Radiating efficiency evaluation method and system for radiator

Abstract

The invention relates to the technical field of data processing, in particular to a heat dissipation efficiency evaluation method and system for a radiator, and the method comprises the steps: collecting a thermal image of the surface of a radiator to be evaluated, carrying out the image reconstruction of the thermal image at the current moment through a total variation regularization model, obtaining a reconstruction image, calculating the abnormal degree of each pixel point, and obtaining the heat dissipation efficiency of the radiator. The method comprises the steps of comprehensively determining temperature abnormal points, dividing temperature abnormal areas based on the temperature abnormal points, calculating fault classification indexes of the temperature abnormal areas, feeding back the fault classification indexes to a total variation regularization model, adjusting regularization coefficients of the total variation regularization model, and calculating the fault classification indexes of the temperature abnormal areas according to the temperature abnormal points and the variation trend of the abnormal degree of each pixel point. And the heat dissipation efficiency is determined on the basis of the temperature anomaly region in the optimized reconstructed image. According to the method, the accuracy of heat dissipation efficiency evaluation is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. Specifically, it relates to a method and system for evaluating the heat dissipation efficiency of a radiator. Background Technology

[0002] As a core piece of equipment in the power system, the normal operation of the transformer is crucial to the stability of the power grid. The internal heat dissipation system of the transformer mainly consists of radiators, which are responsible for dissipating the heat generated by the transformer during operation in a timely manner to prevent overheating from damaging the equipment.

[0003] As transformers operate over long periods, the heat dissipation efficiency of radiators can be affected by various factors, such as dust accumulation, cooling medium leakage, blockage of heat pipes, or fan malfunction. These problems are often latent and difficult to detect in the early stages, but they gradually lead to a decrease in heat dissipation efficiency, thereby affecting the transformer's heat dissipation capacity, and even causing equipment overheating, resulting in electrical faults, and in severe cases, power outages. Therefore, ensuring the stability and efficiency of transformer radiators during long-term operation and regularly evaluating their heat dissipation efficiency is particularly important.

[0004] Existing technologies typically use thermal imaging to evaluate the heat dissipation efficiency of radiators. Specifically, a thermal imager scans the radiator surface to obtain a thermal image. Based on this image, the temperature distribution in different areas of the radiator surface is analyzed, providing a direct view of the radiator's heat dissipation status. This temperature information allows for a preliminary assessment of the radiator's heat dissipation efficiency, such as analyzing the proportion of locally high-temperature areas and the temperature difference between these areas and normal-temperature areas.

[0005] To accurately identify heat dissipation faults, existing technologies typically require image quality enhancement of thermal images. After image quality enhancement, abnormal temperature areas are identified to determine the heat dissipation faults, and then the impact of different heat dissipation faults on the heat dissipation effect is analyzed to determine the final heat dissipation efficiency.

[0006] In the current field of image processing, the total variational regularization model is typically used for image reconstruction. This model balances the smoothness of the image with data fidelity by minimizing the energy function, thereby achieving image quality enhancement.

[0007] However, when using a total variational regularization model to enhance the image quality of thermal images of transformer radiators, the regularization coefficients of traditional total variational regularization models are usually set to fixed empirical values. Heat pipe blockage exhibits high-frequency gradient characteristics with sharp edges, requiring lower smoothing intensity to preserve texture details, while heat dissipation medium leakage exhibits low-frequency diffusion characteristics with blurred edges, requiring higher smoothing intensity to enhance smoothing and noise reduction. If the fixed regularization coefficient is set too low, thermal noise cannot be effectively filtered out, causing weak leakage signals to be submerged; if it is set too high, a step effect occurs, resulting in over-smoothing and erasing the key gradient characteristics of the blockage point. This contradiction between fixed parameters and variable fault characteristics makes it impossible to accurately identify these two types of faults, thus affecting the accuracy of the heat dissipation efficiency assessment results. Summary of the Invention

[0008] To address the problem that existing total variational regularization models, which use fixed regularization coefficients, cannot simultaneously satisfy the requirements of heat pipe blockage for preserving high-frequency features and heat dissipation medium leakage for denoising low-frequency features, leading to inaccurate fault identification and thus affecting the accuracy of heat dissipation efficiency assessment, this invention proposes a method and system for heat dissipation efficiency assessment of radiators.

[0009] On one hand, the present invention provides a method for evaluating the heat dissipation efficiency of a radiator, comprising:

[0010] Collect thermal images of the surface of the radiator to be evaluated during the operation of the transformer, and use the time of heat dissipation efficiency evaluation as the current time. Based on the temperature change rate and temperature gradient magnitude of the pixels in the thermal image at the current moment, as well as the load change rate of the transformer at the current moment, the degree of anomaly of the pixels is determined; based on the degree of anomaly of the pixels and the rate of change of the degree of anomaly of the pixels in historical thermal images, the trend of the degree of anomaly of the pixels is determined. By analyzing the changing trends of pixel anomalies, temperature anomalies are identified. Based on these anomalies, a neighborhood search method is used to delineate the temperature anomaly regions in the current thermal image. Fault classification indices for these regions are calculated and fed back into a total variational regularization model. The regularization coefficients of the total variational regularization model are dynamically adjusted to obtain an optimized model. The optimized model is then used to reconstruct the current thermal image, resulting in a reconstructed image. Based on the temperature anomaly regions in the reconstructed image, the heat dissipation efficiency is determined.

[0011] In the image preprocessing stage, this technical solution introduces a total variational regularization model to reconstruct the thermal image. This is based on the physical fact that thermal imaging data is usually accompanied by Gaussian noise or non-uniform interference. The total variational model utilizes the sparsity prior of the image gradient to effectively preserve the texture features of the heat sink edges and fault points while removing noise, avoiding the blurring caused by traditional filtering and establishing a high-quality data foundation for subsequent analysis. Secondly, in the anomaly detection and trend analysis stage, spatial features, temporal features, and operating condition features are integrated. In particular, the load change rate is introduced as a parameter, solving the problem of false alarms caused by the natural temperature rise of the transformer under normal load fluctuations, and achieving decoupling between fault temperature rise and operating condition temperature rise. By comparing historical thermal images, the system no longer focuses only on instantaneous temperature exceedances, but also on the acceleration of the evolution of the anomaly degree, improving the ability to detect gradual faults. Finally, a feedback mechanism from fault classification to image reconstruction was established: the calculated fault classification index is fed back to the underlying total variation model, and the regularization coefficient is dynamically adjusted. This means that the system can enhance the fault features of the image based on whether the currently determined fault type needs to retain edge blockage features or relatively smooth leakage features, thereby obtaining a reconstructed image that best reflects the real fault situation. The heat dissipation efficiency is calculated based on the reconstructed image, ensuring that the evaluation result is based on accurate data after removing interference and feature enhancement, thus improving the accuracy of heat dissipation efficiency evaluation.

[0012] Preferably, the degree of anomaly of a pixel is determined based on the temperature change rate and temperature gradient amplitude of the pixels in the thermal image at the current moment, as well as the load change rate of the transformer at the current moment. This includes: constructing a temperature anomaly term for each pixel based on its temperature change rate and temperature gradient amplitude; and constructing a dynamic weight adjustment mechanism for the temperature anomaly term and the load change rate to comprehensively determine the degree of anomaly of each pixel. The dynamic weight adjustment mechanism is configured to: establish a negative correlation between the degree of anomaly and the load change rate; and reduce the weight of the temperature anomaly term when the load change rate increases to suppress non-faulty temperature rise interference caused by load fluctuations.

[0013] This technical solution addresses the interference of transformer operating condition fluctuations on thermal fault diagnosis by constructing a dynamic weight adjustment mechanism. At the principle level, transformer temperature and load current have a physical positive correlation. When the load increases drastically, the overall rise in radiator surface temperature is a normal phenomenon consistent with physical laws, not a fault. Judging solely based on temperature thresholds or a single temperature change rate can easily lead to false alarms. The dynamic weight adjustment mechanism can distinguish between temperature rises caused by normal operating conditions and those caused by faults. During peak transformer loads or periods of drastic fluctuations, the system automatically lowers its sensitivity to avoid false alarms; while during periods of stable load, the weight is restored, allowing the system to sensitively detect minute abnormal temperature rises, thus improving the robustness and accuracy of the assessment method in the complex operating environment of actual substations.

[0014] Preferably, the trend of pixel anomaly is determined based on the degree of pixel anomaly and the rate of change of pixel anomaly in historical thermal images: the rate of change of anomaly of each pixel in the current thermal image is calculated, all historical thermal images of the current thermal image are obtained, and the rate of change of pixel anomaly in each historical thermal image is calculated; a time decay weighting mechanism is introduced to perform weighted cumulative calculation of the rate of change of pixel anomaly in each historical thermal image to obtain a weighted cumulative value; the trend of pixel anomaly is determined by the difference between the rate of change of pixel anomaly and the weighted cumulative value: a difference greater than 0 indicates an increasing trend; a difference equal to 0 indicates a stable trend; and a difference less than 0 indicates a decreasing trend.

[0015] This technical solution introduces a time-decay weighted mechanism to quantify the changing trend of anomaly severity. This is an analysis logic based on time-series memory, designed to accurately capture the evolution direction of faults. The rate of change of anomaly severity at a single moment may be affected by instantaneous interference and has randomness. In order to obtain reliable trend judgment, historical data must be introduced as a reference. However, different historical moments have different reference values ​​for the present. The closer the historical data is to the present moment, the more timely the equipment status information it contains. Therefore, a time-decay weighted mechanism is introduced, giving higher weight to recent historical data and lower weight to older data, and calculating a weighted cumulative value as a baseline. By comparing the current rate of change with this historical weighted baseline, the system can determine whether the anomaly is accelerating, remaining stable, or gradually mitigating. This differential comparison logic filters out long-standing background noise and highlights the dynamic evolution characteristics of the anomaly state, enabling the system not only to detect faults but also to predict their development trend, providing maintenance personnel with a more forward-looking decision-making basis.

[0016] A preferred method for determining temperature anomalies by analyzing the changing trend of the anomaly level of each pixel is as follows: if the anomaly level of a certain pixel in the thermal image at the current moment has been increasing in a predetermined number of consecutive historical thermal images prior to the current moment, then the pixel is determined to be a temperature anomaly.

[0017] This technical solution establishes a judgment logic for confirming trends across multiple consecutive frames, reflecting the principle of temporal consistency verification. In actual operation, thermal imaging data may exhibit sporadic jumps or single-frame anomalies. If a fault point is determined based solely on the abnormal trend of one or two frames, the false alarm rate of the system will be too high. Here, it is required that the abnormal trend of a certain pixel must maintain an upward trend in multiple consecutive historical thermal imaging images to be confirmed as a temperature anomaly point. This utilizes the characteristics of physical faults being persistent and irreversible. A true heat dissipation fault will cause heat to accumulate continuously, and its temperature anomaly trend will inevitably be continuously rising. This eliminates false anomalies caused by instantaneous sensor errors or sudden environmental interference, ensuring that the areas subsequently classified as temperature anomaly regions do indeed have continuously deteriorating thermal defects. This ensures that subsequent fault classification and efficiency evaluation are focused on the real fault points.

[0018] The preferred fault classification index for temperature anomaly regions is determined as follows: For each temperature anomaly region, the gradient radiation characteristic value of the region is determined through gradient direction analysis, and the average slope of the temperature change rate of the region is determined based on the average slope of the tangent line of the fitted curve of the temperature change rate over time. The gradient radiation characteristic value and the average slope are synergistically fused using a nonlinear exponential mapping model to serve as the fault classification index for the temperature anomaly region. The nonlinear exponential mapping model is configured such that: when both the gradient radiation characteristic value and the average slope increase, the fault classification index is amplified using an exponential function; when both the gradient radiation characteristic value and the average slope decrease, the fault classification index is reduced using an exponential function.

[0019] This technical solution constructs a fault classification model based on the synergistic fusion of spatiotemporal features, which can accurately distinguish between two specific types of faults. Heat pipe blockage typically results in localized heat accumulation, with its temperature gradient exhibiting a radial pattern that spreads outwards spatially. Due to its small heat capacity, the temperature rises rapidly. In contrast, heat dissipation medium leakage, due to the diffusion and evaporation of oil flow, results in a more discrete heat distribution, and the temperature rise process is relatively slow. The nonlinear exponential function exhibits signal amplification and suppression characteristics: when both characteristic values ​​increase simultaneously, the exponential function grows explosively, rapidly approaching the first preset value, indicating a heat pipe blockage fault; when both decrease simultaneously, the exponential decay effect causes it to rapidly approach the second preset value, indicating a heat dissipation medium leakage fault.

[0020] Preferably, the fault classification index is fed back into the total variation regularization model, and the regularization coefficient of the total variation regularization model is dynamically adjusted to obtain an optimized total variation regularization model. This includes: constructing an adaptive feedback association mechanism between the regularization coefficient and the fault classification index; the adaptive feedback association mechanism is configured such that: when the fault classification index indicates that the heat pipe is blocked in the temperature abnormal area, the regularization coefficient is reduced to decrease the smoothing strength of the total variation regularization model; when the fault classification index indicates that the heat dissipation medium is leaked in the temperature abnormal area, the regularization coefficient is increased to enhance the smoothing strength of the total variation regularization model, thereby obtaining an optimized total variation regularization model.

[0021] This technical solution constructs a closed-loop control system that transmits information backward from high-level semantic cognition to low-level visual reconstruction. The regularization coefficients in the total variation regularization model determine the trade-off between denoising and smoothing and detail preservation in the image reconstruction process. Traditional fixed coefficients are difficult to handle both clear fault features and fuzzy noise interference at the same time. Through an adaptive feedback association mechanism, a confidence-based parameter adjustment strategy is established, which enables the generated reconstructed image to have better fault feature representation capabilities, thereby improving the accuracy of subsequent fault identification.

[0022] Preferably, the thermal image at the current moment is reconstructed using the optimized total variational regularization model to obtain the reconstructed image. This includes: constructing an energy function containing a data fidelity term and a total variational regularization term; the data fidelity term is configured to constrain the pixel value similarity between thermal images to preserve the original temperature information; the total variational regularization term is configured to constrain the smoothness of the thermal image based on the sparsity of the image gradient to suppress noise and preserve edge features; the regularization coefficient is used as the weight parameter of the total variational regularization term, and the energy function is minimized using an optimization iterative algorithm to obtain the reconstructed image.

[0023] Preferably, determining the heat dissipation efficiency based on the temperature anomaly region in the reconstructed graph includes: dividing the temperature anomaly region into a heat pipe blockage region, a heat dissipation medium leakage region, or a region to be confirmed according to the fault classification index; calculating the area ratio and temperature difference ratio of the heat pipe blockage region and the heat dissipation medium leakage region relative to the total surface of the radiator; generating a blockage influence coefficient and a leakage influence coefficient based on the area ratio and temperature difference ratio; and using the blockage influence coefficient and the leakage influence coefficient to reduce and correct the baseline heat dissipation efficiency of the radiator to be evaluated to obtain the heat dissipation efficiency at the current moment.

[0024] This technical solution establishes a mathematical mapping relationship from physical damage to performance degradation. The overall efficiency of the radiator depends on its effective heat dissipation area and the internal and external temperature difference. The existence of the fault area is essentially a reduction in effective heat dissipation capacity. The area ratio and temperature difference ratio of the blocked area and the leakage area are calculated respectively. These two dimensions represent the breadth and intensity of the fault. The area ratio reflects the spatial range of heat dissipation obstruction, and the temperature difference ratio reflects the severity of local heat exchange failure. By converting these physical quantities into blockage influence coefficient and leakage influence coefficient, and linearly reducing the baseline efficiency, an evaluation model based on damage accumulation is realized to achieve accurate evaluation of heat dissipation faults.

[0025] Preferably, the temperature anomaly area is divided into a heat pipe blockage area, a heat dissipation medium leakage area, or a pending confirmation area according to the fault classification index, including: a first threshold and a second threshold of the preset fault classification index. For any temperature anomaly area, if the fault classification index of the temperature anomaly area is greater than or equal to the first threshold, it is classified as a heat pipe blockage area. If the fault classification index of the temperature anomaly area is greater than or equal to the second threshold and less than the first threshold, it is classified as a pending confirmation area. If the fault classification index of the temperature anomaly area is less than the second threshold, it is classified as a heat dissipation medium leakage area.

[0026] On the other hand, the present invention provides a heat dissipation efficiency evaluation system for a radiator, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any heat dissipation efficiency evaluation method.

[0027] The present invention has the following effects: This invention suppresses non-fault-related temperature rise disturbances caused by operating condition fluctuations, and combines a total variational regularization model and a fault classification index feedback mechanism to achieve adaptive adjustment of model parameters. While preserving key fault texture details, it removes noise to the greatest extent, and synergistically integrates temperature time-series evolution features and gradient spatial radiation features to amplify the fault characteristics of heat pipe blockage and medium leakage, improve the sensitivity of capturing early gradual faults, and enhance the accuracy of transformer heat dissipation status assessment. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the three-dimensional temperature distribution of a thermal image acquired under the condition of blocked heat pipes; Figure 3 It uses the existing total variation regularization model to... Figure 2 A schematic diagram of the 3D effect after image reconstruction; Figure 4 It uses the optimized total variational regularization model of this invention to... Figure 2 A schematic diagram of the 3D effect after image reconstruction; Figure 5 It is a schematic diagram of the three-dimensional temperature distribution of the thermal imaging image collected under the condition of heat dissipation medium leakage; Figure 6 It uses the existing total variation regularization model to... Figure 5 A schematic diagram of the 3D effect after image reconstruction; Figure 7 It uses the optimized total variational regularization model of this invention to... Figure 5 A schematic diagram of the 3D effect after image reconstruction. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0030] This invention provides a method for evaluating the heat dissipation efficiency of a radiator, such as... Figure 1 As shown, it includes: S1: Use a thermal imager to acquire a thermal image of the surface of the transformer's heat sink.

[0031] Many factors affect the heat dissipation efficiency of transformer radiators. External factors typically include dust accumulation on the radiator surface, while internal factors include leakage of the heat dissipation medium, blockage of the heat dissipation pipes, and fan failure. Dust accumulation on the radiator surface is relatively easy to handle, and fan failure is also easy to locate. However, the impact of heat dissipation medium leakage and heat dissipation pipe blockage on the radiator's heat dissipation efficiency is more complex and insidious, especially in the early stages of the fault, where it is often difficult to detect.

[0032] Therefore, in order to more accurately evaluate the heat dissipation effect of transformers, the research focus of this invention is to analyze the impact of heat dissipation medium leakage and heat pipe blockage on heat dissipation efficiency, thereby accurately evaluating the heat dissipation efficiency of the radiator.

[0033] Before evaluating heat dissipation efficiency, it is necessary to first eliminate the influence of external factors on heat dissipation efficiency. To this end, the following two preprocessing steps are proposed: The first step is to remove dust from the radiator surface. Dust accumulation on the radiator surface will hinder the effective dissipation of heat, thus affecting the heat dissipation efficiency. Therefore, before evaluation, it is necessary to ensure that there is no dust accumulation on the radiator surface to eliminate the interference of dust on the heat dissipation effect.

[0034] Secondly, check the fan function: Fan failure can also lead to a decrease in heat dissipation efficiency, which may affect the normal operation of the heat sink. Therefore, it is necessary to check the fan to ensure that it is in normal working condition.

[0035] Once it is confirmed that the heat sink surface is free of dust and the fan is functioning normally, the influence of these easily identifiable factors on the heat dissipation effect can be ruled out. Subsequent analysis can then be conducted, focusing on the impact of internal problems such as heat dissipation medium leakage and heat pipe blockage on heat dissipation efficiency.

[0036] Transformers are crucial for power grid stability, operating under heavy loads and requiring long-term operation. For this type of transformer, heat dissipation efficiency is typically assessed monthly for timely maintenance. However, a monthly assessment cycle cannot capture some minor, gradually accumulating temperature faults that may not cause significant temperature changes in the short term and could even be overlooked. Therefore, to enhance the timely identification of early temperature faults, this invention sets up a weekly assessment of heat dissipation efficiency, and after the assessment, the thermal images collected during that week are cleaned.

[0037] Specifically: Thermal imaging is used to collect thermal images of transformers during operation. These images visually display the temperature distribution on the radiator surface, with each location on the radiator surface corresponding to a temperature value. Thermal imaging helps analyze potential localized temperature rises caused by problems such as cooling medium leakage or blockage of heat pipes. Changes in transformer load directly affect its operating temperature; increased load leads to increased heat generation inside the transformer, consequently raising the radiator temperature.

[0038] During the operation of the transformer, a thermal image of the surface of the radiator to be evaluated is collected every hour. For some heat pipe blockage or cooling medium leakage faults, there may not be drastic temperature changes in the early stage, but rather a gradual decline in heat dissipation and temperature rise. High-frequency monitoring can help capture these small, gradual temperature changes and detect early potential faults in a timely manner.

[0039] Each time the heat dissipation efficiency of the heat sink to be evaluated is assessed, a thermal image of the current moment and all thermal images before the current moment are acquired. All thermal images before the current moment are used as historical thermal images of the current moment. A location on the surface of the heat sink to be evaluated corresponds to the same pixel in all thermal images. The temperature value of each pixel in each thermal image is recorded, as well as the load value of the transformer at the acquisition time of each thermal image.

[0040] Set the thermal image at the current time. The temperature value of each pixel is recorded as The load value of the transformer at the current moment is recorded as .

[0041] S2: Determine the degree of abnormality of the pixel based on the temperature change rate and temperature gradient amplitude of the pixel, as well as the load change rate of the transformer.

[0042] First, considering that the rate of temperature change can reflect whether the thermal state of the area where the pixel is located is stable, it is one of the key bases for detecting faults. During the operation of the heat sink, the temperature change of each pixel is usually relatively stable under normal conditions, and the rate of temperature change is small.

[0043] However, when a malfunction occurs (such as a blocked heat pipe or a leaking heat dissipation medium), the temperature of the corresponding area will change significantly. Therefore, if a pixel has a large rate of temperature change in a thermal image, the pixel may be located in a temperature abnormality area. By using the rate of temperature change, we can make a preliminary judgment on the temperature abnormality of the pixel.

[0044] Therefore, the rate of temperature change for each pixel in the thermal image at the current moment is calculated:

[0045] in, for The thermal image at time 1 Temperature change rate per pixel for The thermal image at time 1 Temperature value of each pixel The current thermal image is the first one. The pixel at the th point Temperature values ​​from historical thermal images.

[0046] Then, considering that the temperature gradient can reflect the spatial variation of the surface temperature distribution of the transformer heat sink, the magnitude of the temperature gradient amplitude of the pixel in the horizontal direction reflects the magnitude of the temperature difference of the pixel in the horizontal direction, and the magnitude of the temperature gradient amplitude of the pixel in the vertical direction reflects the magnitude of the temperature difference of the pixel in the vertical direction. By combining the gradient amplitudes in the horizontal and vertical directions of the pixel, the gradient amplitude of the pixel is obtained, which reflects whether the temperature change around the pixel is drastic.

[0047] Under normal circumstances, the temperature distribution on the surface of the heat sink is relatively uniform, and the temperature gradient amplitude of the pixel is small. However, when a fault occurs, a significant temperature difference will form between the faulty area and the surrounding normal area, thereby increasing the temperature gradient amplitude.

[0048] Therefore, if the temperature gradient magnitude of a pixel in a thermal image is larger, it indicates that the pixel is more likely to be in a temperature anomaly region.

[0049] The horizontal temperature gradient magnitude of each pixel in the current thermal image is equal to the temperature difference between its two horizontally adjacent pixels, and the vertical temperature gradient magnitude is equal to the temperature difference between its two vertically adjacent pixels. The temperature gradient magnitude of the pixel is then synthesized from the horizontal and vertical temperature gradient magnitudes.

[0050] Specifically, the method for synthesizing the temperature gradient magnitude of each pixel is as follows:

[0051] In this formula, The current thermal image is the first one. Temperature gradient magnitude of each pixel and These are the first and second thermal images at the current time. The horizontal and vertical temperature gradient values ​​of each pixel.

[0052] Next, considering that the transformer's load change rate reflects the load changes of the transformer at different times, there is a certain correlation between load changes and temperature changes during normal operation. When the load increases, the transformer generates more heat, and the surface temperature of the radiator will rise accordingly; conversely, when the load decreases, the temperature will decrease. Therefore, considering the load change rate allows for a more comprehensive analysis of the relationship between temperature changes and load.

[0053] The rate of change of the transformer's load at the current moment is:

[0054] In this formula, This represents the rate of change of the transformer's load at the current moment. This represents the transformer's load value at the current moment. This represents the load value of the transformer at the corresponding moment in the first historical thermal image.

[0055] Finally, by combining the temperature change rate, temperature gradient magnitude, and load change rate, the anomaly degree of each pixel in the reconstructed image at the current moment is calculated: a temperature anomaly term for each pixel is constructed based on the temperature change rate and temperature gradient magnitude; a dynamic weight adjustment mechanism for the temperature anomaly term and the load change rate is constructed to comprehensively determine the anomaly degree of each pixel; the dynamic weight adjustment mechanism is configured to: establish a negative correlation response relationship between the anomaly degree and the load change rate, and reduce the weight of the temperature anomaly term when the load change rate increases, so as to suppress non-faulty temperature rise interference caused by load fluctuations.

[0056] Specifically, the following relationship is satisfied:

[0057] In this formula, The current thermal image is the first one. The degree of abnormality of each pixel As a regulating factor, The current thermal image is the first one. Temperature change rate per pixel The current thermal image is the first one. Temperature gradient magnitude of each pixel Let || be the rate of change of the transformer's load at the current moment, and || be the absolute value sign.

[0058] In this formula, The smaller the load variation rate of the transformer, the more stable the load. It will get closer to 1, at which point It plays a dominant role in the degree of anomaly of pixels, because Reflects the first The temperature change of each pixel over time The larger it is, the more likely it is to be the first The more likely the temperature value of a single pixel is to be abnormal. The temperature variations of pixels in both the horizontal and vertical directions were taken into account. The larger it is, the more likely it is to be the first The more likely the temperature value of a pixel is to be abnormal, it reflects that when the load changes steadily, the greater the temperature change rate and temperature gradient amplitude of a pixel, the higher the degree of abnormality.

[0059] When the load changes are unstable, the greater the rate of load change of the transformer, the more unstable the load. It will get closer to 0, at which point The temperature change rate and temperature gradient change rate play a dominant role in determining the degree of anomaly in pixels. If these parameters are small, the pixel is more likely to be abnormal. This is because, under normal circumstances, temperature and temperature gradient increase with increasing load. When temperature and temperature gradient do not change with load, the pixel is more likely to be abnormal. This means that even with significant load changes, if a pixel's temperature change rate and temperature gradient amplitude are too small relative to the load change, the degree of anomaly will also increase.

[0060] Using this method, the degree of anomaly of each pixel in the current thermal image is obtained, as well as the degree of anomaly of that pixel in each historical thermal image.

[0061] With this design, the exception handling logic is satisfied as follows: When the transformer load changes slowly, the larger the temperature change and gradient change of the pixel, the more likely the pixel is to be abnormal; when the transformer load changes drastically, the smaller the temperature change and gradient change of the pixel, the more likely the pixel is to be abnormal. When the transformer load changes little, the smaller the temperature change and gradient change of the pixel, the more normal the pixel is. When the transformer load changes greatly, the larger the temperature change and gradient change of the pixel, the more normal the pixel is.

[0062] In summary, the greater the degree of anomaly of a pixel in the thermal image at the current moment, the more likely that the pixel has experienced a temperature anomaly, and the more likely that the heat sink has experienced a heat dissipation failure at that moment.

[0063] S3: Determine the trend of pixel anomaly based on the degree of pixel anomaly and the rate of change of pixel anomaly in historical thermal images.

[0064] Specifically, it includes: Calculate the rate of change of the degree of anomaly for each pixel in the current thermal image, and calculate the rate of change of the degree of anomaly for that pixel in each historical thermal image. The calculation process of the rate of change of the degree of anomaly is the same as the calculation formula of the load change rate of a transformer at the current moment. Only the parameters need to be changed.

[0065] A time decay weighting mechanism is introduced to calculate the weighted cumulative value by weighting the rate of change of the abnormality of the pixel in each historical thermal image. The trend of the abnormality of the pixel is determined by the difference between the rate of change of the abnormality of the pixel and the weighted cumulative value: if the difference is greater than 0, the trend is upward; if the difference is equal to 0, the trend is stable; if the difference is less than 0, the trend is downward.

[0066] Specifically, the following relationship is satisfied:

[0067] In this formula, The current thermal image is the first one. The trend of changes in the degree of anomaly of each pixel The current thermal image is the first one. The rate of change of the degree of anomaly of each pixel This refers to the serial number of the historical thermal image. This represents the total number of historical thermal images. For the first The time decay weight of historical thermal images is assigned, with historical thermal images more recent to the current time being considered more important. The current thermal image is the first one. The pixel at the th point Rate of change of anomaly severity in historical thermal images.

[0068] In this formula, Indicates the first The weighted summation of the anomaly degree of each pixel in all historical thermal images yields an index that comprehensively considers the rate of change of historical anomaly degree. This index reflects the cumulative trend of the anomaly degree of each pixel in the current thermal image over the historical time period. The larger the index, the more obvious the positive cumulative effect of the anomaly degree of that pixel over the historical time period, that is, the anomaly degree of that pixel generally shows a continuously increasing trend.

[0069] like This indicates the current time of the thermal image. The rate of change of the anomaly degree of the _th pixel is greater than the rate of change of the anomaly degree in historical thermal images, indicating that in the current thermal image, the _th _th pixel... The trend of anomaly severity in individual pixels is increasing.

[0070] like This indicates the current time of the thermal image. The rate of change of the anomaly degree of the nth pixel is equal to the rate of change of the anomaly degree in the historical thermal images, indicating that in the current thermal image, the nth pixel... The trend of anomaly degree changes in individual pixels is stable.

[0071] like This indicates the current time of the thermal image. The rate of change of the anomaly degree of the _th pixel is less than the rate of change of the anomaly degree in historical thermal images, indicating that in the current thermal image, the _th _th pixel... The trend of the abnormality level of each pixel is decreasing.

[0072] The time decay weight is determined based on the following formula:

[0073] In this formula, For the first Time decay weights of historical thermal images It is a natural constant. This is a coefficient used to adjust the decay rate. Based on historical experience, a value of 0.6 was used to preserve the long-term trend effects of older thermal images. This represents the total number of historical thermal images. along with As it increases, it decreases. The closer , The larger the image, the more recent the historical thermal image. The closer to 0, The larger it is, the closer it is to 1. The smaller the image, the older the historical thermal image. The smaller the value, the closer it approaches 0. This exponential decay function highlights the importance of recent historical thermal images. This is the normalization function.

[0074] S4: Determine temperature anomaly points by analyzing the changing trends of pixel anomalies, and then use a neighborhood search method to divide the temperature anomaly region of the thermal image at the current moment.

[0075] First, for each pixel in the thermal image at the current moment, if the trend of the degree of abnormality of the pixel in the previous 5 consecutive historical thermal images is all increasing, the pixel is judged to be a temperature anomaly point, and it is believed that the heat sink is likely to have a temperature anomaly at that pixel.

[0076] Then, using each temperature anomaly point in the current reconstructed graph as the center, the eight-neighbor search algorithm is used to divide the temperature anomaly region: For each temperature anomaly, a search is performed within its eight-neighborhood. If any neighboring pixel in the eight-neighborhood is also a temperature anomaly, the temperature anomalies are merged into the same temperature anomaly region. Then, taking each temperature anomaly in the merged temperature anomaly region as the center, a search is performed within its eight-neighborhood (excluding those already searched). Newly found temperature anomalies are also merged into the same temperature anomaly region, and this process is repeated continuously.

[0077] The search stops when no new temperature anomalies are found within the eight-neighborhood of any temperature anomaly point in the merged temperature anomaly region. This means that the temperature anomaly region can no longer expand outwards, and the currently merged temperature anomaly region is a complete temperature anomaly region. Following this method, all temperature anomaly regions in the current reconstructed graph are obtained.

[0078] S5: Calculate the fault classification index for the temperature anomaly region, feed the fault classification index back into the total variation regularization model, dynamically adjust the regularization coefficient of the total variation regularization model, and obtain the optimized total variation regularization model.

[0079] When a heat pipe becomes blocked inside the radiator, the temperature in the blocked area will rise rapidly, and the temperature change rate of the pixels in that area will increase significantly. However, when the heat dissipation medium leaks inside the radiator, the temperature rises relatively slowly, and the temperature change rate is relatively small.

[0080] Blockage of heat pipes can lead to localized high temperatures. This is because a blockage creates a high-temperature point, from which the temperature gradually decreases towards the surrounding area, resulting in a significant temperature gradient. However, when the heat dissipation medium leaks, its uneven distribution prevents efficient heat conduction and dissipation. In the initial stage of leakage, the amount of heat dissipation medium around the leak point decreases, reducing its heat dissipation capacity. This causes the local temperature to rise slowly, exhibiting an uneven temperature distribution. The temperature gradient around the leak point is relatively small, and the temperature does not concentrate at any single high-temperature point.

[0081] Therefore, we first analyze the rate of temperature change for each temperature anomaly region across multiple reconstructed graphs. The faster the change, the more likely the temperature anomaly region is caused by a blocked heat pipe. Next, we analyze the gradient radiation characteristic value for each temperature anomaly region. The larger the gradient radiation characteristic value, the more likely the temperature anomaly region is caused by a blocked heat pipe.

[0082] Specifically, it includes: S51: For each temperature anomaly region, determine the gradient radiation characteristic value of that temperature anomaly region through gradient direction analysis.

[0083] First, obtain the temperature gradient direction path of the temperature anomaly region, including steps S511 to S514: S511: The highest temperature point in each temperature anomaly zone is taken as the center point; S512: Calculate the temperature difference between the center point and each of its neighboring pixels, and obtain a gradient direction by pointing from the center point to the neighboring pixel with the largest temperature difference; S513: Use the neighboring pixel with the largest temperature difference as the updated center point; S514: Repeat steps S512 to S513 until the updated center exceeds the temperature anomaly area, and stop to obtain the gradient direction path of the temperature anomaly area, which reflects the path of the key temperature gradient direction of the temperature anomaly area. The gradient direction path shows the changing trend of the temperature gradient direction of the temperature anomaly area, and the gradient direction path will pass through multiple pixels.

[0084] Then, based on the gradient direction path of each temperature anomaly region, it is determined whether the temperature gradient direction of that temperature anomaly region exhibits a radial pattern, and the gradient radiation characteristic value of each temperature anomaly region is calculated:

[0085] In this formula, For the first Gradient radiation characteristic values ​​of a temperature anomaly region For the first The total number of pixels traversed along the temperature gradient direction path in each temperature anomaly region. For the first The highest temperature point in the temperature anomaly area The path along the direction of the temperature gradient passes through the first The direction vector composed of pixels For the first The path along the temperature gradient direction of the temperature anomaly region passes through the [number]th [region]. The pixel and the The direction vector composed of pixels To achieve the desired mold length, For the first The variance of the Euclidean distance between all pixels along the temperature gradient path in a temperature anomaly region and the highest temperature point.

[0086] In this formula, Indicates the first The consistency of the temperature gradient direction of all pixels along the temperature gradient path in a temperature anomaly region reflects the closeness of the gradient path direction to the direction originating from the center. Greater consistency indicates a more radial temperature distribution characteristic in the temperature anomaly region, because molecules... The dot product operation represents vectors. The denominator is the product of the magnitudes. The overall value of the fraction is the cosine of the angle between the two vectors. The larger the cosine value, the smaller the angle and the better the consistency. Reflects the first The uniformity of the distance from the center to all pixels along the temperature gradient direction path in a temperature anomaly region. The smaller the value, the better the uniformity, indicating that the distance distribution from all pixels along the gradient path to the highest temperature point is more concentrated, and the stronger the radiation characteristics of the temperature gradient direction in the temperature anomaly region.

[0087] In summary, combining the consistency of the temperature gradient direction of all pixels along the temperature gradient path with the uniformity of the distance from all pixels along the temperature gradient path to the center allows for a more comprehensive and rigorous measurement of the radiation characteristic value of the gradient path in an abnormal temperature region. The larger the radiation characteristic value, the more likely the abnormal temperature region is caused by a blocked heat pipe.

[0088] S52: Determine the average slope of the temperature change rate in the temperature anomaly region based on the mean slope of the tangent line of the fitted curve of the temperature change rate over time.

[0089] The average slope of the fitted curve of the temperature change rate of the temperature anomaly region reflects how fast the temperature change rate of the temperature anomaly region changes across multiple thermal imaging maps.

[0090] To obtain the temperature anomaly region's corresponding pixels in the current thermal image, the average temperature change rate of all corresponding pixels in the current thermal image is used as the temperature change rate of that anomaly region in the current thermal image. Similarly, the temperature change rate of that anomaly region in each historical thermal image is obtained, thus obtaining the total temperature anomaly region's corresponding pixels in the current thermal image. Temperature change rate.

[0091] Using polynomial regression to fit the pair Curve fitting is performed on the temperature change rate, with the polynomial order set to 2, to obtain the fitted curve of the temperature change rate. The horizontal axis of the fitted curve represents different thermal images, and the vertical axis represents the temperature change rate of the temperature anomaly area in different thermal images.

[0092] The slope at each point on the fitted curve is calculated using the slope calculation formula, and the average slope of the fitted curve is then calculated.

[0093] S53: The gradient radiation characteristic value and the average slope are synergistically fused using a nonlinear exponential mapping model as a fault classification index for this temperature anomaly region.

[0094] Specifically, based on the average slope of the fitted curve of the temperature change rate in the temperature anomaly region and the gradient radiation characteristic value of the temperature anomaly region, a fault classification index is calculated for each temperature anomaly region:

[0095] In this formula, For the first Fault classification indicators for temperature anomaly areas It is a natural exponential function. For the first The average slope (absolute value, ensuring non-negative) of the fitted curve of the rate of temperature change for each temperature anomaly region. For the first Gradient radiation characteristic values ​​(absolute values, ensuring non-negativity) of each temperature anomaly region.

[0096] In this formula, the molecule part It reflects and The synergistic growth effect means that when the sum of the two increases, the numerator grows exponentially, and the denominator... Follow The increase is significant due to the increase of [the value]. Follow It increases and approaches 0.

[0097] when and When the average is larger, It will be much greater than At this point, both the denominator and numerator are expressed in terms of... As the main force, The larger it is, the more likely it is to be the first The temperature change rate is large in each temperature anomaly region. The larger it is, the more it means the first The more pronounced the temperature gradient radiation characteristics of a temperature anomaly region, the greater the likelihood that the temperature anomaly region is caused by a blocked heat pipe. The closer the value is to 1.

[0098] when and The smaller the average size, The closer to 1, The closer it is to 1, the more... and The smaller the difference, the closer the numerator is to 0 and the closer the denominator is to 2. The smaller the value, the higher the value. The slower the temperature change in a temperature anomaly region, the better. The smaller the value, the higher the value. The less pronounced the temperature gradient radiation characteristics in an area of ​​temperature anomaly, the more consistent with the characteristics of heat dissipation medium leakage, leading to... The closer the value is to 0, the smaller it is The higher the value, the more consistent it is with the characteristics of heat dissipation medium leakage.

[0099] According to this formula, fault classification indicators for all temperature anomaly areas are obtained, and these fault classification indicators are all within... Within the range.

[0100] S54: Feed the fault classification index back into the total variation regularization model and dynamically adjust the regularization coefficient to obtain the optimized total variation regularization model.

[0101] For each temperature anomaly region in the current thermal imaging image, an adaptive feedback correlation mechanism is constructed between the regularization coefficient of the total variational regularization model and the fault classification index of that temperature anomaly region. When the fault classification index indicates that the temperature anomaly region has heat pipe blockage, the regularization coefficient is reduced to decrease the smoothing strength of the total variational regularization model. When the fault classification index indicates that the temperature anomaly region has heat dissipation medium leakage, the regularization coefficient is increased to enhance the smoothing strength of the total variational regularization model. Based on the total variational regularization model after the change in smoothing strength, the optimized temperature anomaly region is obtained. By optimizing all temperature anomaly regions in the current reconstructed image, the optimized temperature anomaly region at the current moment is obtained.

[0102] First, define the energy function of the total variational regularized model: ; In this formula, Let be the total energy function of the total variational regularized model. The reconstructed graph to be solved. This is the thermal image (observation image) at the current moment. The coordinates are the two-dimensional pixel coordinates of the image. The first term on the right side of the formula is the data fidelity term, used to constrain the reconstructed image. Compared to the original image The first term is the pixel value similarity between the two values, and the second term is the total variation regularization term, which uses the gradient sparsity prior to remove noise and preserve edges. The initial regularization coefficient set for the total variation regularization model is set to 0.1 and used as a weight parameter to balance the data fidelity term and the regularization term.

[0103] Next, the formula for dynamically adjusting the regularization coefficient of the total variation regularization model using the fault classification index of this temperature anomaly region is as follows:

[0104] In this formula, The adjusted regularization coefficients are... The initial regularization coefficients are... To adjust the sensitivity coefficient (set to an empirical value, such as 2), The fault classification index for this temperature anomaly area.

[0105] The adjustment logic of this formula perfectly matches the physical characteristics of different faults: When a fault in a region of abnormal temperature tends to be due to heat pipe blockage: At this time Larger, approaching 1 For negative values, the exponential function The value of is less than 1, making A decrease in the regularization coefficient means a reduction in smoothing intensity, focusing on data fidelity, thereby fully preserving the sharp edges and high-frequency gradient features unique to the blocked area and preventing image blurring.

[0106] When a fault in a region of abnormal temperature tends to indicate leakage of the heat dissipation medium: At this time... Smaller, close to 0 For positive values, the exponential function The value of is greater than 1, making An increase in the regularization coefficient means enhanced smoothing strength, focusing on noise reduction, which can effectively suppress the low-frequency diffuse noise common in the leakage area and improve the signal-to-noise ratio of weak leakage signals.

[0107] S6: Use the optimized total variational regularization model to reconstruct the thermal image at the current moment to obtain the reconstructed image, and determine the heat dissipation efficiency based on the temperature anomaly area in the reconstructed image.

[0108] This step no longer performs image reconstruction and enhancement operations on the entire image as in the traditional approach. Instead, it focuses on each temperature anomaly region in the current thermal image and adjusts the total variation regularization model based on the fault characteristics of that temperature anomaly region. This is equivalent to customizing a total variation regularization model that is adapted to each temperature anomaly region.

[0109] First, using the optimized total variational regularized model obtained for each temperature anomaly region according to its own fault characteristics, and following the standard process of image reconstruction enhancement of the total variational regularized model, targeted image reconstruction enhancement operations are performed on each temperature anomaly region, and finally the reconstructed image corresponding to the thermal image at the current moment is obtained.

[0110] Next, the first threshold is set as The second threshold is In the optimized reconstruction graph at the current moment, if the fault classification index of a certain temperature anomaly region is at... The abnormal temperature area is classified as a heat pipe blockage area. If the fault classification index is within the range... The area with abnormal temperature is designated as a region to be confirmed. If the fault classification index is within the range... This area of ​​abnormal temperature is classified as a heat dissipation medium leakage area.

[0111] Using this method, all areas of blocked heat pipes, all areas of leaking heat dissipation medium, and all areas to be confirmed are obtained. Half of the total area of ​​all areas to be confirmed is contributed to the areas of blocked heat pipes, and the other half is contributed to the areas of leaking heat dissipation medium. Half of the temperature difference of all areas to be confirmed is contributed to the areas of blocked heat pipes, and the other half is contributed to the areas of leaking heat dissipation medium.

[0112] Then, calculate the congestion impact coefficient:

[0113] In this formula, The congestion impact coefficient, and These represent the total area and temperature difference (maximum temperature minus minimum temperature) of all blocked heat pipe areas. The total surface area of ​​the heat sink to be evaluated. This represents the temperature difference in a standard thermal image.

[0114] In this formula, It is to utilize right Normalization was performed, and the overall values ​​reflect the intensity of localized heat buildup caused by heat pipe blockage. The more numerous the values, the more significant the temperature anomaly in the blocked area. This reflects the spatial extent of the blockage caused by the heat pipe blockage. near , The larger the value, the more likely it is that a large area of ​​heat pipes has failed, resulting in a significant decrease in overall heat dissipation efficiency.

[0115] Determine the leakage impact factor:

[0116] In this formula, The leakage impact factor is... and These represent the total area and temperature difference of all leakage areas of the heat dissipation medium. The total surface area of ​​the heat sink to be evaluated. This represents the temperature difference in a standard thermal image.

[0117] In this formula, It is to utilize right After normalization, the overall value reflects the degree of decrease in local heat conduction capacity caused by leakage of the heat dissipation medium. Leakage is usually accompanied by loss of medium and temperature difference. The larger the value, the more serious the leakage of the heat dissipation medium, because under normal circumstances the temperature distribution is uniform and the temperature difference is small. This reflects the spatial diffusion range of a leaking heat dissipation medium. If near , The larger the value, the more extensive the leakage of heat dissipation medium, resulting in a significant decrease in overall heat dissipation efficiency.

[0118] Finally, the heat dissipation efficiency of the radiator to be evaluated was determined:

[0119] in, To evaluate the current heat dissipation efficiency of the radiator, The baseline heat dissipation efficiency of the radiator to be evaluated is the radiator's heat dissipation efficiency without any blockages or leaks. This is the blockage impact coefficient, which indicates the impairment of heat dissipation efficiency caused by the blockage area. It is the leakage impact coefficient, which indicates the damage to heat dissipation efficiency caused by the leakage area.

[0120] If there is no blockage of heat pipes or leakage of heat dissipation medium on the surface of the radiator to be evaluated, that is and All are 0. This indicates the heat dissipation efficiency when there are no abnormalities. If only the heat pipe is blocked, If it is 0, then Will be The reduction will be carried out, and the extent of the reduction depends on... The size. If only the heat dissipation medium leaks, If it is 0, then Will be The reduction will be carried out, and the extent of the reduction depends on... The size. If both exist simultaneously, It will be affected by both factors simultaneously, and the degree of reduction is determined by the combined influence coefficients of both factors. Through this method, an accurate assessment of the heat dissipation efficiency of the radiator to be evaluated can be achieved.

[0121] To further verify the accuracy of the image reconstruction effect and heat dissipation efficiency assessment of this invention under complex fault conditions, this embodiment selected two typical scenarios, "heat pipe blockage" and "heat dissipation medium leakage," for a three-dimensional thermal imaging reconstruction comparison experiment. The experimental results are as follows: Figures 2 to 7 As shown.

[0122] like Figure 2 As shown, due to the physical blockage inside the heat pipe, heat cannot be effectively conducted, resulting in a highly characteristic local high temperature peak at coordinates (X≈100mm, Y≈100mm). However, due to the sensor noise of the thermal imager and environmental interference, there are a lot of burr-like random noise on the surface of the original image. This noise, mixed with the high-frequency blockage peak signal, can easily interfere with the subsequent accurate calculation of the temperature difference ratio.

[0123] like Figure 3 As shown, existing technologies typically use fixed empirical values ​​as regularization coefficients, failing to detect local physical fault characteristics in images. When processing congestion signals with high-frequency gradient characteristics, the fixed smoothing intensity is excessive, leading to a significant peak-clipping effect. (Comparison) Figure 2 and Figure 3 It is evident that the originally sharp high-temperature peak has been significantly flattened, with the peak temperature dropping from approximately 120°C to around 80°C. Furthermore, the width of the peak base has been incorrectly widened. This excessive smoothing has resulted in the loss of crucial temperature difference information.

[0124] like Figure 4 As shown, based on the nonlinear exponential mapping model described in step S5 of this invention, the system identifies the gradient radiation characteristic values ​​of the region. With the average slope of the rate of temperature change The calculated fault classification indices are all relatively large. The value approaching 1 indicates heat pipe blockage. Therefore, based on the adaptive feedback correlation mechanism in step S6, the regularization coefficient is automatically reduced to decrease the smoothing intensity, removing the basis noise while fully preserving the sharp peak characteristics above 120°C, without any... Figure 3The peak-shaving phenomenon observed in the data indicates that this invention can effectively decouple noise from fault characteristics, ensuring the accuracy of temperature difference data in heat dissipation efficiency evaluation.

[0125] like Figure 5 As shown, unlike the blockage condition, the medium leakage is manifested as the uneven diffusion of hot oil on the radiator surface. In three-dimensional space, it presents as a slope with blurred edges and a slow upward slope, which is a low-frequency diffusion characteristic. Since the leakage area involves fluid flow, its original image is often accompanied by a larger amplitude of non-uniform noise, which masks the true diffusion boundary.

[0126] like Figure 6 As shown, the traditional total variation model exposes its inherent step effect defect. The originally smooth and gradual temperature diffusion surface is reconstructed into several discontinuous step-like planes. This non-physical step-like artifact is caused by the total variation term tending to produce piecewise constant solutions when dealing with smooth gradient regions. This can lead to misjudgment of the leakage area area, splitting the continuous leakage area into multiple discrete regions.

[0127] like Figure 7 As shown, the system detected that the gradient radiation characteristic value and average slope of this region are both small, indicating a low fault classification index. Approaching 0 indicates leakage of the heat dissipation medium. Based on the adaptive feedback mechanism, the system automatically increases the regularization coefficient to enhance the smoothness. The reconstructed surface exhibits smooth and continuous characteristics, effectively eliminating noise and suppressing [other issues]. Figure 6 The step effect in the model restores the natural diffusion pattern of the medium leakage. By dynamically adjusting the model parameters, this invention solves the step artifact problem in the low-frequency diffusion region of the total variation model, thereby improving the accuracy of subsequent assessment of heat dissipation medium leakage faults.

[0128] In addition, the present invention provides a heat dissipation efficiency evaluation system for a heat sink, the heat dissipation efficiency evaluation system including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of any heat dissipation efficiency evaluation method.

Claims

1. A method for evaluating the heat dissipation efficiency of a radiator, characterized in that, include: Collect thermal images of the surface of the radiator to be evaluated during the operation of the transformer, and use the time of heat dissipation efficiency evaluation as the current time. Based on the temperature change rate and temperature gradient magnitude of the pixels in the thermal image at the current moment, as well as the load change rate of the transformer at the current moment, the degree of anomaly of the pixels is determined; based on the degree of anomaly of the pixels and the rate of change of the degree of anomaly of the pixels in historical thermal images, the trend of the degree of anomaly of the pixels is determined. By analyzing the changing trends of pixel anomalies, temperature anomalies are identified. Based on these anomalies, a neighborhood search method is used to delineate the temperature anomaly regions in the current thermal image. Fault classification indices for these regions are calculated and fed back into a total variational regularization model. The regularization coefficients of the total variational regularization model are dynamically adjusted to obtain an optimized model. The optimized model is then used to reconstruct the current thermal image, resulting in a reconstructed image. Based on the temperature anomaly regions in the reconstructed image, the heat dissipation efficiency is determined.

2. The method for evaluating the heat dissipation efficiency of a radiator according to claim 1, characterized in that, Based on the temperature change rate and temperature gradient magnitude of the pixels in the thermal image at the current moment, and the load change rate of the transformer at the current moment, the degree of anomaly of the pixels is determined, including: A temperature anomaly term for each pixel is constructed based on its temperature change rate and temperature gradient magnitude. A dynamic weight adjustment mechanism for the temperature anomaly term and the load change rate is constructed to comprehensively determine the degree of anomaly for each pixel. The dynamic weight adjustment mechanism is configured to establish a negative correlation between the degree of anomaly and the load change rate. When the load change rate increases, the weight of the temperature anomaly term is reduced to suppress non-faulty temperature rise interference caused by load fluctuations.

3. The method for evaluating the heat dissipation efficiency of a radiator according to claim 1, characterized in that, Based on the degree of anomaly of each pixel and the rate of change of its anomaly degree in historical thermal images, the trend of pixel anomaly degree is determined: The rate of change of anomaly degree for each pixel in the current thermal image is calculated; all historical thermal images of the current thermal image are obtained, and the rate of change of anomaly degree for that pixel in each historical thermal image is calculated; a time decay weighting mechanism is introduced to perform a weighted cumulative calculation of the rate of change of anomaly degree for that pixel in each historical thermal image, obtaining a weighted cumulative value; the trend of pixel anomaly degree is determined by the difference between the rate of change of anomaly degree for that pixel and the weighted cumulative value: if the difference is greater than 0, the trend is upward; The difference is equal to 0, and the trend is stable. The difference is less than 0, and the trend is downward.

4. The method for evaluating the heat dissipation efficiency of a radiator according to claim 1, characterized in that, The method for identifying temperature anomaly points by analyzing the changing trend of the anomaly level of each pixel is as follows: If the abnormality of a pixel in the current thermal image shows an increasing trend in the number of consecutive historical thermal images prior to the current moment, then that pixel is determined to be a temperature anomaly.

5. The method for evaluating the heat dissipation efficiency of a radiator according to claim 1, characterized in that, The fault classification indicators for areas with abnormal temperatures are determined based on the following method: For each temperature anomaly region, the gradient radiation characteristic value of the region is determined through gradient direction analysis, and the average slope of the temperature change rate of the region is determined based on the mean slope of the tangent line of the fitted curve of the temperature change rate over time. The gradient radiation characteristic value and the average slope are synergistically fused using a nonlinear exponential mapping model to serve as the fault classification index for the temperature anomaly region. The nonlinear exponential mapping model is configured such that: when both the gradient radiation characteristic value and the average slope increase, the fault classification index is amplified using an exponential function; when both the gradient radiation characteristic value and the average slope decrease, the fault classification index is reduced using an exponential function.

6. The method for evaluating the heat dissipation efficiency of a radiator according to claim 1, characterized in that, The fault classification index is fed back into the total variation regularization model, and the regularization coefficients of the total variation regularization model are dynamically adjusted to obtain the optimized total variation regularization model, including: An adaptive feedback correlation mechanism is constructed between the regularization coefficient and the fault classification index. The adaptive feedback correlation mechanism is configured as follows: when the fault classification index indicates that the heat pipe is blocked in the temperature abnormal area, the regularization coefficient is reduced to reduce the smoothing strength of the total variation regularization model; when the fault classification index indicates that the heat dissipation medium is leaked in the temperature abnormal area, the regularization coefficient is increased to enhance the smoothing strength of the total variation regularization model, thus obtaining the optimized total variation regularization model.

7. The method for evaluating the heat dissipation efficiency of a radiator according to claim 1, characterized in that, The thermal image at the current moment is reconstructed using the optimized total variational regularized model, resulting in a reconstructed image, including: An energy function is constructed that includes a data fidelity term and a total variational regularization term. The data fidelity term is configured to constrain the pixel value similarity between thermal images to preserve the original temperature information. The total variational regularization term is configured to constrain the smoothness of the thermal image based on the sparsity of the image gradient to suppress noise and preserve edge features. The regularization coefficient is used as the weight parameter of the total variational regularization term, and the energy function is minimized using an optimization iterative algorithm to obtain the reconstructed image.

8. The method for evaluating the heat dissipation efficiency of a radiator according to claim 1, characterized in that, Determining heat dissipation efficiency based on temperature anomaly regions in the reconstructed graph includes: Based on the fault classification index, the temperature abnormality area is divided into a heat pipe blockage area, a heat dissipation medium leakage area, or an area to be confirmed; the area ratio and temperature difference ratio of the heat pipe blockage area and the heat dissipation medium leakage area relative to the total surface of the heat sink are calculated respectively; based on the area ratio and temperature difference ratio, the blockage influence coefficient and the leakage influence coefficient are generated respectively; the blockage influence coefficient and the leakage influence coefficient are used to reduce and correct the benchmark heat dissipation efficiency of the heat sink to be evaluated, so as to obtain the heat dissipation efficiency at the current moment.

9. The method for evaluating the heat dissipation efficiency of a radiator according to claim 8, characterized in that, Based on the fault classification indicators, the temperature anomaly area is divided into areas of blocked heat pipes, leaking heat dissipation medium, or areas awaiting confirmation, including: The system presets a first threshold and a second threshold for fault classification indicators. For any temperature anomaly area, if the fault classification indicator of the temperature anomaly area is greater than or equal to the first threshold, it is classified as a heat pipe blockage area. If the fault classification indicator of the temperature anomaly area is greater than or equal to the second threshold and less than the first threshold, it is classified as a region to be confirmed. If the fault classification indicator of the temperature anomaly area is less than the second threshold, it is classified as a heat dissipation medium leakage area.

10. A system for evaluating the heat dissipation efficiency of a radiator, characterized in that, The heat dissipation efficiency evaluation system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the heat dissipation efficiency evaluation method as described in any one of claims 1-9.

Images