Early warning method for equipment temperature rise based on infrared temperature field persistent homology analysis

Abstract

The application belongs to the technical field of infrared thermal imaging and equipment state monitoring, and particularly relates to an early warning method for temperature rise of equipment based on persistent homology analysis of infrared temperature field, which comprises the following steps: continuously collecting infrared thermal image sequences of the equipment to be measured, and generating temperature matrices after pretreatment; performing multi-threshold sub-level set decomposition on each temperature matrix to extract connected components under each threshold; performing persistent homology calculation to track the birth and death process of the connected components with the change of the threshold, and generating a persistence diagram; extracting topological time sequence features such as betti zero-dimensional maximum lifetime, betti one-dimensional maximum lifetime and watershain distance from each persistence diagram; and determining the equipment state grade based on the calculation of the topological features and outputting graded warning information. The application realizes high-sensitivity early warning of the temperature rise anomaly of the equipment by monitoring the evolution of the topological structure of the temperature field.

Description

Technical Field

[0001] This invention relates to the fields of infrared thermal imaging technology and equipment condition monitoring technology, and in particular to an early warning method for equipment temperature rise based on persistent coherence analysis of infrared temperature fields. Background Technology

[0002] Infrared thermal imaging technology acquires the temperature field distribution of a target in a non-contact manner and has been widely used in fields such as power equipment condition monitoring and industrial equipment fault diagnosis. Existing infrared thermal imaging temperature rise assessment methods mainly fall into the following categories.

[0003] The first category is static evaluation methods based on temperature thresholds. These methods compare the highest temperature or relative temperature difference with a preset threshold, triggering an alarm when the temperature exceeds the threshold. This method only triggers an alarm after the temperature rises to the preset threshold, making it a reactive alarm rather than a proactive warning; furthermore, fixed thresholds are difficult to adapt to differences in equipment temperature rise under different ambient temperatures and load conditions.

[0004] The second category is dynamic evaluation methods based on the rate of temperature rise. These methods determine the temperature change trend of the equipment by calculating the rate of temperature change between adjacent time points, and issue an early warning when the rate of temperature rise exceeds a set threshold. This method is sensitive to measurement noise; instantaneous fluctuations in temperature data can easily lead to sudden changes in the rate of temperature rise, resulting in false alarms.

[0005] The third category is data-driven trend prediction methods. In recent years, some studies have introduced deep learning models to predict temperature rise trends. However, these methods are black-box models, and the prediction results lack physical interpretability; model training relies on a large amount of labeled data, resulting in high implementation barriers and deployment costs.

[0006] The fourth category comprises methods based on topological data analysis that have emerged in recent years. Existing research has applied persistent cohomology to the identification and classification of photovoltaic infrared hotspot images, achieving static identification of hotspots by extracting the topological features of the images. However, the above applications belong to static analysis or analysis based on a single time section, and do not involve dynamic tracking and early warning of the continuous evolution of the temperature field topology over time.

[0007] In summary, existing infrared thermal imaging temperature rise assessment techniques still have significant shortcomings in terms of detection sensitivity for weak signals in the early stages, noise immunity, and the physical interpretability of assessment results. How to identify abnormal signals in advance from changes in the spatial structure of the temperature field before the temperature value has risen significantly is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0008] This invention proposes an early warning method for equipment temperature rise based on persistent coherence analysis of infrared temperature field. It aims to solve the technical problem that existing infrared thermal imaging temperature rise assessment methods are unable to identify early abnormal signals from changes in the spatial structure of the temperature field before the temperature value increases significantly, so as to achieve early warning.

[0009] This invention provides an early warning method for equipment temperature rise based on persistent coherence analysis of infrared temperature fields, including: Step S1: Continuously acquire infrared thermal image sequences of the device under test, and generate a temperature matrix after performing non-uniformity correction, blind pixel replacement and temperature calibration on each frame of infrared image; Step S2: Perform multi-threshold sub-level set decomposition on the temperature matrix of each frame, extract the upper level set under multiple temperature thresholds, and obtain the connected component set under each temperature threshold. Step S3: Perform persistent cohomology calculation on the temperature matrix of each frame, sort all pixels in the temperature matrix from low to high according to temperature value and activate them in sequence, record the birth and death process of the connected components as the temperature threshold changes, and generate a persistence map of the temperature matrix of each frame. Step S4: Extract topological time series features from the persistence graph of each frame. The topological time series features include the Betty zero-dimensional maximum lifetime sequence, the Betty one-dimensional maximum lifetime sequence, and the Wasserstein distance sequence between the current persistence graph and the reference persistence graph. Step S5: Calculate the comprehensive anomaly index based on the topological time series characteristics, and determine the equipment status level based on the comprehensive anomaly index; Step S6: Output corresponding graded early warning information according to the device status level.

[0010] Further, in step S2, the upper level set is determined according to the following formula: ; in, for Coordinates in the time-temperature matrix Temperature value at that location, Temperature threshold This indicates that the temperature is greater than or equal to the threshold. The set consisting of all the pixels.

[0011] Further, in step S3, the conditions for the birth of a connected component are: when a new cell is activated and all its neighboring cells are not activated, a new connected component is born; the conditions for the death of a connected component are: when two connected components are connected through a newly activated cell, the connected component born later dies; each connected component is represented as a point in the persistence graph, the horizontal coordinate of which is the temperature threshold at the time of birth, and the vertical coordinate is the temperature threshold at the time of death.

[0012] Furthermore, the adjacent pixels are pixels within an eight-neighborhood range centered on the target pixel.

[0013] Further, in step S4, the Betty zero-dimensional maximum lifetime sequence is the maximum lifetime value of all Betty zero-dimensional features in each frame of the persistence graph, and the lifetime is the difference between the extinction temperature threshold and the birth temperature threshold; the Betty one-dimensional maximum lifetime sequence is the maximum lifetime value of all Betty one-dimensional features in each frame of the persistence graph.

[0014] Furthermore, in step S4, the Wasserstein distance between the current persistence map and the reference persistence map is obtained by calculating the optimal matching distance between all topological feature points in the two persistence maps, and the reference persistence map is the persistence map of the temperature matrix of the device under normal operating conditions.

[0015] Furthermore, in step S5, the comprehensive anomaly index ; in, for Betty's zero-dimensional maximum lifespan at any given moment. The first lifetime threshold, for The distance of Wasserstein at any moment The first distance threshold, and These are the weighting coefficients, and .

[0016] Furthermore, the tiered early warning method is as follows: when When the value is less than the first exponential threshold, it is considered a normal state; when... A value greater than or equal to the first exponential threshold and less than the second exponential threshold is considered to be in a state of interest; when A warning state is determined when the value is greater than or equal to the second exponential threshold and less than the third exponential threshold; when An alarm state is determined when the value is greater than or equal to the third exponential threshold.

[0017] Further, in step S1, the non-uniformity correction includes: correcting the response value of each pixel according to the gain correction coefficient and the bias correction coefficient of each pixel; the blind pixel replacement includes: identifying abnormal response pixels based on a pre-established blind pixel table, and replacing and repairing the abnormal response pixels using a neighborhood interpolation method; the temperature calibration includes: converting the corrected grayscale image into the temperature matrix based on the mapping relationship between pixel grayscale values ​​and temperature values ​​established by the blackbody radiation source calibration experiment.

[0018] Furthermore, the resolution of the infrared thermal image sequence is 384 x 288, and the sampling frequency is greater than or equal to 1 Hz; the device under test is a power transformer, a high-voltage motor, a switch cabinet, or an electrical connector.

[0019] The technical effects of the disclosed technical solution are as follows: First, this invention converts each frame of the infrared temperature matrix into a persistent map through persistent coherence analysis, characterizing the spatial morphology of the temperature field from the perspective of its topological structure. Early equipment failures first alter the spatial distribution of the temperature field before changing its numerical value. Therefore, changes in the topological structure occur earlier than a significant increase in the temperature value, enabling this invention to detect identifiable topological anomalies when the temperature change is small, significantly improving the warning time compared to traditional threshold methods.

[0020] Second, this invention achieves continuous tracking of the evolution of the temperature field topology over time by extracting topological time-series features such as the Betty zero-dimensional maximum lifetime, the Betty one-dimensional maximum lifetime, and the Wasserstein distance from the persistence maps of each frame. The Betty zero-dimensional maximum lifetime reflects the persistence of the most significant high-temperature connected component in the temperature field. During normal operation, the temperature field is smoothly connected, resulting in a short Betty zero-dimensional maximum lifetime. When local hotspots appear, isolated high-temperature connected components emerge, leading to a significant increase in the Betty zero-dimensional maximum lifetime. The Wasserstein distance quantifies the overall difference between the current temperature field topology and the normal state topology. These three features characterize the evolution of the temperature field topology from different dimensions and complement each other.

[0021] Third, this invention achieves a comprehensive quantitative assessment of temperature field topological anomalies by weightedly fusing the Betty zero-dimensional maximum lifetime and Wasserstein distance into a comprehensive anomaly index. The Betty zero-dimensional maximum lifetime is sensitive to the occurrence of isolated high-temperature connected components, while the Wasserstein distance is sensitive to the overall changes in the temperature field topology. Combining the two ensures both sensitivity and reliability in detection.

[0022] Fourth, the topological features of this invention, such as the persistence graph, Betti number, and Wasserstein distance, all have strict mathematical definitions and clear physical correspondences, and the evaluation results are interpretable and traceable. Furthermore, the persistence homology calculation only involves sorting the temperature matrix, connectivity analysis, and optimal matching, with controllable computational complexity. It does not rely on large-scale neural network training and can be deployed in real-time in embedded systems. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the early warning method for equipment temperature rise based on persistent coherence analysis of infrared temperature field proposed in this embodiment of the invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] To address the technical problem mentioned in the background art, existing infrared thermal imaging temperature rise assessment methods struggle to identify early abnormal signals from spatial structure changes in the temperature field before a significant temperature increase, thus hindering early warning. This invention provides an early warning method for equipment temperature rise based on persistent coherence analysis of the infrared temperature field, referencing... Figure 1 As shown, the specific steps include: Step S1: Infrared thermal image acquisition and temperature matrix generation.

[0026] (i) Infrared Image Acquisition: Continuously acquire infrared thermal image sequences of the device under test. In this embodiment, an infrared focal plane array detector with a resolution of 384 x 288 is used, and the sampling frequency is set to 1 Hz to continuously acquire infrared thermal image sequences of the transformer high-voltage bushing area. Each frame of the infrared image contains grayscale data of 384 x 288, which equals 110,592 pixels.

[0027] As a better technical solution, the sampling frequency can be adjusted according to the importance of the equipment and the rate of temperature change.

[0028] (II) Non-uniformity correction: Non-uniformity correction is performed on each frame of the acquired infrared image. The purpose of non-uniformity correction is to eliminate fixed pattern noise caused by differences in the response of each pixel in the detector. In this embodiment, a two-point correction method is used. A high-temperature reference blackbody and a low-temperature reference blackbody with known temperatures are set inside the infrared thermal imager. For each pixel, the gain correction coefficient and the offset correction coefficient are calculated based on its actual response values ​​to the high-temperature reference blackbody and the low-temperature reference blackbody. The gain correction coefficient is calculated as follows: the difference between the theoretical response value of the high-temperature reference blackbody and the theoretical response value of the low-temperature reference blackbody, divided by the difference between the actual response value of the pixel to the high-temperature reference blackbody and the actual response value of the pixel to the low-temperature reference blackbody. The offset correction coefficient is calculated as follows: the theoretical response value of the low-temperature reference blackbody minus the gain correction coefficient multiplied by the actual response value of the pixel to the low-temperature reference blackbody. The corrected pixel response value is equal to the gain correction coefficient multiplied by the original response value of the pixel plus the offset correction coefficient.

[0029] (iii) Blind Pixel Replacement: Blind pixel replacement is performed on the corrected infrared image. A blind pixel refers to a pixel in the infrared focal plane array that has an abnormal response. In this embodiment, the blind pixel table is pre-calibrated using a blackbody uniform radiation field test. For each blind pixel location, a weighted average interpolation is performed using the gray values ​​of its eight surrounding effective pixels.

[0030] (iv) Temperature calibration and temperature matrix generation: The corrected and replaced grayscale image is converted into a temperature matrix. Temperature calibration establishes the mapping relationship between pixel grayscale values ​​and temperature values ​​based on blackbody radiation source calibration experiments. In this embodiment, a third-order polynomial is used to fit the grayscale temperature curve, and the temperature value is determined according to the following formula: ; in, For pixel grayscale values, For polynomial coefficients, These represent temperature values. The coefficients of each polynomial were obtained by fitting calibration experimental data of a blackbody radiation source at multiple known temperature points. After transformation, the temperature matrix is ​​obtained. Each element in the temperature matrix express Time coordinates The temperature value of each pixel. The temperature matrix for each frame has a dimension of 384 rows by 288 columns.

[0031] Through step S1 above, the original infrared image is converted into a temperature matrix that can be used for subsequent topology analysis. Non-uniformity correction and blind pixel replacement effectively suppress the influence of detector noise on temperature data. Temperature calibration establishes an accurate mapping relationship between grayscale values ​​and temperature values, providing accurate and reliable temperature matrix data for subsequent persistent coherence analysis.

[0032] Step S2: Multi-threshold sub-level set decomposition.

[0033] For each frame temperature matrix Perform multi-threshold sublevel set decomposition. Sublevel set decomposition is a prerequisite for persistent homology computation. Its purpose is to transform the continuous temperature field into a set of discrete connected components at different temperature thresholds, thereby laying the foundation for topology analysis.

[0034] For each frame temperature matrix At multiple temperature thresholds The upper level set is extracted from each of the following steps. The upper level set is determined according to the following formula: ; in, for Coordinates in the time-temperature matrix Temperature value at that location, Temperature threshold This indicates that the temperature is greater than or equal to the threshold. The set consisting of all the pixels.

[0035] In this embodiment, the temperature threshold The value ranges from the lowest to the highest temperature value in the temperature matrix, with a step size of 0.1 degrees Celsius. As... The temperature gradually increases from the lowest temperature to the highest temperature. The entire image region gradually shrinks into several isolated high-temperature connected components, eventually becoming an empty set.

[0036] upper level set at different temperature thresholds The connectivity will change when At lower temperatures, high-temperature regions interconnect to form a large connected component; when When the elevation reaches a certain level, large connected components break into several isolated connected components; when As the threshold continues to rise, isolated connected components disappear one by one. This evolution of connected components with varying thresholds is the core object that persistent cohomology seeks to track.

[0037] Through step S2, each frame of the two-dimensional continuous temperature field is transformed into a set of discrete connected components that vary with the threshold λ, providing a data foundation for the persistent cohomology calculation in step S3. By using multi-threshold sub-level set decomposition, the continuous temperature field is transformed into a discrete topological object, enabling the spatial structural characteristics of the temperature field to be described and analyzed using rigorous mathematical methods.

[0038] Step S3: Persistent Coherence Calculation and Persistence Graph Generation. For each frame's temperature matrix... Persistent cohomology calculations are performed to track the birth and death of connected components as the temperature threshold changes, generating a persistent graph of the temperature matrix for each frame.

[0039] (a) Construction of the filtering sequence: All pixels in the temperature matrix are sorted from low to high temperature values, and each pixel is activated sequentially. Specifically, starting with the pixel with the lowest temperature, each pixel is marked as activated. As the threshold... As the level increases, the set of activated pixels is exactly equal to the upper level set. This process of gradually activating pixels as the threshold increases from low to high constitutes a filtering sequence.

[0040] (ii) Records of the birth and death of connected components. During the construction of the filter sequence, the following two situations may occur when a new pixel is activated.

[0041] In the first scenario, when all neighboring pixels of a newly activated pixel are not activated, the pixel independently forms a new connected component, which is generated at this temperature threshold. In this embodiment, neighboring pixels are defined as pixels within an eight-neighborhood range centered on the target pixel.

[0042] In the second scenario, when a newly activated cell connects to two or more existing connected components, these connected components merge into a single connected component. According to the rules of persistent cohomology, the later-born connected component dies and merges into the earliest-born connected component. The demise temperature threshold of this later-born connected component is the current temperature threshold.

[0043] (III) Tracking Betty's zero-dimensional features, which correspond to the birth and death of connected components. The steps above record Betty's zero-dimensional features. Each connected component is represented as a point in the persistence graph, denoted as (birth, death), where birth is the temperature threshold at which the connected component is born, and death is the temperature threshold at which the connected component is destroyed. The lifetime L of the connected component is determined by the following formula: L = death-birth; The lifetime L reflects the temperature range in which the connected component survives changes in the temperature threshold. The longer the lifetime, the more significant and stable the connected component (high-temperature region) is.

[0044] (iv) Tracking of Betty's one-dimensional features, which correspond to the birth and death of one-dimensional holes. A one-dimensional hole is a closed ring structure formed by activated pixels, containing inactive pixels inside. In this embodiment, a one-dimensional hole is born when activated pixels form a ring structure; the hole disappears when the interior of the ring is filled by subsequently activated pixels. Each Betty one-dimensional feature is also represented as a point (birth, death) in the persistence graph.

[0045] (v) Generation of the persistence graph: All Betty zero-dimensional feature points and Betty one-dimensional feature points are plotted on a two-dimensional plane with "birth temperature threshold" as the horizontal axis and "death temperature threshold" as the vertical axis, thus obtaining the persistence graph D(t) of the temperature matrix of this frame. The vertical distance of each point in the persistence graph from the diagonal (i.e., death-birth) characterizes the salience of the topological feature.

[0046] Through step S3 above, each frame of the temperature matrix is ​​transformed from a set of temperature values ​​into a set of topological features, i.e., a persistent graph. By performing persistent cohomology calculations, each frame of the temperature matrix is ​​converted into a persistent graph, realizing a mathematical expression of the spatial structural features of the temperature field. Changes in topological features reflect changes in the internal thermal state of the equipment earlier than changes in temperature values, laying the foundation for early warning.

[0047] Step S4: Extraction of topological time series features.

[0048] All persistent graphs generated from step S3 Extract the following topological time series features.

[0049] (a) Betty's zero-dimensional maximum lifetime sequence: For each frame of the persistence graph D(t), extract the lifetime values ​​of all Betty zero-dimensional feature points in the graph, take the maximum value, and determine it according to the following formula: ; in, Let be Betty's zero-dimensional feature set. The physical meaning is: the lifetime of the most significant high-temperature connected component in the current temperature field. During normal operation, the temperature field is smooth, and there are no isolated high-temperature islands. It remains at a low level. When a localized hotspot occurs in the device, an isolated high-temperature connectivity component is created, which persists over a relatively wide temperature threshold range. Significantly increased.

[0050] (ii) Betty's one-dimensional maximum lifetime sequence: For each frame of the persistence graph D(t), extract the lifetime values ​​of all Betty's one-dimensional feature points in the graph, and take the maximum value among them, which is determined according to the following formula: ; in, Let be Betty's one-dimensional feature set. The physical meaning is: the lifetime of the most prominent ring structure in the current temperature field. When the equipment is operating normally, there is no ring structure in the temperature field, and the Betty one-dimensional feature set is empty. =0.

[0051] (iii) The Wasserstein distance sequence between persistence graphs, with the persistence graph of the equipment under normal operating conditions as the reference persistence graph. The baseline persistence map is selected as follows: In the initial stage after the equipment is put into operation, after confirming that the equipment has no abnormal operating conditions, select the temperature matrix with the most representative temperature distribution from the acquired infrared image sequence, and calculate its persistence map as the baseline persistence map.

[0052] For each frame of the persistence graph D(t), calculate D(t) and... The Wasserstein distance between them is determined by the following formula: ; The Wasserstein distance is the sum of the optimal matching costs for all pairs of topological feature points between two persistent graphs. Specifically, for all points in two persistent graphs, an optimal bijective matching is found that minimizes the sum of distances between all matched pairs; this minimum sum of distances is the Wasserstein distance.

[0053] The physical meaning is: the degree of overall difference between the current temperature field topology and the normal temperature field topology. When the equipment is running normally, It remains within a small range of fluctuations. When equipment malfunctions, the temperature field topology changes. Significantly increased.

[0054] Through step S4 above, the topological information of each frame of the persistence graph is extracted into three time-series features. , and By extracting topological time-series features, the continuous evolution of the temperature field's topological structure over time was quantitatively tracked. The three features reflect changes in the temperature field's topological structure from different dimensions and are highly complementary.

[0055] Step S5: Calculation of comprehensive anomaly index and determination of status.

[0056] Based on the topological time series features extracted in step S4, a comprehensive anomaly index is calculated, and the equipment status level is determined according to the comprehensive anomaly index.

[0057] (I) Feature normalization and calculation of comprehensive anomaly index, and Normalization was performed separately. Divide by the first lifetime threshold , Divide by the first distance threshold .

[0058] First lifespan threshold The method for obtaining the value is as follows: collect 24 hours of temperature data under normal equipment operating conditions, and calculate the value within that period. mean and standard deviation ,Pick .

[0059] First distance threshold The method for obtaining the value is as follows: Collect 24 hours of temperature data while the equipment is operating normally, and calculate the value within that period. mean and standard deviation ,Pick .

[0060] The comprehensive anomaly index is calculated using the following formula: ; in, for Betty's zero-dimensional maximum lifespan at any given moment. The first lifetime threshold, for The distance of Wasserstein at any moment The first distance threshold, and These are the weighting coefficients, and In this embodiment, =0.5, =0.5.

[0061] (ii) Consistency of physical dimensions. In the above formula, and The dimensions of all values ​​are in degrees Celsius. and The dimensions of all are degrees Celsius, therefore and All are dimensionless numbers. and The coefficients are dimensionless weighting coefficients, and the comprehensive anomaly index Φ(t) is a dimensionless number. The dimensions on both sides of the equation are consistent.

[0062] (III) Setting the grading threshold and determining the state level, setting the first index threshold. Second exponential threshold and the third exponential threshold ,and In this embodiment, .

[0063] According to the comprehensive anomaly index The relationship between the device status and various thresholds is used to determine the device status level. when When this occurs, it is considered a normal state.

[0064] when When this happens, it is considered to be in a state of being monitored.

[0065] when When this occurs, it is determined to be a warning state.

[0066] when When this occurs, it is determined to be an alarm state.

[0067] By combining the Betty zero-dimensional maximum lifetime and Wasserstein distance weighted into a comprehensive anomaly index, a comprehensive quantitative assessment of temperature field topological anomalies is achieved, taking into account both detection sensitivity and reliability.

[0068] Step S6: Output of graded early warning information.

[0069] Based on the device status level determined in step S5, output the corresponding graded early warning information.

[0070] When the status level is normal, no warning information is output, and monitoring continues.

[0071] When the status level is "Attention," an alert is output, including device identification information, current status level, current comprehensive anomaly index, current maximum lifespan of the Betty zero-dimensional device, current Wasserstein distance, and recommended measures to strengthen monitoring. The alert is sent via pop-up windows on the monitoring system interface and SMS push notifications.

[0072] When the status level is warning, a warning message is output, including equipment identification information, current status level, current comprehensive anomaly index, current maximum lifespan of the Betty zero dimension, current Wasserstein distance, and suggested maintenance measures. The warning message is sent via a pop-up window on the monitoring system interface, SMS push notifications, and audible and visual alarms.

[0073] When the status level is alarm, alarm information is output, including device identification information, current status level, current comprehensive anomaly index, current maximum lifespan of the Betty zero dimension, current Wasserstein distance, and recommended measures for immediate shutdown and inspection. Alarm information is sent via monitoring system interface pop-ups, SMS push notifications, audible and visual alarms, and telephone notifications.

[0074] By outputting differentiated early warning information in a tiered manner, precise reminders and tiered response guidance are provided to operation and maintenance personnel.

[0075] In a specific application scenario, when a partial discharge defect occurs in the high-voltage bushing of a transformer, the temperature field begins to change slowly.

[0076] In the early stages of defect development (minutes 0 to 90), the highest temperature of the transformer's high-voltage bushing rose from 48.0 degrees Celsius to 48.6 degrees Celsius, an increase of only 0.6 degrees Celsius, comparable to normal fluctuations. However, persistence plot analysis showed that around minute 45, β0(t) rose from approximately 0.8 degrees Celsius during normal operation to 1.5 degrees Celsius, indicating the emergence of a prolonged connected component in the temperature field, i.e., the formation of an isolated high-temperature island. At this point... The index rose slowly from approximately 0.5 to 0.8. (Comprehensive Abnormality Index) The value increased from approximately 0.6 to approximately 0.9, but was still less than Φ1=1.0, so the system determined it to be in a normal state.

[0077] Between the 90th and 180th minutes, the temperature rose from 48.6 degrees Celsius to 49.5 degrees Celsius. β0(t) increased from 1.5 degrees Celsius to 3.2 degrees Celsius, and W_dist(t) increased from 0.8 to 1.5. The comprehensive anomaly index was calculated using the following formula: =0.5×(3.2 / 2.0)+0.5×(1.5 / 1.2)=1.425; =1.425 exceeded =1.0 but less than =1.8, the system determines it to be in a state of concern and issues a warning. At this time, the absolute temperature rise of the transformer is only 1.5 degrees Celsius.

[0078] Between the 180th and 360th minutes, the temperature rose from 49.5 degrees Celsius to 52.0 degrees Celsius. The temperature rose from 3.2 degrees Celsius to 6.5 degrees Celsius. The index rose from 1.5 to 3.8. The comprehensive anomaly index is calculated using the following formula: =0.5×(6.5 / 2.0)+0.5×(3.8 / 1.2)=3.208 =3.208 exceeded =1.8 and =2.8, the system determines that it is in an alarm state and issues an alarm message.

[0079] Upon receiving the alarm, maintenance personnel immediately arranged for repairs and discovered localized overheating caused by a partial discharge defect inside the high-voltage bushing. Timely intervention prevented the fault from escalating. In this case, the traditional threshold method would not trigger an alarm until the temperature reached 60 degrees Celsius (approximately 540 minutes). This invention, through persistent coherence analysis of the temperature field topology evolution, detected the abnormal signal approximately 360 minutes earlier.

[0080] When the ambient temperature rises briefly due to sunlight, the temperature field heats up uniformly overall, and the topological structure of the temperature field does not change substantially. The persistence map remains similar to the baseline persistence map. and No significant increases were observed, and the system did not trigger false alarms.

[0081] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. An early warning method for equipment temperature rise based on persistent coherence analysis of infrared temperature field, characterized in that, include: Step S1: Continuously acquire infrared thermal image sequences of the device under test, and generate a temperature matrix after performing non-uniformity correction, blind pixel replacement and temperature calibration on each frame of infrared image; Step S2: Perform multi-threshold sub-level set decomposition on the temperature matrix of each frame, extract the upper level set under multiple temperature thresholds, and obtain the connected component set under each temperature threshold. Step S3: Perform persistent cohomology calculation on the temperature matrix of each frame, sort all pixels in the temperature matrix from low to high according to temperature value and activate them in sequence, record the birth and death process of the connected components as the temperature threshold changes, and generate a persistence map of the temperature matrix of each frame. Step S4: Extract topological time series features from the persistence graph of each frame. The topological time series features include the Betty zero-dimensional maximum lifetime sequence, the Betty one-dimensional maximum lifetime sequence, and the Wasserstein distance sequence between the current persistence graph and the reference persistence graph. Step S5: Calculate the comprehensive anomaly index based on the topological time series characteristics, and determine the equipment status level based on the comprehensive anomaly index; Step S6: Output corresponding graded early warning information according to the device status level.

2. The method according to claim 1, characterized in that, In step S2, the upper level set is determined according to the following formula: ； in, for Coordinates in the time-temperature matrix Temperature value at that location, Temperature threshold This indicates that the temperature is greater than or equal to the threshold. The set consisting of all the pixels.

3. The method according to claim 1, characterized in that, In step S3, the conditions for the birth of a connected component are: when a new cell is activated and all its neighboring cells are not activated, a new connected component is born; the conditions for the death of a connected component are: when two connected components are connected through a newly activated cell, the connected component born later is destroyed; each connected component is represented as a point in the persistence graph, with the horizontal coordinate of the point being the temperature threshold at the time of birth and the vertical coordinate being the temperature threshold at the time of death.

4. The method according to claim 3, characterized in that, The adjacent pixels are the pixels within an eight-neighbor area centered on the target pixel.

5. The method according to claim 1, characterized in that, In step S4, the Betty zero-dimensional maximum lifetime sequence is the maximum lifetime value of all Betty zero-dimensional features in each frame of the persistence graph, and the lifetime is the difference between the extinction temperature threshold and the birth temperature threshold; the Betty one-dimensional maximum lifetime sequence is the maximum lifetime value of all Betty one-dimensional features in each frame of the persistence graph.

6. The method according to claim 1, characterized in that, In step S4, the Wasserstein distance between the current persistence map and the reference persistence map is obtained by calculating the optimal matching distance between all topological feature points in the two persistence maps. The reference persistence map is the persistence map of the temperature matrix of the device under normal operating conditions.

7. The method according to claim 1, characterized in that, In step S5, the comprehensive anomaly index ; in, for Betty's zero-dimensional maximum lifespan at any given moment. The first lifetime threshold, for The distance of Wasserstein at any moment The first distance threshold, and These are the weighting coefficients, and .

8. The method according to claim 7, characterized in that, The tiered early warning method is as follows: when When the value is less than the first exponential threshold, it is considered a normal state; when... A value greater than or equal to the first exponential threshold and less than the second exponential threshold is considered to be in a state of interest; when A warning state is determined when the value is greater than or equal to the second exponential threshold and less than the third exponential threshold; when An alarm state is determined when the value is greater than or equal to the third exponential threshold.

9. The method according to claim 1, characterized in that, In step S1, the non-uniformity correction includes: correcting the response value of each pixel according to the gain correction coefficient and the bias correction coefficient of each pixel; the blind pixel replacement includes: identifying abnormal response pixels based on a pre-established blind pixel table, and replacing and repairing the abnormal response pixels using a neighborhood interpolation method; the temperature calibration includes: converting the corrected grayscale image into the temperature matrix based on the mapping relationship between pixel grayscale values ​​and temperature values ​​established by the blackbody radiation source calibration experiment.

10. The method according to claim 1, characterized in that, The infrared thermal image sequence has a resolution of 384 x 288 and a sampling frequency greater than or equal to 1 Hz; the device under test is a power transformer, a high-voltage motor, a switch cabinet, or an electrical connector.

Images