heat flow density of each key path point to generate heat flow density vector field data. Among them, the heat flow density vector field data includes the heat flow density and heat flow direction of each key path point. The heat flow density represents the intensity and speed of heat flow corresponding to the key path point. Specifically, it can be the heat passing through a unit area per unit time corresponding to this key path point, that is, how much heat flows per second and per area unit. The heat flow direction represents the heat transfer direction, that is, the direction in which the heat flows from this key path point.

[0099] Among them, for each key path point, the heat flow density of this key path point can be calculated according to the material thermal conductivity of this key path point. For example, based on the gradient magnitude of the key path point and the material thermal conductivity, through the following heat flow density formula, the heat flow density of the key path point can be calculated: ; Among them, q represents the heat flow density of the key path point. λ represents the material thermal conductivity. For example, the material thermal conductivity of aluminum is λ = 237 W / (m·K). ∇T represents the gradient magnitude of the temperature of the key path point.

[0100] Among them, the heat flow direction of the key path point can be the gradient direction, indicating that the heat flows from high temperature to low temperature. Based on this, the electronic device generates heat flow density vector field data according to the heat flow density and heat flow direction of each key path point. Each key path point can be represented as a vector including magnitude (that is, heat flow density) and direction (that is, heat flow direction).

[0101] Step 3: Based on each key path point in the hot spot distribution map, construct a thermal resistance network model.

[0102] The thermal resistance network model can be a mathematical model used to simulate the heat conduction path of a capacitor; the thermal resistance network model can be used to describe the transfer situation of heat flow in the target area where the capacitor is located.

[0103] The electronic device can use the Dijkstra algorithm for construction. Among them, the electronic device can determine the edge weight of the corresponding material according to the thermal conductivity of the capacitor material. Take each key path point as a node, connect adjacent nodes to get an edge, and determine the edge weight of each edge according to the material corresponding to each edge. Based on each node, the edges corresponding to each node, and the edge weights of each edge, use the Dijkstra algorithm to perform the shortest path search to obtain the thermal resistance network model.

[0104] It should be noted that in the thermal resistance network model, each edge can represent the heat conduction path between two nodes. The edge weight can be the thermal resistance value of the heat conduction path connecting two nodes in the thermal resistance network model, used to characterize the magnitude of the obstruction of the heat conduction path to the heat flow. The edge weight can be an attribute of this edge, that is, the magnitude of the thermal resistance value. The larger the edge weight, the smaller the heat flow allowed to pass through this path. The edge weight can be understood as the ease of heat flow passing through this heat conduction path. The smaller the edge weight, the easier the heat flow passes through.

[0105] For example, for the aluminum material area with a thermal conductivity of λ = 237 W / (m·K), its edge weight can be set to 1 / λ, that is, 1 / 237. For the plastic area, the heat conduction is poor and the thermal resistance is large. The edge weight of the plastic area = 1 / 0.2.

[0106] Among them, the electronic device can use the Dijkstra algorithm to perform path search between each node, that is, find the path with the minimum thermal resistance of the heat flow, so as to obtain the thermal resistance network model. That is, starting from the high-temperature point, find the path with the "least resistance". Among them, when performing the shortest path search for abnormal paths, a relaxation factor α = 85 can be set. Through this relaxation factor, a small amount of detour can be allowed to balance the path length and thermal resistance; based on this, it helps to better adjust the path during the path search process and find a better solution. By constructing the thermal resistance network model, the machine lays a foundation for subsequent analysis of the heat flow path.

[0107] In a possible way, when using the Dijkstra algorithm for path search, the path search can be combined with the heat flux density vector field data. For example, if the vector field shows that the heat flux density in a certain area is high, the thermal resistance weight of this area can be dynamically reduced, that is, the edge weight of the edge in this area is reduced, so that the algorithm is more inclined to select the edge in this area when performing path search.

[0108] Step 4: Based on the gradient magnitudes and directions of each key path point and the thermal resistance network model, perform heat flow transfer path detection to generate an initial set of heat flow transfer paths.

[0109] In this step, heat flux density vector field data can be generated based on the gradient magnitudes and directions of each key path point; the Runge-Kutta fourth-order method is used to perform path tracing according to the heat flux density vector field data to determine the initial heat flow transfer path. For example, the heat transfer path in the target area can be accurately traced along the direction of the heat flux density vector field.

[0110] Among them, during the path tracing process, the electronic device can input the heat flux density vector field data and the thermal resistance network model, and perform path tracing according to a preset step size. For example, it can advance at a step size of 1 mm, that is, move 1 mm along the path each time, and the tracing direction is determined by the gradient direction of the key path point (that is, the vector direction of the key path point in the vector field), ensuring that the path extends along the direction of the fastest temperature change. Based on this, path tracing is performed to determine the initial heat flow transfer path, and an initial set of heat flow transfer paths including at least one initial heat flow transfer path is obtained.

[0111] In addition, too large a temperature difference may lead to inaccurate paths. Therefore, when the temperature difference between adjacent two points is detected to exceed 8 °C, quadratic spline interpolation compensation can be triggered to more accurately fit the path and make the tracing of the heat flow path more precise. By continuously advancing and compensating, the path of the heat flow is gradually determined.

[0112] Step 5: For each path bifurcation point in the initial heat flow transfer path, based on the area where each path bifurcation point is located, use the entropy value analysis method to identify heat flow anomalies at each path bifurcation point, and identify the heat flow anomaly convergence areas in the areas where each path bifurcation point is located.

[0113] Among them, the electronic device can mark the heat flow anomaly convergence areas.

[0114] In some possible cases, during the process of tracing the heat flow path, the situation of path bifurcation may be encountered. To determine which path bifurcation points are abnormal heat flow convergence points, the entropy value analysis method can be used. When a path bifurcation point is detected, a 5×5 pixel neighborhood centered on the path bifurcation point can be selected as the area where the path bifurcation point is located, and the ratio between the standard deviation and the mean value of the temperatures of each pixel point within the selected neighborhood range is calculated. This ratio can also be called the entropy value. If this ratio is greater than 3, the electronic device can mark the area where the path bifurcation point is located as an abnormal heat flow convergence point. If this ratio is less than or equal to 3, it indicates that the temperature distribution at this point is uniform and the heat flow transfer is normal. Among them, the standard deviation reflects the degree of temperature dispersion, and the mean value represents the average temperature. When their ratio is large, it indicates that the temperature in this area fluctuates violently, and abnormal heat accumulation may be caused by material interfaces, holes or local heat sources, and it is very likely to be a place where abnormal heat flow converges.

[0115] Through the entropy value analysis method, abnormal areas in the heat flow path can be effectively identified. In addition, the abnormal heat flow convergence area may also be the location of capacitor design or material defects. Based on this, the positioning of capacitor design or material defects can also be effectively carried out for subsequent improvement and enhancement.

[0116] Step 6: Smooth the abnormal heat flow convergence areas in each initial heat flow transfer path to obtain each heat flow transfer path.

[0117] In this step, for each initial heat flow transfer path, a sliding window filtering process can be performed on the abnormal heat flow convergence area in the initial heat flow transfer path; and, the outlier points in the abnormal heat flow convergence area are identified, and smoothing processing is performed on the identified outlier points to obtain the heat flow transfer path.

[0118] Among them, performing a sliding window filtering process on the abnormal heat flow convergence area includes: for each key path point included in the abnormal heat flow convergence area in the initial heat flow transfer path, a sliding window with a size of 5 points (including the current key path point and two points before and after it) can be used, and the arithmetic mean of the coordinates of the 5 points within the window is taken as the coordinate of the current key path point. For example, if sawtooth noise points appear at (100, 80) to (105, 85) in the initial path due to sensor noise, the path can be corrected to a continuous curve after smoothing.

[0119] Among them, outliers are identified and smoothed for the outliers, including: for each key path point in the area where heat flow abnormally converges in the initial heat flow transfer path, if the position deviation of the key path point from the coordinate mean of the window where the key path point is located exceeds a first threshold, or the temperature difference from the temperature mean of the window where it is located exceeds a second threshold, then the key path point is marked as an outlier. For example, the first threshold can be 2 pixels and the second threshold can be 5 °C, that is, if the position deviation of a certain point from the window mean exceeds 2 pixels or the temperature difference exceeds 5 °C, it is marked as an outlier.

[0120] For outliers, a linear interpolation algorithm can be used to smooth the outliers. For example, if the temperature difference between a certain point i and the window mean where it is located exceeds 5 °C, the following formula can be used to calculate the new temperature value of point i: P fixed i =( P i −1]+ P i +1]) / 2. That is, based on the adjacent point i - 1 and the adjacent point i + 1 of point i, the temperature value of point i is calculated.

[0121] Step 7. Generate an abnormal heat flow path diagram of the capacitor based on each heat flow transfer path.

[0122] In this step, in one possible way, an intuitive and visual heat flow path diagram can be generated according to each heat flow transfer path. Among them, the preset structure diagram of the capacitor can also be obtained, and the preset structure diagram and each heat flow transfer path are superimposed to generate the abnormal heat flow path diagram.

[0123] Among them, the preset structure diagram can be the structure diagram corresponding to the CAD (Computer Aided Design) design model of the capacitor. Based on this, the design structure of the capacitor and the actual heat flow transfer path can be clearly displayed, so as to more intuitively display the relationship between the actual heat dissipation path and the design structure of the capacitor physical object.

[0124] Specifically, step 7 can be implemented through the following steps 7-1 to 7-3.

[0125] Step 7-1. Align each heat flow transfer path with the preset structure of the capacitor to obtain the deviation result between each heat flow transfer path and the heat dissipation channel of the capacitor, and mark each heat flow transfer path based on the deviation result.

[0126] ​​​Among them, the preset structure of the capacitor can be the CAD design model of the capacitor. Among them, the preset structure includes the vector coordinates for designing the heat dissipation channels, such as the center line of the heat sink; the preset structure can also include the material partition boundary, such as the boundary line between the metal and plastic regions.

[0127] In this step, each heat transfer path can be aligned with the preset structure of the capacitor first. For example, at least 3 common feature points can be selected for alignment, and the rotation matrix and translation vector are calculated by the least squares method. Based on the rotation matrix and translation vector, the coordinates of the key path points included in each heat transfer path are transformed into the coordinate system corresponding to the preset structure. For example, the key path points in each heat transfer path are transformed into the coordinate system of the CAD design model.

[0128] Then, based on the preset structure, the heat dissipation channels of the capacitor are identified, and deviation detection and analysis are carried out between each heat transfer path and the heat dissipation channels of the capacitor. For example, for each key path point in the heat transfer path, the Euclidean distance from the key path point to the nearest heat dissipation channel can be calculated to obtain the deviation result of each heat transfer path.

[0129] Furthermore, marking can also be carried out according to the deviation results.

[0130] For example, if the deviation of the key path point ≤ 1mm, record it but do not alarm, and mark the key path point and the nearest heat dissipation channel corresponding to the key path point as green, indicating normal; If 1mm < the deviation of the key path point ≤ 3mm, prompt manual review, and mark the key path point and the nearest heat dissipation channel corresponding to the key path point as yellow for warning; If the deviation of the key path point > 3mm, trigger the optimization design process, and mark the key path point and the nearest heat dissipation channel corresponding to the key path point as red, indicating that there are defects in the design of the capacitor.

[0131] In addition, the above deviation situations can also be recorded. For example, the actual coordinates of a certain heat dissipation channel are (102.3, 80.5), the design expectation is (100, 80), the deviation is 2.5mm, marked as yellow warning, and it is found through inspection that there is an assembly offset.

[0132] Furthermore, the superposition of each heat transfer path and the CAD design structure can also be carried out. Based on this, the differences between each heat dissipation channel and the actual heat transfer path in the design structure can be better represented.

[0133] Step 7-2: Based on each heat transfer path and its deviation result, adjust and optimize the marking of each heat transfer path.

[0134] For example, for the area marked in red, the temperature range of the red mark can be adjusted according to the area of the region. For example, if the area of the red-marked region exceeds the threshold, the color scale range corresponding to the red can be narrowed. For example, for the region where the deviation of the critical path point > 3 mm, the red-marked range can be updated from the temperature range of 0 - 120 °C to 80 - 120 °C. Based on this, the risk within the region where the deviation of the critical path point > 3 mm can be focused on the temperature range of 80 - 120 °C, highlighting the high-risk region more significantly.

[0135] Also, for example, the transparency of different color-marked regions can be adjusted. For example, the transparency of the green region (the deviation of the critical path point ≤ 1 mm) is set to 50%; the transparency of the yellow region (the deviation range of the critical path point is from 1 mm to 3 mm) is set to 30%; the red region (the deviation range of the critical path point > 3 mm) is set to opaque.

[0136] Furthermore, each heat transfer path can be distinguished according to the temperature gradient difference of each path. For example, if the temperature gradient difference ΔT corresponding to the path ≥ 10 °C, then this heat transfer path is the main path. If the temperature gradient difference ΔT corresponding to the path ≤ 5 °C (ΔT < 10 °C), then this heat transfer path is the secondary path.

[0137] Based on this, different markings can also be made for the main path and the secondary path. For example, for the main path, it can be represented by a gradient color band with a width of 3 pixels, and the color transitions from dark red (ΔT ≥ 10 °C) to light yellow (ΔT ≤ 2 °C), reflecting the main channel of heat transfer. For the secondary path, it can be represented by a light gray line with a width of 1 pixel, indicating the branch path with low heat flux density.

[0138] Step 7 - 3: Generate an abnormal heat flow path diagram of the capacitor based on each adjusted and optimized heat transfer path.

[0139] Among them, based on the optimized heat transfer path, the electronic device can further analyze the correlation between the abnormal heat flow and the gradient direction. For example, analyze the distribution of the abnormal heat flow in the gradient direction, the change relationship between the abnormal heat flow and the gradient direction, etc. Based on the analysis process, judge the overall trend and local trend of heat transfer. The overall trend can be the overall heat transfer trend of the capacitor; the local trend can be the transfer value of a certain heat transfer path, or the capacitor, or a partial path region in a certain path, etc. The transfer trend can include but is not limited to: in which direction the heat is transferred, whether the transfer speed is accelerating or slowing down, etc. Based on this, trend characteristic data are obtained, and these data describe the trend and characteristics of heat transfer.

[0140] Then, based on the trend feature data, the electronic device can match the spatial features of each heat flow transfer path with the design structure of the capacitor to obtain an abnormal heat flow path map of the capacitor. For example, it can detect whether the change of the heat flow transfer path in space conforms to the heat transfer trend corresponding to the design structure. If not, the heat flow transfer path can be adjusted. After repeated matching and adjustment, the final abnormal heat flow path map is determined.

[0141] For example, in this abnormal heat flow path map, the main path is represented by a gradient color band with a width of 3 pixels, and the secondary path has a width of 1 pixel. The color of the color band transitions from dark red (ΔT≥10℃) to light yellow (ΔT≤2℃), which can visually display the importance degree of the heat flow path and the temperature change situation.

[0142] In some possible implementation manners, the heat flow anomaly index corresponding to each heat flow transfer path can also be obtained, and a correlation analysis is performed on the heat flow anomaly index and the working state of the capacitor to obtain a prediction model and correlation parameters corresponding to the heat flow anomaly index and the working state.

[0143] Among them, the working state can include states such as the power, ambient temperature, and rotation speed of the capacitor. Specifically, the working states such as the power of the capacitor, the ambient temperature of the capacitor, and the rotation speed can be obtained from the device work log of the capacitor.

[0144] The heat flow anomaly index can include but is not limited to: the area of the high-temperature region of the capacitor, the maximum gradient amplitude value ΔT corresponding to each heat flow transfer path, etc.

[0145] For example, the following formula can be used to calculate the correlation coefficient between the thermal anomaly index and the working state: Among them, X i represents the heat flow anomaly index, for example, the area of the high-temperature region. Y i represents the device parameter, that is, the working state, for example, the power. 、 respectively represent the mean values of the heat flow anomaly index and the device parameter, for example, the mean value of the area of the high-temperature region in multiple detection processes, and the mean value of the power in multiple detection processes.

[0146] Among them, the determination criterion can be: if ∣R∣≥0.7, then it is strongly correlated. For example, when the power increases, the high-temperature region expands. If 0.3≤∣R∣<0.7, then it is moderately correlated. If ∣R∣<0.3, then it is weakly correlated.

[0147] Furthermore, a model prediction can also be performed on the heat flow anomaly index and the working state. For example, using a regression model, a linear or non-linear relationship model between the heat flow anomaly and the device parameter can be established. For example, the relationship between the predicted high-temperature region area S and the power P can be S = 0.2P + 5.

[0148] Based on this, an abnormal heat flow path diagram of the capacitor can be generated based on each adjusted and optimized heat flow transfer path, as well as the prediction model and correlation parameters corresponding to the heat flow anomaly index and the working state. For example, visual markings are made in the high-temperature area. If the red area is clicked, the coordinates, deviation, and a label such as "the power increase causes the heat flow to deviate from the design channel" will be displayed.

[0149] Step S105: The electronic device extracts key feature points from the abnormal heat flow path diagram, and determines the fault detection result of the capacitor based on the difference between the key feature points and the preset template of the capacitor.

[0150] In this step, the electronic device extracts each feature point in the abnormal heat flow path diagram where the temperature gradient change reaches the target condition; compares each feature point with the preset template constructed based on the Gaussian mixture model to identify the abnormal feature points among the feature points; performs clustering processing on each abnormal feature point, and generates the fault detection result of the capacitor based on the clustering result.

[0151] The following is a specific introduction to each process.

[0152] Step 1: The electronic device extracts each feature point in the abnormal heat flow path diagram where the temperature gradient change reaches the target condition.

[0153] Among them, since the areas with significant temperature gradient changes in the abnormal heat flow path diagram may contain important information. For example, significant temperature gradient changes are often related to possible faults of the capacitor. Therefore, the electronic device can use the Harris corner detection algorithm to find the areas with significant temperature gradient changes.

[0154] Specifically, the electronic device can traverse each key path point in the abnormal heat flow path diagram, calculate the corner response function for each key path point, and analyze the temperature change situation around each key path point to determine whether it has an obvious temperature gradient change. If the threshold condition is met, this position is marked as a candidate feature point. Based on this, a batch of areas with prominent temperature changes can be quickly screened out, laying a foundation for more accurate feature point extraction in the follow-up. Among them, the electronic device can calculate the temperature gradients in the horizontal and vertical directions for each key path point, and construct a second-order matrix M through the following formula: Among them, the second-order matrix M can describe the gradient distribution of the local area. are the temperature gradients in the x and y directions respectively. is a 5×5 Gaussian weight window, which can be the position weight of the points (x,y) within the 5×5 window.

[0155] Then, according to the second-order matrix M, calculate the corner response value R of the critical path points through the following formula: ; where R represents the corner response value. det(M) is the determinant of the matrix, trace(M) is the trace of the matrix, and k is an empirical constant (usually taken as 0.04 - 0.06).

[0156] Among them, the target condition may include: the corner response value of the critical path point is greater than a preset corner response function threshold.

[0157] In this step, feature points can be screened based on the preset corner response function threshold and the corner response values of each critical path point. For example, the corner response function threshold is 0.1. When the corner response value of a certain critical path point is greater than 0.1, it is a feature point. In a possible implementation, the feature points can be further screened by combining the temperature mean and standard deviation of the points within the neighborhood of the critical path point. The mean represents the average temperature level of the area, while the standard deviation reflects the degree of temperature dispersion, that is, the severity of temperature change.

[0158] The target condition may include: the corner response value of the critical path point is greater than a preset corner response function threshold, and moreover, the standard deviation corresponding to the neighborhood of the critical path point is greater than a preset standard deviation threshold. For example, the critical path points with corner response values greater than 0.1 are used as candidate feature points, and the standard deviation corresponding to the neighborhood of each candidate feature point is obtained. When the standard deviation is greater than 2°C, the candidate feature point is used as a feature point.

[0159] It should be noted that if the standard deviation is large, it indicates that the temperature change in this area is relatively complex and severe, and it is more likely to contain information related to faults. If the standard deviation is small, it means that the temperature is relatively stable and it is less likely to be the key position where faults occur, so this candidate point can be discarded. In this way, the most representative points that can best reflect the abnormal heat flow situation can be selected, and these points will be used as important bases for subsequent fault detection. Step 2: Compare each feature point with a preset template constructed based on the Gaussian mixture model to identify the abnormal feature points among each feature point.

[0160] Among them, after determining the feature points, it is necessary to judge whether these feature points are normal and whether there are features related to faults. In this step, a preset template constructed based on the Gaussian mixture model can be used to judge each feature point.

[0161] The preset template is constructed based on the Gaussian mixture model and can represent the data distribution of the path points of the heat flow path diagram in the normal working mode of the capacitor. For example, based on the historical heat flow path diagrams of normal operation, characteristic points can be extracted as training data, and the Gaussian mixture model is modeled using the training data to obtain the preset template. When constructing the preset template, the number of clusters can be set to 3, and the data in the normal working mode is divided into 3 categories, representing the main heat modes in the normal working mode; for example, uniform heat dissipation, local heat dissipation channels, and edge heat dissipation. The covariance type is set to "full", indicating that each category of data has its own independent covariance matrix, which can more comprehensively describe the distribution characteristics of the data. Different clusters are allowed to freely vary in shape and orientation.

[0162] After the preset template is constructed, the similarity between them can be judged by calculating the Euclidean distance between the characteristic points and the preset template. For each characteristic point, calculate the distance between the characteristic point and the cluster centers of the preset model. The closer the distance, the more similar the characteristic point is to the normal mode; the farther the distance, the more likely there is an anomaly. Among them, if the minimum distance among the distances between the characteristic point and the cluster centers is greater than the preset distance threshold, the characteristic point is marked as an abnormal characteristic point. When the Euclidean distance between the characteristic point and the preset template exceeds the preset distance threshold, the characteristic point is marked as an abnormal characteristic point. For example, the preset distance threshold is 5. When the Euclidean distance between the characteristic point and the template exceeds 5, it is an abnormal characteristic point. Based on this, it is possible to quickly identify which characteristic points deviate from the normal working heat mode, providing clues for further judging the fault heat characteristics. Step 3: Perform clustering processing on each abnormal characteristic point, and generate a fault detection result of the capacitor based on the clustering result.

[0163] Among them, the number of clusters can be set to 2, and the number of iterations is set to 100. The abnormal characteristic points are clustered to obtain 2 clusters. One cluster represents the fault area, and the other cluster represents the non-fault area; for example, one cluster may be the area related to the fault; the other cluster may be a relatively normal non-fault area, but there are also certain anomalies. In this step, the temperature difference between the cluster centers of these two clusters can also be determined. If the temperature difference exceeds the preset temperature difference threshold, it is determined that the area where one cluster is located is the fault area. Among them, it can be determined that the area where the cluster with a higher cluster center temperature is located is the fault area. The fault detection result includes the fault area corresponding to the capacitor. The fault detection result of the capacitor includes the location of the fault area of the capacitor.

[0164] In this step, this is because a large temperature difference indicates that the heat flow characteristics between different categories are significantly different, and it is very likely that there is a fault resulting in uneven heat flow distribution, thus determining this area as the place where the fault occurs.

[0165] Among them, for the fault detection results, different - colored rectangular frames can be used to mark the abnormal area and the normal area. For the fault area, the machine marks it with a red rectangular frame; for the normal area, it is marked with a green rectangular frame. For example, the size of the rectangular frame is set to 10×10 pixels, so that the fault and normal area ranges can be clearly shown in the figure.

[0166] At the same time, in order to better distinguish different states, the machine will also fill the rectangular frame, and the transparency of the filling color is set to 5. Through this marking method, the staff can quickly and intuitively see which places have faults and which places are normal on the abnormal heat - flow path diagram, providing a clear and definite basis for subsequent fault troubleshooting and handling.

[0167] In one possible way, before performing step 3, the principal - component analysis method can also be used to reduce the dimension of the abnormal feature points to obtain the fault determination results of each abnormal feature point. Correspondingly, step 3 may include: clustering the candidate fault points among each abnormal feature point according to the fault determination results of each abnormal feature point to obtain the fault area of the capacitor.

[0168] Among them, the fault determination result indicates whether the abnormal feature point belongs to the candidate fault point. For example, the fault determination result is: the abnormal feature point belongs to the candidate fault point, or the abnormal feature point does not belong to the candidate fault point.

[0169] In this step, based on the multi - dimensional feature data of each abnormal feature point, the PCA (Principal Components Analysis) method is used to reduce the dimension of the abnormal feature points and extract the first two principal components; among them, the multi - dimensional features of the abnormal feature points include but are not limited to: the coordinate position, temperature, temperature - gradient amplitude, gradient direction, heat - flow density, etc. of the abnormal feature point.

[0170] Then, calculate the contribution rate of each abnormal feature point to each of the first two principal components; if the contribution rate of the first principal component of the abnormal feature point is lower than the preset contribution - rate threshold, then this abnormal feature point is regarded as a candidate fault point. Among them, the first principal component is the principal component with a larger contribution rate among the two principal components of each abnormal feature point. Among them, the preset contribution - rate threshold can be 70%; when the contribution rate of the first principal component is lower than 70%, it is determined that the feature corresponding to this abnormal feature point is a fault heat feature, and this abnormal feature point is a candidate fault point.

[0171] In this step, if the contribution rate of the first principal component is low, it means that the data distribution is relatively scattered and difficult to be explained by a single main factor. It is very likely that due to the fault, complex abnormal changes in heat flow occur, thus determining that this feature is related to the fault. The present invention provides a method for detecting a capacitor. By obtaining the original thermal image set of the capacitor in the energized state, the temperature change rate of each point of the capacitor is determined based on this original thermal image set. The temperature change rate distribution map obtained therefrom can detect the dynamic thermal distribution of the capacitor in real time, capture the transient characteristics of the capacitor in real time, so as to accurately describe the dynamic change of the capacitor temperature, and lay a foundation for the subsequent dynamic analysis of the capacitor. Then, based on this temperature change rate distribution map, the abnormal points of the capacitor are identified to obtain the hot spot distribution map corresponding to the capacitor; and based on this hot spot distribution map, the heat flow transfer path analysis of the capacitor is carried out to determine the propagation trajectory of the temperature change rate of the capacitor, and the abnormal heat flow path map of the capacitor is obtained; based on the abnormal heat flow path map obtained therefrom, the abnormal conditions in the dynamic thermal process of the capacitor can be accurately captured, so as to accurately identify the abnormal heat flow and internal hot spots. Based on this, the key feature points are extracted from the abnormal heat flow path map, and based on the difference between the key feature points and the preset template of the capacitor, the fault detection result of the capacitor is determined. Since the detection is based on the key feature points in the abnormal heat flow path map with high accuracy, especially for the working scenarios of capacitors under complex working conditions, the potential faults in the working process of the capacitor can be accurately and effectively detected, improving the accuracy of capacitor detection.

[0172] Figure 2 It is a schematic structural diagram of a capacitor detection system provided by an embodiment of the present invention. As Figure 2 shown, the system includes: An acquisition module 201, configured to acquire an original thermal image set of the capacitor in the energized state. The original thermal image set includes a plurality of original thermal images collected based on a target frequency, and each original thermal image includes temperature data corresponding to each pixel point in the original thermal image; A first determination module 202, configured to determine the temperature change rate of each point of the capacitor based on the original thermal image set, and obtain a temperature change rate distribution map corresponding to the capacitor. One point of the capacitor corresponds to one pixel point of each original thermal image; A second determination module 203, configured to identify abnormal points of the capacitor based on the temperature change rate distribution map, and obtain a hot spot distribution map corresponding to the capacitor; A third determination module 204, configured to perform a heat flow transfer path analysis on the capacitor based on the hot spot distribution map, determine the propagation trajectory of the temperature change rate of the capacitor, and obtain an abnormal heat flow path map of the capacitor; A detection module 205, configured to extract key feature points from the abnormal heat flow path map, and determine the fault detection result of the capacitor based on the difference between the key feature points and the preset template of the capacitor.

[0173] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the embodiment of the above capacitor detection method are implemented.

[0174] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, the device where the computer-readable storage medium is located is controlled to implement the steps of the embodiment of the above capacitor detection method.

[0175] In the above specific embodiments, the purpose, technical solution, and beneficial effects of the present invention are further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting a capacitor, characterized in that, Including: Obtain an original thermal image set of the capacitor in the energized state, where the original thermal image set includes a plurality of original thermal images collected based on a target frequency, and each original thermal image includes temperature data corresponding to each pixel point in the original thermal image; Based on the original thermal image set, determine the temperature change rate of each point of the capacitor to obtain a temperature change rate distribution map corresponding to the capacitor, where one point of the capacitor corresponds to one pixel point of each original thermal image; Based on the temperature change rate distribution map, identify abnormal points of the capacitor to obtain a hot spot distribution map corresponding to the capacitor; Based on the hot spot distribution map, analyze the heat flow transfer path of the capacitor, determine the propagation trajectory of the temperature change rate of the capacitor, and obtain an abnormal heat flow path map of the capacitor; Extract key feature points from the abnormal heat flow path map, and based on the difference between the key feature points and a preset template of the capacitor, determine the fault detection result of the capacitor.

2. The detection method of the capacitor according to claim 1, wherein The step of identifying abnormal points of the capacitor based on the temperature change rate distribution map to obtain a hot spot distribution map corresponding to the capacitor includes: Based on the temperature change rate distribution map, identify abnormal regions of the capacitor to identify a set of abnormal heat flow candidate regions corresponding to the capacitor; Cluster the abnormal points in the set of abnormal heat flow candidate regions to obtain an internal hot spot distribution map corresponding to the capacitor; Correct the internal hot spot distribution map according to environmental temperature fluctuation data and capacitor material characteristics, and based on the corrected internal hot spot distribution map, obtain the hot spot distribution map of the capacitor.

3. The detection method of the capacitor according to claim 2, characterized in that, The step of correcting the internal hot spot distribution map according to environmental temperature fluctuation data and capacitor material characteristics and obtaining the hot spot distribution map of the capacitor based on the corrected internal hot spot distribution map includes: Use an infrared thermal imager to collect infrared thermal images of the capacitor, and use a distributed temperature sensor network to record the environmental temperature fluctuation of the capacitor to obtain an environmental temperature change matrix; Based on a material emissivity database, establish a target look-up table, and based on the target look-up table and the infrared thermal image of the capacitor, obtain the standard emissivity of each pixel point in the internal hot spot distribution map; Perform spatial correction on the standard emissivity of each pixel point in the internal hot spot distribution map through a bilinear interpolation algorithm to obtain a corrected emissivity, and based on the corrected emissivity, compensate the temperature data of each pixel point in the internal hot spot distribution map to obtain compensated temperature data; Based on the compensated temperature data corresponding to the internal hot spot distribution map, perform coupled calculation with a finite element heat conduction model to obtain the three-dimensional temperature field distribution inside the material of the capacitor; Extract isothermal surfaces from the three-dimensional temperature field distribution of the capacitor, and based on the extracted isothermal surfaces, obtain a three-dimensional hot spot distribution map, and use the three-dimensional hot spot distribution map as the hot spot distribution map corresponding to the capacitor.

4. The detection method of the capacitor according to claim 1, wherein The step of analyzing the heat flow transfer path of the capacitor based on the hot spot distribution map, determining the propagation trajectory of the temperature change rate of the capacitor, and obtaining an abnormal heat flow path map of the capacitor includes: Calculate the temperature gradient field of the hot spot distribution map using the Sobel operator to obtain the gradient amplitude and direction of each pixel point in the hot spot distribution map; Based on the gradient amplitude and direction of each pixel point in the hot spot distribution map, screen out each key path point from each pixel point in the hot spot distribution map through the non-maximum suppression algorithm; Construct a thermal resistance network model based on each key path point in the hot spot distribution map; Perform heat flow transfer path detection based on the gradient amplitude and direction of each key path point and the thermal resistance network model to generate an initial heat flow transfer path set; For each path bifurcation point in each initial heat flow transfer path, based on the area where each path bifurcation point is located, use the entropy value analysis method to identify heat flow anomalies at each path bifurcation point, and identify heat flow anomaly convergence areas in the areas where each path bifurcation point is located; Smooth the heat flow anomaly convergence areas in each initial heat flow transfer path to obtain each heat flow transfer path; Generate an abnormal heat flow path map of the capacitor based on each heat flow transfer path; 5. The detection method of the capacitor according to claim 4, characterized in that, The generating an abnormal heat flow path map of the capacitor based on each heat flow transfer path includes: Align each heat flow transfer path with the preset structure of the capacitor to obtain the deviation result between each heat flow transfer path and the heat dissipation channel of the capacitor, and mark each heat flow transfer path based on the deviation result; Adjust and optimize the marks of each heat flow transfer path based on each heat flow transfer path and its deviation result; Generate the abnormal heat flow path map of the capacitor based on each heat flow transfer path after adjustment and optimization; 6. The detection method of the capacitor according to claim 1, characterized in that, The extracting key feature points from the abnormal heat flow path map and determining the fault detection result of the capacitor based on the difference between the key feature points and the preset template of the capacitor includes: Extract each feature point in the abnormal heat flow path map where the temperature gradient change reaches the target condition; Compare each feature point with a preset template constructed based on the Gaussian mixture model to identify abnormal feature points among each feature point; Perform clustering processing on each abnormal feature point, and generate a fault detection result of the capacitor based on the clustering result; 7. The detection method of the capacitor according to claim 6, wherein Before the performing clustering processing on each abnormal feature point and generating a fault detection result of the capacitor based on the clustering result, the method further includes: Use the principal component analysis method to reduce the dimension of each abnormal feature point to obtain the fault determination result of each abnormal feature point, and the fault determination result represents whether the abnormal feature point belongs to a candidate fault point; Correspondingly, the performing clustering processing on each abnormal feature point and generating a fault detection result of the capacitor based on the clustering result includes: According to the fault determination result of each abnormal feature point, perform clustering processing on the candidate fault points among each abnormal feature point to obtain the fault area of the capacitor; 8. The detection method of the capacitor according to claim 1, characterized in that, The determining the temperature change rate of each point of the capacitor based on the original thermal image set to obtain the temperature change rate distribution map corresponding to the capacitor includes: Determine the temperature change rate of each point of the capacitor based on the original thermal image set; Using the inter-frame difference algorithm, dynamic features are extracted from the set of original thermal images to obtain a sequence of differential thermal images, and the transient change rate of the temperature at each point of the capacitor is obtained based on the sequence of differential thermal images; the sequence of differential thermal images characterizes the dynamic change features of the temperature distribution between adjacent images in the set of original thermal images; Based on the temperature change rate and the transient change rate at each point of the capacitor, a temperature change rate distribution map corresponding to the capacitor is obtained.

9. A detection system for a capacitor, characterized in that, It includes: An acquisition module, configured to acquire a set of original thermal images of the capacitor in the energized state, the set of original thermal images including a plurality of original thermal images collected based on a target frequency, and each original thermal image including temperature data corresponding to each pixel point in the original thermal image; A first determination module, configured to determine the temperature change rate at each point of the capacitor based on the set of original thermal images, and obtain a temperature change rate distribution map corresponding to the capacitor, where one point of the capacitor corresponds to one pixel point of each original thermal image; A second determination module, configured to identify abnormal points of the capacitor based on the temperature change rate distribution map, and obtain a hot spot distribution map corresponding to the capacitor; A third determination module, configured to analyze the heat flow transfer path of the capacitor based on the hot spot distribution map, determine the propagation trajectory of the temperature change rate of the capacitor, and obtain an abnormal heat flow path map of the capacitor; A detection module, configured to extract key feature points from the abnormal heat flow path map, and determine the fault detection result of the capacitor based on the difference between the key feature points and a preset template of the capacitor.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the capacitor detection method according to any one of claims 1 to 8.

Images