Infrared fault diagnosis method for power equipment combined with image recognition

Abstract

The present application relates to the technical field of power equipment fault diagnosis, in particular to an infrared fault diagnosis method of power equipment combined with image recognition, comprising: S1, obtaining multi-source synchronous data flow of a target power equipment in a first diagnosis period; S2, generating a structural feature map of the equipment based on the response relationship of the infrared thermal image sequence in the multi-source synchronous data flow to the change of the environmental parameters; S3, constructing a thermal behavior image of the equipment based on the structural feature map and the infrared thermal image sequence, deducing the theoretical benchmark thermal state of the equipment based on the thermal behavior image and the current environmental parameters; S4, locating the thermal abnormal source by comparing the real-time thermal image with the theoretical benchmark thermal state and performing abnormal backtracking. By taking the natural fluctuation of the natural environmental parameters as an excitation signal, the temperature response difference of each spatial position on the surface of the equipment to the excitation signal is analyzed, effectively solving the pain point that the stock equipment is difficult to implement diagnosis due to the lack of design data, and reducing the threshold and cost of technology popularization.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault diagnosis technology, and more specifically to an infrared fault diagnosis method for power equipment that combines image recognition. Background Technology

[0002] During long-term operation, power equipment inevitably experiences various heat-related faults due to factors such as load impact, environmental corrosion, and material aging. Infrared thermal imaging technology, with its advantages of being non-contact, uninterrupted power supply, and capable of large-area rapid scanning, has become an important tool for condition monitoring and fault diagnosis of power equipment.

[0003] Current infrared fault diagnosis methods for power equipment mainly fall into two categories: The first is the direct comparison method based on temperature thresholds. This method identifies faults by determining whether the absolute temperature or relative temperature difference of the equipment surface exceeds the limits specified in regulations. While simple to implement, this method heavily relies on the accurate setting of empirical thresholds and cannot effectively distinguish whether the temperature rise on the equipment surface originates from internal faults or external environmental factors (such as sunlight, wind speed, and ambient temperature). It is prone to misjudgment or missed judgment under complex weather conditions. The second category is pattern recognition methods based on image processing. These methods extract the texture, shape, and statistical features of infrared thermal images and use classifiers to identify fault types. This type of method abstracts the fault diagnosis problem into a simple image classification problem, severing the inherent connection between thermal images and the equipment's physical structure, material properties, and operating conditions. This results in opaque diagnostic logic and extremely high requirements for the completeness of training samples, making it difficult to effectively handle unseen equipment types or novel fault modes. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing an infrared fault diagnosis method for power equipment that combines image recognition.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for infrared fault diagnosis of power equipment combined with image recognition, comprising the following steps:

[0006] S1. Obtain the multi-source synchronous data stream of the target power equipment during the first diagnostic cycle;

[0007] S2. Based on the response relationship of infrared thermal image sequences in multi-source synchronous data streams to changes in environmental parameters, generate a structural feature map of the equipment;

[0008] S3. Based on the structural feature map and infrared thermal image sequence, construct the thermal behavior image of the equipment, and based on the thermal behavior image and current environmental parameters, deduce the theoretical baseline thermal state of the equipment;

[0009] S4. By comparing real-time thermal images with theoretical baseline thermal states and tracing anomalies, the source of generated thermal anomalies is located.

[0010] In a preferred embodiment, in S1, based on the thermal inertia of the target power equipment and the expected environmental disturbance change pattern, the duration of the first diagnostic cycle is set, and multi-source synchronous data streams are continuously and synchronously acquired within the duration of the first diagnostic cycle. Infrared thermal image sequences of the target power equipment are continuously acquired at a first preset frequency, and visible light image sequences of the target power equipment at the same viewing angle are synchronously and continuously acquired at a second preset frequency.

[0011] The system synchronously and continuously acquires first-class and second-class environmental parameters of the target device's environment. The first-class environmental parameters include ambient temperature and solar radiation intensity, while the second-class environmental parameters include wind speed and wind direction. Based on a unified time reference, the infrared thermal image sequence, visible light image sequence, first-class environmental parameters, and second-class environmental parameters are time-aligned, and the infrared thermal image and visible light image are spatially registered. The data after temporal and spatial alignment and registration are encapsulated and output as a multi-source synchronous data stream.

[0012] In a preferred embodiment, in step S2, the numerical sequences of the first type of environmental parameters and the numerical sequences of the second type of environmental parameters continuously recorded within the first diagnostic cycle are read from the multi-source synchronous data stream. The standard deviation of the numerical fluctuation of the first type of environmental parameters in the stable period without environmental disturbance is calculated. Three times the standard deviation of the numerical fluctuation is used as the first preset threshold. Based on the thermal diffusivity and geometric dimensions of the main components of the target power equipment, its thermal time constant is estimated. The reciprocal of the thermal time constant is used as the second preset threshold. The first diagnostic cycle is traversed to identify continuous time intervals where the numerical change amplitude of the first type of environmental parameter or the second type of environmental parameter exceeds the first preset threshold and the rate of change per unit time exceeds the second preset threshold. The time interval between the start and end times of the continuous time interval is determined as the period of significant change in environmental disturbance.

[0013] Infrared thermal image sequences are read from multi-source synchronous data streams, and the spatial alignment parameters between the established infrared thermal images and visible light images in the multi-source synchronous data streams are read synchronously. Using the first frame of the infrared thermal image sequence as the reference coordinate system, rigid or non-rigid image registration is performed on each frame of the infrared thermal image sequence based on the spatial alignment parameters to eliminate pixel offset caused by small displacements of the device or camera. From the infrared thermal image sequence after image registration, all thermal image frames whose acquisition time coordinates fall completely between the start and end times of the period of significant environmental disturbance are extracted, and the order of each thermal image frame on the original time axis is strictly maintained, and they are arranged in sequence to form a subset of the dynamic response of the thermal image.

[0014] Using the first frame of the visible light image sequence or the first frame of the infrared thermal image sequence in the multi-source synchronous data stream as the reference image, a rectangular grid partitioning scheme with a fixed number of rows and columns is established on the reference image. Each grid cell in the rectangular grid is defined as a pixel region, and each pixel region is assigned a unique spatial index identifier. The grid partitioning scheme maintains a constant spatial position and resolution during periods of significant environmental disturbance. For each pixel region, based on the spatial index identifier, the temperature values ​​of all pixels within the grid cell coverage area are extracted from each frame of thermal image in the thermal image dynamic response subset. The arithmetic mean of the temperature values ​​of all pixels in the grid cell on the frame thermal image is calculated. The arithmetic mean is used as the representative temperature value of the pixel region at each frame. According to the time order of the thermal image frames, the representative temperature values ​​obtained in each frame are arranged sequentially to form the temperature change trajectory of the pixel region during the period of significant environmental disturbance.

[0015] The instantaneous values ​​of the first type of environmental parameter and the second type of environmental parameter, which are precisely corresponding to the time of each frame of the dynamic response subset of thermal imaging during the period of significant change in environmental disturbance, are read from the multi-source synchronous data stream. The instantaneous value sequence of the first type of environmental parameter is arranged into the trajectory of change of the first environmental parameter, and the instantaneous value sequence of the second type of environmental parameter is arranged into the trajectory of change of the second environmental parameter, with time as the horizontal axis and parameter value as the vertical axis.

[0016] For the temperature change trajectory of each pixel region, a correlation feature extraction operation is performed, specifically including the following steps:

[0017] The temperature change trajectory, the first environmental parameter change trajectory, and the second environmental parameter change trajectory were respectively subjected to zero-mean processing by subtracting the mean value of their respective trajectories to eliminate the influence of the DC component.

[0018] The temperature change trajectory after zero-mean processing and the first environmental parameter change trajectory after zero-mean processing are subjected to point-by-point sliding correlation operation on the time axis to obtain the first cross-correlation function sequence. In the first cross-correlation function sequence, the sliding offset corresponding to the function value reaching the maximum is found, and the sliding offset is determined as the first time delay value of the temperature change of the pixel area relative to the change of the first type of environmental parameter.

[0019] The temperature change trajectory after zero-mean processing and the second environmental parameter change trajectory after zero-mean processing are subjected to point-by-point sliding correlation operation on the time axis to obtain the second cross-correlation function sequence. In the second cross-correlation function sequence, the sliding offset corresponding to the function value reaching the maximum is found, and the sliding offset is determined as the second time delay value of the temperature change of the pixel area relative to the change of the second type of environmental parameter.

[0020] The temperature change trajectory is obtained by subtracting the temperature value at the beginning of the period of significant environmental disturbance from the temperature change trajectory. The first environmental parameter change trajectory is obtained by subtracting the parameter value at the beginning of the period of significant environmental disturbance from the first environmental parameter change trajectory. The first environmental parameter change trajectory is obtained by using the value of the first environmental parameter change trajectory as the independent variable and the value of the temperature change trajectory as the dependent variable. A univariate linear regression is performed to fit the curve, and the slope of the fitted line is calculated. The slope is determined as the first proportionality coefficient between the temperature change amplitude of the pixel area and the change amplitude of the first type of environmental parameter.

[0021] The temperature residual trajectory is obtained by subtracting the first environmental contribution component, represented by the multiplication of the first proportional coefficient and the first environmental parameter change trajectory, from the temperature change trajectory. The parameter value of the second environmental parameter change trajectory at the beginning of the period of significant environmental disturbance is subtracted from the second environmental parameter change trajectory to obtain the second environmental parameter change trajectory. A univariate linear regression is performed with the value of the second environmental parameter change trajectory as the independent variable and the value of the temperature residual trajectory as the dependent variable. The slope of the fitted line is calculated and the slope value is determined as the second proportional coefficient between the temperature change amplitude of the pixel region and the change amplitude of the second type of environmental parameter.

[0022] The first time delay value, the second time delay value, the first scaling factor, and the second scaling factor are arranged in a fixed order to form a four-dimensional correlation feature vector of the pixel region.

[0023] The four-dimensional correlation feature vectors of all pixel regions are normalized to map the numerical range of each dimension to the same interval. The Euclidean distance between the normalized feature vectors of any two pixel regions is calculated. The reciprocal or negative correlation transformation value of the Euclidean distance is used as the similarity measure of the correlation features between the two pixel regions. According to the total number of pixel regions covered by the thermal dynamic response subset and the preset feature region division granularity requirements, a third preset threshold is set. Pixel regions with similarity measures greater than or equal to the third preset threshold are judged to have the same correlation feature attributes and are classified into the same initial cluster. Pixel regions that are adjacent in the image space and belong to the same initial cluster are marked and merged to form a first candidate region with continuous spatial boundaries. Isolated pixel regions that are not adjacent in the image space but belong to the same initial cluster and have the same correlation feature attributes as the first candidate region are merged into the nearest first candidate region to form the final first feature region and second feature region.

[0024] The boundary contours, spatial positions, and mutual topological relationships of the first and second feature regions in the coordinate system of the reference image are digitally encoded to generate a marker image with the same pixel size as the reference image. Each pixel in the marker image is assigned a unique identifier for its respective feature region. The marker image is then used as the structural feature map of the device.

[0025] In a preferred embodiment, in step S3, the thermal behavior profile of the device is constructed by fusing the structural feature map and the infrared thermal image sequence, and the theoretical baseline thermal state of the device is deduced based on the thermal behavior profile and the current environmental parameters. The spatial position correspondence between the infrared thermal image sequence and the visible light image sequence is read from the multi-source synchronous data stream and recorded as a list of pixel coordinates of feature points in the two images. At the same time, the spatial position correspondence between the first frame of the visible light image sequence and the structural feature map is read from the auxiliary information of the structural feature map. The correspondence is formed by drawing the first frame of the visible light image sequence as the base map when generating the structural feature map, so that each pixel position of the structural feature map corresponds one-to-one with the pixel position of the first frame of the visible light image sequence.

[0026] Based on the correspondence between feature points of infrared thermal images and visible light images, and the one-to-one correspondence between pixel positions of the first frame of the visible light image and the structural feature map, a position mapping table is established from the pixel position of each frame of infrared thermal image to the pixel position of the structural feature map. The position mapping table records the target row coordinates and target column coordinates of each pixel of the infrared thermal image on the structural feature map. According to the position mapping table, the temperature values ​​of each pixel in each frame of the infrared thermal image sequence are moved to the target pixel position corresponding to the structural feature map. For empty pixels that do not have temperature values ​​after moving, the temperature values ​​of the four neighboring pixels around the pixel are taken and weighted averaged to fill the missing pixels, thereby obtaining a spatially registered infrared thermal image sequence with the same number of pixel rows and pixel columns as the structural feature map.

[0027] Using the acquisition time of each frame of the spatially registered infrared thermal image sequence as the time index, a time sequence list of the sequence is established, and the partition boundary of the structural feature map is kept unchanged at each moment of the time sequence list. Thus, the spatiotemporal alignment of the structural feature map with the spatially registered infrared thermal image sequence is completed.

[0028] Subsequently, for each feature region in the structural feature map, based on the temperature performance of the feature region in the spatially registered infrared thermal image sequence and associated with the correlation features recorded by each pixel region in the feature region, the thermal behavior parameter set of the feature region is calculated. The temperature values ​​of all pixels covered by the feature region in each frame of thermal image are extracted from the spatially registered infrared thermal image sequence. Invalid fill values ​​caused by position movement are removed. The average temperature of the remaining valid pixels in the frame of thermal image is calculated. The average value is used as the regional representative temperature of the feature region in the frame of thermal image. According to the time sequence of each frame of thermal image, the regional representative temperatures are arranged sequentially to form the regional temperature change curve of the feature region in the first diagnostic cycle. At the same time, the curves of the first type of environmental parameter values ​​and the curves of the second type of environmental parameter values ​​corresponding to the time of the first diagnostic cycle are read from the multi-source synchronous data stream.

[0029] From the generated and saved pixel region association feature record table, all pixel region identifiers belonging to the current feature region are selected. The first time delay value of each of these pixel regions is read. If the feature region contains multiple pixel regions, the average of the first time delay values ​​of all pixel regions is calculated, and the average value is used as the basic thermal inertia parameter of the feature region. If the feature region contains only one pixel region, the first time delay value of the pixel region is directly used as the basic thermal inertia parameter. From the pixel region association feature record table, all pixel regions belonging to the current feature region are selected. The first proportional coefficient of each of these pixel regions is read, and its average value is calculated. The average value is used as the first environmental response sensitivity parameter of the feature region. The second proportional coefficient of each of these pixel regions is read, and its average value is calculated. The average value is used as the second environmental response sensitivity parameter of the feature region.

[0030] The complete numerical sequences of the first and second environmental parameters continuously recorded during the first diagnostic cycle are read from the multi-source synchronous data stream. Simultaneously, the equipment operation logs and fault alarm records contained in the multi-source synchronous data stream are read. The thermal time constant of the component with the largest thermal time constant among the main components of the target power equipment is multiplied by three, and the resulting value is set as the fourth preset threshold. This value is used to determine the shortest continuous duration required to achieve steady state. The standard deviations of the fluctuations of the first and second environmental parameters during the stable period without environmental disturbances are calculated respectively. Twice the standard deviations of the fluctuations are set as the fifth preset thresholds for the first and second environmental parameters, respectively. Set a threshold, iterate through each moment of the first diagnostic cycle, find all time periods where the continuous duration exceeds the fourth preset threshold, and during this period, the fluctuation range of the instantaneous value of the first type of environmental parameter relative to the average value of the time period is always less than the fifth preset threshold of the first type of environmental parameter, and at the same time, the fluctuation range of the instantaneous value of the second type of environmental parameter relative to the average value of the time period is always less than the fifth preset threshold of the second type of environmental parameter. List these time periods as candidate steady-state time periods. Remove any time period from the candidate steady-state time periods that overlaps with the fault occurrence time interval recorded in the equipment operation log or fault alarm record. Determine the time period with the longest duration among the remaining candidate steady-state time periods as the steady-state time period.

[0031] For each feature region, the representative temperature of the feature region on all frames of thermal images during the steady-state period is extracted from the spatially registered infrared thermal image sequence. The average value of these representative temperatures is calculated and used as the steady-state operating temperature of the feature region. Instantaneous ambient temperature values ​​collected at the same time as the thermal image frames during the steady-state period are extracted from the multi-source synchronous data stream. The average value of these instantaneous ambient temperature values ​​is calculated and used as the ambient reference temperature for the steady-state period. The steady-state operating temperature of the feature region is subtracted from the ambient reference temperature, and the difference is determined as the steady-state thermal balance level parameter of the feature region.

[0032] The basic thermal inertia parameters, the first environmental response sensitivity parameters, the second environmental response sensitivity parameters, and the steady-state thermal equilibrium level parameters are compiled into a set of data as the thermal behavior parameter set of the characteristic region.

[0033] Using the structural feature map as the basis for spatial division, a unique number is assigned to each feature region in the structural feature map, and the set of thermal behavior parameters of the feature region is associated with the unique number and stored to form a lookup table of feature region numbers and thermal behavior parameter sets. The lookup table and the image data of the structural feature map are saved together to generate the thermal behavior image of the device.

[0034] When using the theoretical baseline thermal state of the thermal behavior profile and the current environmental parameter extrapolation equipment, the theoretical temperature value of each characteristic region at the current moment is first determined. The instantaneous values ​​of the first type of environmental parameter and the instantaneous values ​​of the second type of environmental parameter at the current moment are read from the multi-source synchronous data stream. The environmental baseline temperature corresponding to the steady-state period is read from the reference table of the thermal behavior profile. The environmental baseline temperature is used as the baseline value of the first type of environmental parameter. The average value of the second type of environmental parameter corresponding to the steady-state period is read from the reference table of the thermal behavior profile. The average value is used as the baseline value of the second type of environmental parameter.

[0035] Subtract the baseline value of the first type of environmental parameter from the instantaneous value of the first type of environmental parameter at the current moment to obtain the change of the first type of environmental parameter. Subtract the baseline value of the second type of environmental parameter from the instantaneous value of the second type of environmental parameter at the current moment to obtain the change of the second type of environmental parameter. For each characteristic region, use the unique number of the characteristic region as an index to look up and read the steady-state thermal balance level parameter, the first environmental response sensitivity parameter and the second environmental response sensitivity parameter of the characteristic region from the reference table of thermal behavior profile. Multiply the change of the first type of environmental parameter by the first environmental response sensitivity parameter, and the calculation result is used as the first environmental contribution temperature increment of the characteristic region.

[0036] Multiply the current change in the second type of environmental parameter by the second environmental response sensitivity parameter, and use the result as the second environmental contribution temperature increment of the characteristic region. Add the steady-state thermal equilibrium level parameter of the characteristic region, the first environmental contribution temperature increment, and the second environmental contribution temperature increment, and the sum is determined as the theoretical temperature value of the characteristic region at the current moment.

[0037] Finally, the theoretical temperature values ​​of all feature regions are integrated to generate the theoretical reference thermal state of the device at the current moment. Based on the number of pixel rows and columns of the structural feature map, a blank image is created. Each pixel position in the blank image corresponds one-to-one with the pixel position in the structural feature map. The initial value of all pixels is set to zero. Each pixel position in the blank image is scanned row by row to check the unique number of the feature region to which the pixel belongs in the corresponding position in the structural feature map.

[0038] Based on the unique number, the theoretical temperature value of the feature region at the current moment is obtained from the calculated theoretical temperature values ​​of each feature region, and the theoretical temperature value is directly written to the pixel position. For the background pixel position in the structural feature map that is not included in any feature region, the instantaneous value of the ambient temperature at the current moment is read from the multi-source synchronous data stream, and the instantaneous value of the ambient temperature is written to the pixel position. The acquisition time mark and temperature unit mark are added to the image after all pixel assignments are completed, and the image is determined as the theoretical reference thermal state.

[0039] In a preferred embodiment, in step S4, the real-time infrared thermal image aligned with time and space is subjected to pixel-by-pixel subtraction with the theoretical reference thermal state. The real-time temperature value of each pixel position is subtracted from the theoretical temperature value of the corresponding pixel position, and the resulting difference image is determined as the global thermal deviation field.

[0040] Using the structural feature map as the spatial division benchmark, the distribution of the global thermal deviation field in each feature region is statistically and morphologically analyzed. For each feature region, the average value, maximum value and the percentage of pixels with thermal deviation exceeding the sixth preset threshold are statistically analyzed. Pixels exceeding the sixth preset threshold are marked as connected components, and at least one thermal deviation cluster region is identified. The geometric center coordinates and spatial range of each thermal deviation cluster region are recorded.

[0041] Based on the thermal behavior parameter set of each feature region in the thermal behavior profile, thermal conduction logic verification is performed on the thermal deviation cluster region. Deviation regions that cannot be explained by known thermal conduction relationships are marked. For each thermal deviation cluster region, the basic thermal inertia parameters and environmental response sensitivity parameters of the feature region where the region is located and all its spatially adjacent feature regions are read from the thermal behavior profile. The expected time delay order of each feature region's response to thermal conduction is determined according to the magnitude of the basic thermal inertia parameters. The thermal deviation evolution curves of the cluster region and its adjacent regions within at least one complete environmental disturbance cycle before the real-time acquisition time are extracted from the continuous time series thermal image. The actual time order of the thermal deviation peaks of the two is compared.

[0042] Simultaneously, the steady-state thermal balance level parameters and environmental response sensitivity parameters of the characteristic region where the thermal deviation accumulation area is located are read from the thermal behavior profile. The theoretical environmental contribution temperature of the region at the current moment is calculated based on the change in environmental parameters at the current moment. The theoretical environmental contribution temperature and steady-state thermal balance level parameters are subtracted from the actual thermal deviation peak value of the thermal deviation accumulation area to obtain the residual thermal deviation.

[0043] If the peak value of thermal deviation in the thermal deviation accumulation area appears earlier than that in its adjacent areas, or if the absolute value of the residual thermal deviation exceeds the seventh preset threshold, then the thermal deviation accumulation area is determined to be inexplicable by the heat conduction of the adjacent areas or its own historical thermal behavior, and the complete feature area where the thermal deviation accumulation area is located is marked as a deviation area that cannot be explained by the known heat conduction relationship.

[0044] Within the marked deviation region, at least one pixel-level heat accumulation core is determined by extreme value search and clustering of the spatial distribution of thermal deviation, serving as an endogenous thermal anomaly source. For each marked deviation region, the local maximum value of thermal deviation at each pixel position is calculated, and the pixel with the largest value in the local neighborhood is retained as a candidate core point. Spatial proximity clustering is performed on the candidate core points, and adjacent candidate core points with an Euclidean distance less than the eighth preset threshold are merged into the same heat accumulation core. The arithmetic mean of the pixel coordinates of all merged core points is used as the final positioning coordinates of the core.

[0045] For each heat accumulation core, verify whether the residual thermal deviation at the core location exceeds the seventh preset threshold and the actual thermal deviation at the core location, after deducting the residual conduction deviation contributed by the expected conduction from the adjacent feature regions, exceeds the ninth preset threshold. Heat accumulation cores that simultaneously meet the above conditions are identified as endogenous thermal anomaly sources, and the pixel coordinates, thermal deviation peak value, and feature region identifier of the anomaly source are recorded. Endogenous thermal anomaly sources are output as anomalies.

[0046] The beneficial effects of this invention are as follows: by using the natural fluctuations of environmental parameters as an excitation signal, the temperature response differences of various spatial locations on the equipment surface to the excitation signal are analyzed. It is not necessary to know the equipment model, component name, geometric dimensions, and material parameters in advance, which effectively solves the pain point of existing equipment being difficult to diagnose due to the lack of design data, and lowers the threshold and cost of technology promotion. By separating the contribution of environmental disturbances and normal heat conduction to the temperature distribution on the equipment surface, the early weak fault signals are effectively enhanced. By independently establishing environmental response sensitivity parameters and steady-state thermal balance level parameters for each characteristic area, the theoretical contribution of environmental factors to the temperature of the area can be deducted, so that the residual thermal deviation is directly focused on the net heat generated by the fault itself. Attached Figure Description

[0047] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0049] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0050] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0051] like Figure 1 This embodiment provides a method for infrared fault diagnosis of power equipment combined with image recognition, including the following steps:

[0052] S1. Obtain the multi-source synchronous data stream of the target power equipment during the first diagnostic cycle;

[0053] Furthermore, in S1, based on the thermal inertia of the target power equipment and the expected environmental disturbance change pattern, the duration of the first diagnostic cycle is set, and multi-source synchronous data streams are continuously and synchronously acquired within the duration of the first diagnostic cycle. Infrared thermal image sequences of the target power equipment are continuously acquired at a first preset frequency. The first preset frequency is set based on the ability to capture the main dynamic changes in the surface temperature of the equipment. Visible light image sequences of the target power equipment at the same viewing angle are synchronously and continuously acquired at a second preset frequency.

[0054] The system synchronously and continuously acquires first-class and second-class environmental parameters of the target device's environment. The first-class environmental parameters include ambient temperature and solar radiation intensity, while the second-class environmental parameters include wind speed and wind direction. Based on a unified time reference, the system performs time alignment on the infrared thermal image sequence, the visible light image sequence, the first-class environmental parameters, and the second-class environmental parameters, and performs spatial registration on the infrared thermal image and the visible light image. The data after temporal and spatial alignment and registration is encapsulated and output as a multi-source synchronous data stream.

[0055] When S1 is used, the duration setting of the first diagnostic cycle is to ensure that the first diagnostic cycle can naturally cover at least one or more processes in which environmental parameters (such as wind speed and solar radiation intensity) undergo significant and non-transient changes. The principle of starting synchronous acquisition is to establish a unified time reference origin to ensure that all subsequent heterogeneous data streams have comparable time labels. This is the fundamental premise for realizing the fusion and alignment of multi-source data in the time dimension.

[0056] By sampling at a frequency no lower than the first preset frequency, and at a rate higher than the main time constant of thermal diffusion on the device surface, the spatiotemporal evolution of its temperature field is continuously captured, thereby recording complete dynamic information of the device's thermal state. The second preset frequency provides continuous recording of the device's geometric structure and appearance texture from the same spatial perspective. Its synchronization principle requires that the infrared and visible light acquisition systems be triggered by the same clock signal or have a definite time offset compensation relationship, thereby ensuring that any infrared thermal image frame can be associated with the corresponding visible light image frame in time. The specific values ​​of the first preset frequency and the second preset frequency may be adapted and set according to the thermal inertia and diagnostic accuracy requirements of the target device, ensuring that each data stream can effectively reflect the changes of the physical quantities of interest (temperature, visible image) and provide sufficient data points for the spatiotemporal alignment of the two.

[0057] The first type of environmental parameters (including at least ambient temperature and solar radiation intensity) mainly characterizes the driving potential or heat source intensity in radiation and convection heat transfer, directly affecting the overall thermal balance level of the equipment. The second type of environmental parameters (including at least wind speed and wind direction) mainly characterizes the intensity and directionality of forced convection heat transfer, directly affecting the local heat dissipation efficiency of the equipment surface. The true state of the main environmental driving forces and boundary conditions acting on the equipment surface is recorded at each moment of infrared thermal imaging sequence acquisition. Based on the timestamps carried by each data stream, interpolation or resampling techniques are used to unify the infrared thermal imaging sequence, visible light image sequence, and environmental parameter sequence to the same standard. On the time axis, the problem of time asynchrony caused by slight differences in the start time of acquisition or sampling period is eliminated. Spatial alignment is to establish the mapping relationship of pixels or features between the visible light image coordinate system and the infrared thermal image coordinate system through image registration technology, so that the spatial position information obtained based on visible light analysis can be accurately mapped onto the thermal image. Finally, encapsulation is to integrate all data that have completed spatiotemporal alignment and have clear correspondence (including the aligned thermal image sequence, the aligned visible light image sequence, the time-aligned environmental parameter sequence, and the spatial alignment parameters) into a structured data packet, namely a multi-source synchronous data stream.

[0058] S2. Based on the response relationship of infrared thermal image sequences in multi-source synchronous data streams to changes in environmental parameters, generate a structural feature map of the equipment;

[0059] Furthermore, in S2, the numerical sequences of the first type of environmental parameters and the second type of environmental parameters continuously recorded within the first diagnostic cycle are read from the multi-source synchronous data stream. The standard deviation of the numerical fluctuation of the first type of environmental parameter in the stable period without environmental disturbance is calculated. Three times the standard deviation of the numerical fluctuation is used as the first preset threshold. Based on the thermal diffusivity and geometric dimensions of the main components of the target power equipment, its thermal time constant is estimated. The reciprocal of the thermal time constant is used as the second preset threshold. The first diagnostic cycle is traversed to identify the continuous time interval where the numerical change amplitude of the first type of environmental parameter or the second type of environmental parameter exceeds the first preset threshold and the rate of change per unit time exceeds the second preset threshold. The time interval between the start and end times of the continuous time interval is determined as the period of significant change in environmental disturbance.

[0060] Infrared thermal image sequences are read from multi-source synchronous data streams, and the spatial alignment parameters between the established infrared thermal images and visible light images in the multi-source synchronous data streams are read synchronously. Using the first frame of the infrared thermal image sequence as the reference coordinate system, rigid or non-rigid image registration is performed on each frame of the infrared thermal image sequence based on the spatial alignment parameters to eliminate pixel offset caused by small displacements of the device or camera. From the infrared thermal image sequence after image registration, all thermal image frames whose acquisition time coordinates fall completely between the start and end times of the period of significant environmental disturbance are extracted, and the order of each thermal image frame on the original time axis is strictly maintained, and they are arranged in sequence to form a subset of the dynamic response of the thermal image.

[0061] Using the first frame of the visible light image sequence or the first frame of the infrared thermal image sequence in the multi-source synchronous data stream as the reference image, a rectangular grid partitioning scheme with a fixed number of rows and columns is established on the reference image. Each grid cell in the rectangular grid is defined as a pixel region, and each pixel region is assigned a unique spatial index identifier. The grid partitioning scheme maintains a constant spatial position and resolution during periods of significant environmental disturbance. For each pixel region, based on the spatial index identifier, the temperature values ​​of all pixels within the grid cell coverage area are extracted from each frame of thermal image in the thermal image dynamic response subset. The arithmetic mean of the temperature values ​​of all pixels in the grid cell on the frame thermal image is calculated. The arithmetic mean is used as the representative temperature value of the pixel region at each frame. According to the time order of the thermal image frames, the representative temperature values ​​obtained in each frame are arranged sequentially to form the temperature change trajectory of the pixel region during the period of significant environmental disturbance.

[0062] The instantaneous values ​​of the first type of environmental parameter and the second type of environmental parameter, which are precisely corresponding to the time of each frame of the dynamic response subset of thermal imaging during the period of significant change in environmental disturbance, are read from the multi-source synchronous data stream. The instantaneous value sequence of the first type of environmental parameter is arranged into the trajectory of change of the first environmental parameter, and the instantaneous value sequence of the second type of environmental parameter is arranged into the trajectory of change of the second environmental parameter, with time as the horizontal axis and parameter value as the vertical axis.

[0063] For the temperature change trajectory of each pixel region, a correlation feature extraction operation is performed, specifically including the following steps:

[0064] The temperature change trajectory, the first environmental parameter change trajectory, and the second environmental parameter change trajectory were respectively subjected to zero-mean processing by subtracting the mean value of their respective trajectories to eliminate the influence of the DC component.

[0065] The temperature change trajectory after zero-mean processing and the first environmental parameter change trajectory after zero-mean processing are subjected to point-by-point sliding correlation operation on the time axis to obtain the first cross-correlation function sequence. In the first cross-correlation function sequence, the sliding offset corresponding to the function value reaching the maximum is found, and the sliding offset is determined as the first time delay value of the temperature change of the pixel area relative to the change of the first type of environmental parameter.

[0066] The temperature change trajectory after zero-mean processing and the second environmental parameter change trajectory after zero-mean processing are subjected to point-by-point sliding correlation operation on the time axis to obtain the second cross-correlation function sequence. In the second cross-correlation function sequence, the sliding offset corresponding to the function value reaching the maximum is found, and the sliding offset is determined as the second time delay value of the temperature change of the pixel area relative to the change of the second type of environmental parameter.

[0067] The temperature change trajectory is obtained by subtracting the temperature value at the beginning of the period of significant environmental disturbance from the temperature change trajectory. The first environmental parameter change trajectory is obtained by subtracting the parameter value at the beginning of the period of significant environmental disturbance from the first environmental parameter change trajectory. The first environmental parameter change trajectory is obtained by using the value of the first environmental parameter change trajectory as the independent variable and the value of the temperature change trajectory as the dependent variable. A univariate linear regression is performed to fit the curve, and the slope of the fitted line is calculated. The slope is determined as the first proportionality coefficient between the temperature change amplitude of the pixel area and the change amplitude of the first type of environmental parameter.

[0068] The temperature residual trajectory is obtained by subtracting the first environmental contribution component, represented by the multiplication of the first proportional coefficient and the first environmental parameter change trajectory, from the temperature change trajectory. The parameter value of the second environmental parameter change trajectory at the beginning of the period of significant environmental disturbance is subtracted from the second environmental parameter change trajectory to obtain the second environmental parameter change trajectory. A univariate linear regression is performed with the value of the second environmental parameter change trajectory as the independent variable and the value of the temperature residual trajectory as the dependent variable. The slope of the fitted line is calculated and the slope value is determined as the second proportional coefficient between the temperature change amplitude of the pixel region and the change amplitude of the second type of environmental parameter.

[0069] The first time delay value, the second time delay value, the first scaling factor, and the second scaling factor are arranged in a fixed order to form a four-dimensional correlation feature vector of the pixel region.

[0070] The four-dimensional correlation feature vectors of all pixel regions are normalized to map the numerical range of each dimension to the same interval. The Euclidean distance between the normalized feature vectors of any two pixel regions is calculated. The reciprocal or negative correlation transformation value of the Euclidean distance is used as the similarity measure of the correlation features between the two pixel regions. According to the total number of pixel regions covered by the thermal dynamic response subset and the preset feature region division granularity requirements, a third preset threshold is set. Pixel regions with similarity measures greater than or equal to the third preset threshold are judged to have the same correlation feature attributes and are classified into the same initial cluster. Pixel regions that are adjacent in the image space and belong to the same initial cluster are marked and merged to form a first candidate region with continuous spatial boundaries. Isolated pixel regions that are not adjacent in the image space but belong to the same initial cluster and have the same correlation feature attributes as the first candidate region are merged into the nearest first candidate region to form the final first feature region and second feature region.

[0071] The boundary contours, spatial positions, and mutual topological relationships of the first and second feature regions in the coordinate system of the reference image are digitally encoded to generate a marker image with the same pixel size as the reference image. Each pixel in the marker image is assigned a unique identifier for its respective feature region. The marker image is then used as the structural feature map of the device.

[0072] It should be noted that the specific steps of the point-by-point sliding related calculations are as follows:

[0073] For the generation of the first cross-correlation function sequence, the two input sequences are: the temperature change trajectory of the pixel region after zero-mean processing, and the first environmental parameter change trajectory after zero-mean processing. The temperature change trajectory uses the acquisition time of each frame of thermal image in the thermal image dynamic response subset as the time index, and each time corresponds to a representative temperature value. The first environmental parameter change trajectory uses the acquisition time that is strictly synchronized with each frame of thermal image in the thermal image dynamic response subset as the time index, and each time corresponds to an instantaneous value of a first type of environmental parameter. The two sequences have the same number of time points, denoted as N, and the time interval between adjacent time points is the reciprocal of the first preset frequency of the infrared thermal image sequence. Zero-mean processing is performed, and the arithmetic mean of the temperature change trajectory during the entire period of significant environmental disturbance change, and the arithmetic mean of the first environmental parameter change trajectory during the same period are calculated respectively. The temperature representative value at each time in the temperature change trajectory is subtracted from the temperature arithmetic mean to obtain the zero-mean temperature trajectory. The parameter value at each time in the first environmental parameter change trajectory is subtracted from the first environmental parameter arithmetic mean to obtain the zero-mean first environmental parameter trajectory.

[0074] This involves translating one sequence relative to another on the time axis and calculating the sum of the point-by-point products of the overlapping portions of the two sequences after each translation. First, the range of the sliding offset is defined. The sliding offset k is an integer representing the number of sampling points the zero-mean temperature trajectory moves backward or forward on the time axis relative to the zero-mean first environmental parameter trajectory. A positive value of k indicates a positive translation of the zero-mean temperature trajectory in the positive direction of the time axis, assuming that temperature changes lag behind environmental parameter changes; a negative value of k indicates a negative translation of the zero-mean temperature trajectory in the negative direction of the time axis, assuming that temperature changes lead environmental parameter changes. The range of k is determined based on the maximum expected thermal time constant of the main components of the target power equipment, typically set to an integer range from -(N-1) to (N-1). However, in engineering implementation, to reduce invalid calculations, k can be limited to the preset maximum number of delayed sampling points M, i.e., the range of k is [-M, ...]. [M], where M is less than N and is set according to the empirical value of the equipment's thermal inertia. For each determined sliding offset k, it is necessary to determine the effective overlapping area of ​​the two sequences under the offset. When k≥0, the data points with indices k+1 to N in the zero-mean temperature trajectory overlap one-to-one with the data points with indices 1 to Nk in the zero-mean first environmental parameter trajectory. When k<0, the data points with indices 1 to N+k in the zero-mean temperature trajectory overlap one-to-one with the data points with indices |k|+1 to N in the zero-mean first environmental parameter trajectory. Each set of overlapping data points constitutes a sliding window under the sliding offset. The number of data point pairs in the window is denoted as L(k), where L(k) = N - |k|;

[0075] For each sliding offset k, point-by-point product and accumulation operations are performed. First, each pair of corresponding data points within the sliding window is extracted. Each data pair consists of a temperature value on a zero-mean temperature trajectory and a parameter value on a zero-mean first environmental parameter trajectory. Then, the temperature value and the parameter value are multiplied to obtain the instantaneous product of the data point pair. The instantaneous products of all L(k) data point pairs within the sliding window are accumulated to obtain the accumulated sum at sliding offset k. Finally, the accumulated sum is divided by L(k) for normalization. The accumulated sum or the normalized accumulated sum is the cross-correlation function value corresponding to the sliding offset k. This process is repeated for each k pair. For all integer values ​​from -M to +M, the process of constructing a sliding window, performing point-by-point multiplication, and summing is repeated to obtain a series of numerical sequences with the sliding offset k as the independent variable and the cross-correlation function value as the dependent variable. This numerical sequence is the first cross-correlation function sequence. The horizontal axis of the first cross-correlation function sequence represents the number of sampling points with time delay, and the vertical axis represents the correlation between the two sequences under that delay. The point-by-point multiplication and summing process involves using one sequence as a template and performing sliding matching on another sequence. The matching degree is calculated once for each sliding step, ultimately forming a complete spectrum of correlation as a function of time delay.

[0076] After obtaining the first cross-correlation function sequence, a peak detection operation is performed to determine the maximum value of the sequence and its corresponding sliding offset. First, the maximum correlation value is initialized to negative infinity, and the optimal offset is initialized to 0. Then, the cross-correlation function value corresponding to each sliding offset k in the first cross-correlation function sequence is traversed. The currently traversed cross-correlation function value is compared with the recorded maximum correlation value. If the current cross-correlation function value is greater than the recorded maximum correlation value, the maximum correlation value is updated to the current cross-correlation function value, and the optimal offset is updated to the current k value. If the current cross-correlation function value is equal to the recorded maximum correlation value, then a preset priority rule (such as prioritizing the offset with the smallest delay, prioritizing the offset with a positive delay, etc.) is used to determine whether to update the optimal offset. After the traversal is completed, the obtained... The optimal offset is the number of time offset sampling points corresponding to the strongest correlation between the temperature change trajectory of the pixel area and the change trajectory of the first environmental parameter. The optimal offset k_max is multiplied by the time interval between adjacent frames of the infrared thermal image sequence (i.e., the reciprocal of the first preset frequency). The time interval unit is converted from the number of sampling points to the physical time unit to obtain the physical time delay value. The physical time delay value is the first time delay value of the temperature change of the pixel area relative to the change of the first type of environmental parameter. The first time delay value represents the degree of lag in the response of the device component corresponding to the pixel area to the change of the first type of environmental parameter (such as solar intensity or ambient temperature). The degree of lag is determined by the thermal inertia of the component, the thermal diffusivity of the material, and the geometric dimensions. It is a key characteristic parameter characterizing the inherent physical properties of the component.

[0077] For the generation of the second cross-correlation function sequence and the determination of the second time delay value for the second type of environmental parameters, a computational logic completely symmetrical to the first cross-correlation function sequence is executed. First, the temperature change trajectory of the same pixel region after zero-mean processing is used as the first input sequence, and the change trajectory of the second environmental parameter after zero-mean processing is used as the second input sequence. The change trajectory of the second environmental parameter originates from the instantaneous numerical sequence of wind speed or wind direction parameters synchronously acquired with the thermal imaging dynamic response subset. Then, the same sliding offset value range [-M, Following the same sliding window construction rules, point-by-point multiplication, and summation method, a second cross-correlation function sequence with the sliding offset k as the independent variable is generated. Peak detection is performed in the second cross-correlation function sequence to find the sliding offset corresponding to the maximum cross-correlation function value. Finally, the sliding offset is multiplied by the time interval of the infrared thermal image sequence and converted into physical time units to obtain the second time delay value of the temperature change of the pixel area relative to the change of the second type of environmental parameter. The second time delay value represents the degree of lag in the response of the device component corresponding to the pixel area to changes in forced convection heat transfer conditions such as wind speed or wind direction. It should be noted that all the above calculation processes are based on the same set of basic data, namely the temperature change trajectory of the same pixel area, and independent sliding correlation analysis is performed with the two types of environmental parameter change trajectories with different physical meanings to obtain two mutually orthogonal independent time delay features that reflect the response characteristics of the pixel area to different environmental stimuli.

[0078] When using S2, a period of significant environmental disturbance change refers to a period in which the change of at least one type of environmental parameter exceeds a preset threshold and the rate of change is sufficient to produce an observable response on the surface temperature field of the equipment. If the environmental parameters are in a stable state, the surface temperature distribution of the equipment mainly reflects steady-state thermal equilibrium. At this time, the temperature difference between different structural regions may be due to material differences, operating conditions, or fault heat sources, but the dynamic response characteristics of each region to environmental excitation cannot be separated. Only when the environmental parameters change significantly will each region show distinguishable differences in temperature change trajectory due to differences in essential properties such as thermal inertia and heat transfer coefficient. Step S2, based on the time series data of environmental parameters continuously recorded in the first diagnostic cycle, locates the continuous time interval with the greatest intensity and most drastic change of environmental disturbance through sliding window comparison, rate of change analysis, or pattern recognition. Its output is at least one time period label with a clear start time and end time, which constitutes the time domain constraint condition for subsequent thermal image screening. This ensures that the input data for subsequent analysis is in the information-rich area with the most significant dynamic response of the system, avoiding invalid feature extraction during steady-state periods.

[0079] Based on the output time period label, all thermal image frames whose time coordinates fall within the time period range are selected from the complete infrared thermal image sequence to form a new data subset. The thermal image dynamic response subset has dual attributes: in the time dimension, it records the complete dynamic evolution process of the device surface temperature field from the start of the disturbance to the development of the disturbance and until it tends to a new equilibrium; in the spatial dimension, it completely preserves the continuous change trajectory of the temperature value of each pixel position of the device over time, strictly maintaining the temporal order relationship between thermal image frames, so that each pixel position can form a continuous change curve with time as the independent variable and temperature as the dependent variable. The continuous change curve carries the essential thermal attribute information of the device component corresponding to the spatial point. Areas with high thermal inertia (such as thick metal components) have slow temperature changes and obvious response delays, while areas with low thermal inertia (such as thin-walled shells and insulating media) have rapid temperature changes and strong synchronization with environmental parameter fluctuations.

[0080] Each pixel or pixel neighborhood in the output thermal image dynamic response subset is regarded as an independent thermal response observation unit. The environmental parameter sequence within the corresponding time period (including solar radiation intensity in the first type of environmental parameters and wind speed and direction in the second type of environmental parameters) is regarded as a multidimensional excitation signal applied to the unit. The correlation does not refer to a simple instantaneous numerical correspondence, but to the essential relationship between the temperature change trajectory and the environmental parameter change trajectory at the dynamic characteristic level. Specifically, it includes, but is not limited to: the time delay characteristics between temperature change and environmental parameter change, the proportional characteristics between the temperature change rate and the environmental parameter change amplitude, the deformation characteristics between the temperature response waveform and the environmental excitation waveform, and the weight distribution characteristics of the contribution of each parameter under the combined action of multiple environmental parameters.

[0081] Using the thermal response feature vector of each pixel unit as input, unsupervised clustering is performed in the image space dimension. If two or more pixel units show statistically high similarity in all dimensions of the association, it is presumed that the device entities corresponding to these pixel units have the same or similar material composition, geometric shape and thermal boundary conditions, and should belong to the same feature region. The first feature region and the second feature region formed correspond to the set of physical units on the device with the same thermal response behavior pattern. The final output is a structural feature map, which is essentially a classification map that recolors the pixels of the original infrared image. Regions with the same color in the map represent parts or parts of parts on the device with consistent thermal behavior attributes.

[0082] S3. Based on the structural feature map and infrared thermal image sequence, construct the thermal behavior image of the equipment, and based on the thermal behavior image and current environmental parameters, deduce the theoretical baseline thermal state of the equipment;

[0083] Furthermore, in S3, the thermal behavior profile of the device is constructed by fusing the structural feature map and the infrared thermal image sequence, and the theoretical baseline thermal state of the device is inferred based on the thermal behavior profile and the current environmental parameters. The spatial position correspondence between the infrared thermal image sequence and the visible light image sequence is read from the multi-source synchronous data stream. The spatial position correspondence is obtained by calibrating the same device feature points in the infrared thermal image and the visible light image, and recorded as a list of pixel coordinates of the feature points in the two images. At the same time, the spatial position correspondence between the first frame of the visible light image sequence and the structural feature map is read from the auxiliary information of the structural feature map. The correspondence is formed by drawing the first frame of the visible light image sequence as the base map when generating the structural feature map, so that each pixel position of the structural feature map corresponds one-to-one with the pixel position of the first frame of the visible light image sequence.

[0084] Based on the correspondence between feature points of infrared thermal images and visible light images, and the one-to-one correspondence between pixel positions of the first frame of the visible light image and the structural feature map, a position mapping table is established from the pixel position of each frame of infrared thermal image to the pixel position of the structural feature map. The position mapping table records the target row coordinates and target column coordinates of each pixel of the infrared thermal image on the structural feature map. According to the position mapping table, the temperature values ​​of each pixel in each frame of the infrared thermal image sequence are moved to the target pixel position corresponding to the structural feature map. For empty pixels that do not have temperature values ​​after moving, the temperature values ​​of the four neighboring pixels around the pixel are taken and weighted averaged to fill the missing pixels, thereby obtaining a spatially registered infrared thermal image sequence with the same number of pixel rows and pixel columns as the structural feature map.

[0085] Using the acquisition time of each frame of the spatially registered infrared thermal image sequence as the time index, a time sequence list of the sequence is established, and the partition boundary of the structural feature map is kept unchanged at each moment of the time sequence list. Thus, the spatiotemporal alignment of the structural feature map with the spatially registered infrared thermal image sequence is completed.

[0086] Subsequently, for each feature region in the structural feature map, based on the temperature performance of the feature region in the spatially registered infrared thermal image sequence and associated with the correlation features recorded by each pixel region in the feature region, the thermal behavior parameter set of the feature region is calculated. The temperature values ​​of all pixels covered by the feature region in each frame of thermal image are extracted from the spatially registered infrared thermal image sequence. Invalid fill values ​​caused by position movement are removed. The average temperature of the remaining valid pixels in the frame of thermal image is calculated. The average value is used as the regional representative temperature of the feature region in the frame of thermal image. According to the time sequence of each frame of thermal image, the regional representative temperatures are arranged sequentially to form the regional temperature change curve of the feature region in the first diagnostic cycle. At the same time, the curves of the first type of environmental parameter values ​​and the curves of the second type of environmental parameter values ​​corresponding to the time of the first diagnostic cycle are read from the multi-source synchronous data stream.

[0087] From the generated and saved pixel region association feature record table, all pixel region identifiers belonging to the current feature region are selected. The first time delay value of each of these pixel regions is read. If the feature region contains multiple pixel regions, the average of the first time delay values ​​of all pixel regions is calculated, and the average value is used as the basic thermal inertia parameter of the feature region. If the feature region contains only one pixel region, the first time delay value of the pixel region is directly used as the basic thermal inertia parameter. From the pixel region association feature record table, all pixel regions belonging to the current feature region are selected. The first proportional coefficient of each of these pixel regions is read, and its average value is calculated. The average value is used as the first environmental response sensitivity parameter of the feature region. The second proportional coefficient of each of these pixel regions is read, and its average value is calculated. The average value is used as the second environmental response sensitivity parameter of the feature region.

[0088] The complete numerical sequences of the first and second environmental parameters continuously recorded during the first diagnostic cycle are read from the multi-source synchronous data stream. Simultaneously, the equipment operation logs and fault alarm records contained in the multi-source synchronous data stream are read. The thermal time constant of the component with the largest thermal time constant among the main components of the target power equipment is multiplied by three, and the resulting value is set as the fourth preset threshold. This value is used to determine the shortest continuous duration required to achieve steady state. The standard deviations of the fluctuations of the first and second environmental parameters during the stable period without environmental disturbances are calculated respectively. Twice the standard deviations of the fluctuations are set as the fifth preset thresholds for the first and second environmental parameters, respectively. Set a threshold, iterate through each moment of the first diagnostic cycle, find all time periods where the continuous duration exceeds the fourth preset threshold, and during this period, the fluctuation range of the instantaneous value of the first type of environmental parameter relative to the average value of the time period is always less than the fifth preset threshold of the first type of environmental parameter, and at the same time, the fluctuation range of the instantaneous value of the second type of environmental parameter relative to the average value of the time period is always less than the fifth preset threshold of the second type of environmental parameter. List these time periods as candidate steady-state time periods. Remove any time period from the candidate steady-state time periods that overlaps with the fault occurrence time interval recorded in the equipment operation log or fault alarm record. Determine the time period with the longest duration among the remaining candidate steady-state time periods as the steady-state time period.

[0089] For each feature region, the representative temperature of the feature region on all frames of thermal images during the steady-state period is extracted from the spatially registered infrared thermal image sequence. The average value of these representative temperatures is calculated and used as the steady-state operating temperature of the feature region. Instantaneous ambient temperature values ​​collected at the same time as the thermal image frames during the steady-state period are extracted from the multi-source synchronous data stream. The average value of these instantaneous ambient temperature values ​​is calculated and used as the ambient reference temperature for the steady-state period. The steady-state operating temperature of the feature region is subtracted from the ambient reference temperature, and the difference is determined as the steady-state thermal balance level parameter of the feature region.

[0090] The basic thermal inertia parameters, the first environmental response sensitivity parameters, the second environmental response sensitivity parameters, and the steady-state thermal equilibrium level parameters are compiled into a set of data as the thermal behavior parameter set of the characteristic region.

[0091] Using the structural feature map as the basis for spatial division, a unique number is assigned to each feature region in the structural feature map, and the set of thermal behavior parameters of the feature region is associated with the unique number and stored to form a lookup table of feature region numbers and thermal behavior parameter sets. The lookup table and the image data of the structural feature map are saved together to generate the thermal behavior image of the device.

[0092] When using the theoretical baseline thermal state of the thermal behavior profile and the current environmental parameter extrapolation equipment, the theoretical temperature value of each characteristic region at the current moment is first determined. The instantaneous values ​​of the first type of environmental parameter and the instantaneous values ​​of the second type of environmental parameter at the current moment are read from the multi-source synchronous data stream. The environmental baseline temperature corresponding to the steady-state period is read from the reference table of the thermal behavior profile. The environmental baseline temperature is used as the baseline value of the first type of environmental parameter. The average value of the second type of environmental parameter corresponding to the steady-state period is read from the reference table of the thermal behavior profile. The average value is used as the baseline value of the second type of environmental parameter.

[0093] Subtract the baseline value of the first type of environmental parameter from the instantaneous value of the first type of environmental parameter at the current moment to obtain the change of the first type of environmental parameter. Subtract the baseline value of the second type of environmental parameter from the instantaneous value of the second type of environmental parameter at the current moment to obtain the change of the second type of environmental parameter. For each characteristic region, use the unique number of the characteristic region as an index to look up and read the steady-state thermal balance level parameter, the first environmental response sensitivity parameter and the second environmental response sensitivity parameter of the characteristic region from the reference table of thermal behavior profile. Multiply the change of the first type of environmental parameter by the first environmental response sensitivity parameter, and the calculation result is used as the first environmental contribution temperature increment of the characteristic region.

[0094] Multiply the current change in the second type of environmental parameter by the second environmental response sensitivity parameter, and use the result as the second environmental contribution temperature increment of the characteristic region. Add the steady-state thermal equilibrium level parameter of the characteristic region, the first environmental contribution temperature increment, and the second environmental contribution temperature increment, and the sum is determined as the theoretical temperature value of the characteristic region at the current moment.

[0095] Finally, the theoretical temperature values ​​of all feature regions are integrated to generate the theoretical reference thermal state of the device at the current moment. Based on the number of pixel rows and columns of the structural feature map, a blank image is created. Each pixel position in the blank image corresponds one-to-one with the pixel position in the structural feature map. The initial value of all pixels is set to zero. Each pixel position in the blank image is scanned row by row to check the unique number of the feature region to which the pixel belongs in the corresponding position in the structural feature map.

[0096] Based on the unique number, the theoretical temperature value of the feature region at the current moment is obtained from the calculated theoretical temperature values ​​of each feature region, and the theoretical temperature value is directly written to the pixel position. For the background pixel position in the structural feature map that is not included in any feature region, the instantaneous value of the ambient temperature at the current moment is read from the multi-source synchronous data stream, and the instantaneous value of the ambient temperature is written to the pixel position. The acquisition time mark and temperature unit mark are added to the image after all pixel assignments are completed, and the image is determined as the theoretical reference thermal state.

[0097] S4. By comparing real-time thermal images with theoretical baseline thermal states and tracing anomalies, the source of generated thermal anomalies can be located.

[0098] Furthermore, in S4, the real-time infrared thermal image aligned in time and space is subjected to pixel-by-pixel subtraction with the theoretical reference thermal state. The real-time temperature value of each pixel position is subtracted from the theoretical temperature value of the corresponding pixel position, and the resulting difference image is determined as the global thermal deviation field.

[0099] Using the structural feature map as the spatial division benchmark, the distribution of the global thermal deviation field in each feature region is statistically and morphologically analyzed. For each feature region, the average value, maximum value and percentage of pixels with thermal deviation exceeding the sixth preset threshold are statistically analyzed. Pixels exceeding the sixth preset threshold are marked as connected components to identify at least one thermal deviation cluster region. The geometric center coordinates and spatial range of each cluster region are recorded. The sixth preset threshold is determined based on the upper limit of the statistical fluctuation range of thermal deviation in the feature region during the fault-free historical period.

[0100] Based on the thermal behavior parameter set of each feature region in the thermal behavior profile, thermal conduction logic verification is performed on the thermal deviation cluster region. Deviation regions that cannot be explained by known thermal conduction relationships are marked. For each thermal deviation cluster region, the basic thermal inertia parameters and environmental response sensitivity parameters of the feature region where the region is located and all its spatially adjacent feature regions are read from the thermal behavior profile. The expected time delay order of each feature region's response to thermal conduction is determined according to the magnitude of the basic thermal inertia parameters. The thermal deviation evolution curves of the thermal deviation cluster region and its adjacent regions within at least one complete environmental disturbance cycle before the real-time acquisition time are extracted from the continuous time series thermal image. The actual time order of the thermal deviation peaks of the two is compared.

[0101] Simultaneously, the steady-state thermal balance level parameters and environmental response sensitivity parameters of the characteristic region where the thermal deviation accumulation area is located are read from the thermal behavior profile. The theoretical environmental contribution temperature of the region at the current moment is calculated based on the change in environmental parameters at the current moment. The theoretical environmental contribution temperature and steady-state thermal balance level parameters are subtracted from the actual thermal deviation peak value of the thermal deviation accumulation area to obtain the residual thermal deviation.

[0102] If the peak value of thermal deviation in the thermal deviation accumulation area occurs earlier than that in its adjacent areas, or if the absolute value of the residual thermal deviation exceeds the seventh preset threshold, then the thermal deviation accumulation area is determined to be inexplicable by the heat conduction of the adjacent areas or its own historical thermal behavior. The complete feature area where the thermal deviation accumulation area is located is marked as a deviation area that cannot be explained by the known heat conduction relationship. The seventh preset threshold is determined based on three times the statistical standard deviation of the residual thermal deviation of the feature area during the fault-free historical period.

[0103] Within the marked deviation region, at least one pixel-level heat accumulation core is determined by extreme value search and clustering of the spatial distribution of thermal deviation, serving as an endogenous thermal anomaly source. For each marked deviation region, the local maximum value of thermal deviation at each pixel position is calculated, and the pixel with the largest value in the local neighborhood is retained as a candidate core point. Spatial proximity clustering is performed on the candidate core points, and adjacent candidate core points with an Euclidean distance less than the eighth preset threshold are merged into the same heat accumulation core. The arithmetic mean of the pixel coordinates of all merged core points is used as the final positioning coordinates of the core.

[0104] The eighth preset threshold is pre-calibrated based on the spatial resolution of the infrared thermal imager and the minimum fault feature size of the equipment. For each heat accumulation core, it is verified whether the residual thermal deviation at the core location exceeds the seventh preset threshold and the actual thermal deviation at the core location, after deducting the residual conduction deviation contributed by the expected conduction from the adjacent feature regions, exceeds the ninth preset threshold. Heat accumulation cores that simultaneously meet the above conditions are identified as endogenous thermal anomaly sources, and the pixel coordinates, thermal deviation peak value, and feature region identifier of the anomaly source are recorded. Endogenous thermal anomaly sources are output as anomalies.

[0105] It should be noted that all preset thresholds in this invention can be adapted and set according to the type of target power equipment, operating conditions and fault diagnosis accuracy requirements, and each threshold is a different value. The preset thresholds are obtained by those skilled in the art by collecting multiple sets of sample data and setting a corresponding preset ratio coefficient for each set of sample data; substituting the set preset ratio coefficient and the collected sample data into the formula, any two formulas form a system of two first equations, and the calculated coefficients are filtered and averaged to obtain the value.

[0106] The size of the coefficient is a specific value obtained by quantifying each parameter to facilitate subsequent comparison. The size of the coefficient depends on the amount of sample data and the preset proportional coefficient initially set by those skilled in the art for each set of sample data; as long as it does not affect the proportional relationship between the parameter and the quantified value.

[0107] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0108] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0110] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0112] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0113] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for infrared fault diagnosis of power equipment combining image recognition, characterized in that, Includes the following steps: S1. Obtain the multi-source synchronous data stream of the target power equipment during the first diagnostic cycle; Based on the thermal inertia of the target power equipment and the expected environmental disturbance change pattern, the duration of the first diagnostic cycle is set. During the duration of the first diagnostic cycle, multi-source synchronous data streams are continuously and synchronously acquired. Infrared thermal image sequences of the target power equipment are continuously acquired at a first preset frequency, and visible light image sequences of the target power equipment at the same viewing angle are synchronously and continuously acquired at a second preset frequency. The system synchronously and continuously acquires first-class and second-class environmental parameters of the target device's environment. The first-class environmental parameters include ambient temperature and solar radiation intensity, while the second-class environmental parameters include wind speed and wind direction. Based on a unified time reference, the system performs time alignment on the infrared thermal image sequence, the visible light image sequence, the first-class environmental parameters, and the second-class environmental parameters, and performs spatial registration on the infrared thermal image and the visible light image. The data after temporal and spatial alignment and registration is encapsulated and output as a multi-source synchronous data stream. S2. Based on the response relationship of infrared thermal image sequences in multi-source synchronous data streams to changes in environmental parameters, a structural feature map of the device is generated; the structural feature map specifically includes: a classification map generated by unsupervised clustering based on the thermal response feature vectors of each pixel on the device surface to changes in environmental parameters, and the original infrared image pixels are recolored; wherein, regions with the same color represent components or parts of components on the device with consistent thermal behavior attributes. S3. Based on the structural feature map and infrared thermal image sequence, construct the thermal behavior image of the equipment, and based on the thermal behavior image and current environmental parameters, deduce the theoretical baseline thermal state of the equipment; S4. By comparing real-time thermal images with theoretical baseline thermal states and tracing anomalies, the source of generated thermal anomalies is located.

2. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 1, characterized in that, In step S2, the numerical sequences of the first type of environmental parameters and the second type of environmental parameters continuously recorded within the first diagnostic cycle are read from the multi-source synchronous data stream. The standard deviations of the numerical fluctuations of the first type of environmental parameters and the second type of environmental parameters during the stable period without environmental disturbances are calculated respectively. Three times the respective standard deviations of the numerical fluctuations are used as the first preset thresholds for the first type of environmental parameters and the second type of environmental parameters respectively. Based on the thermal diffusivity and geometric dimensions of the main components of the target power equipment, its thermal time constant is estimated. The reciprocal of the thermal time constant is used as the second preset threshold. The first diagnostic cycle is traversed to identify continuous time intervals under at least one of the following two conditions: First scenario: The magnitude of change of the value of the first type of environmental parameter exceeds the first preset threshold corresponding to the first type of environmental parameter, and the rate of change exceeds the second preset threshold; Second scenario: The magnitude of the change in the value of the second type of environmental parameter exceeds the first preset threshold corresponding to the second type of environmental parameter, and the rate of change exceeds the second preset threshold; The period between the start and end times of a continuous time interval is defined as the period of significant environmental disturbance change.

3. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 2, characterized in that, Infrared thermal image sequences are read from multi-source synchronous data streams, and the spatial alignment parameters between the established infrared thermal images and visible light images in the multi-source synchronous data streams are read synchronously. Using the first frame of the infrared thermal image sequence as the reference coordinate system, rigid or non-rigid image registration is performed on each frame of the infrared thermal image sequence based on the spatial alignment parameters to eliminate pixel offset caused by small displacements of the device or camera. From the infrared thermal image sequence after image registration, all thermal image frames whose acquisition time coordinates fall completely between the start and end times of the period of significant environmental disturbance are extracted, and the order of each thermal image frame on the original time axis is strictly maintained, and they are arranged in sequence to form a subset of the dynamic response of the thermal image. Using the first frame of the visible light image sequence or the first frame of the infrared thermal image sequence in the multi-source synchronous data stream as the reference image, a rectangular grid partitioning scheme with a fixed number of rows and columns is established on the reference image. Each grid cell in the rectangular grid is defined as a pixel region, and each pixel region is assigned a unique spatial index identifier. The grid partitioning scheme maintains a constant spatial position and resolution during periods of significant environmental disturbance. For each pixel region, based on the spatial index identifier, the temperature values ​​of all pixels within the grid cell coverage area are extracted from each frame of thermal image in the thermal image dynamic response subset. The arithmetic mean of the temperature values ​​of all pixels in the grid cell on the frame thermal image is calculated, and the arithmetic mean is used as the representative temperature value of the pixel region at each frame. According to the time order of the thermal image frames, the representative temperature values ​​obtained in each frame are arranged sequentially to form the temperature change trajectory of the pixel region during the period of significant environmental disturbance.

4. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 3, characterized in that, The instantaneous values ​​of the first type of environmental parameter and the second type of environmental parameter, which are precisely corresponding to the time of each frame of the dynamic response subset of thermal imaging during the period of significant change in environmental disturbance, are read from the multi-source synchronous data stream. The instantaneous value sequence of the first type of environmental parameter is arranged into the trajectory of change of the first environmental parameter, and the instantaneous value sequence of the second type of environmental parameter is arranged into the trajectory of change of the second environmental parameter, with time as the horizontal axis and parameter value as the vertical axis. For the temperature change trajectory of each pixel region, a correlation feature extraction operation is performed, specifically including the following steps: The temperature change trajectory, the first environmental parameter change trajectory, and the second environmental parameter change trajectory were respectively subjected to zero-mean processing by subtracting the mean value of their respective trajectories to eliminate the influence of the DC component. The temperature change trajectory after zero-mean processing and the first environmental parameter change trajectory after zero-mean processing are subjected to point-by-point sliding correlation operation on the time axis to obtain the first cross-correlation function sequence. In the first cross-correlation function sequence, the sliding offset corresponding to the function value reaching the maximum is found, and the sliding offset is determined as the first time delay value of the temperature change of the pixel area relative to the change of the first type of environmental parameter. The temperature change trajectory after zero-mean processing and the second environmental parameter change trajectory after zero-mean processing are subjected to point-by-point sliding correlation operation on the time axis to obtain the second cross-correlation function sequence. In the second cross-correlation function sequence, the sliding offset corresponding to the function value reaching the maximum is found, and the sliding offset is determined as the second time delay value of the temperature change of the pixel area relative to the change of the second type of environmental parameter. The temperature change trajectory is obtained by subtracting the temperature value at the beginning of the period of significant environmental disturbance from the temperature change trajectory. The first environmental parameter change trajectory is obtained by subtracting the parameter value at the beginning of the period of significant environmental disturbance from the first environmental parameter change trajectory. The first environmental parameter change trajectory is obtained by using the value of the first environmental parameter change trajectory as the independent variable and the value of the temperature change trajectory as the dependent variable. A univariate linear regression is performed to fit the curve, and the slope of the fitted line is calculated. The slope is determined as the first proportionality coefficient between the temperature change amplitude of the pixel area and the change amplitude of the first type of environmental parameter. The temperature residual trajectory is obtained by subtracting the first environmental contribution component, represented by the multiplication of the first proportional coefficient and the first environmental parameter change trajectory, from the temperature change trajectory. The parameter value of the second environmental parameter change trajectory at the beginning of the period of significant environmental disturbance change is subtracted from the second environmental parameter change trajectory to obtain the second environmental parameter change trajectory. A univariate linear regression is performed with the value of the second environmental parameter change trajectory as the independent variable and the value of the temperature residual trajectory as the dependent variable. The slope of the fitted line is calculated and the slope value is determined as the second proportional coefficient between the temperature change amplitude and the change amplitude of the second type of environmental parameter in the pixel area.

5. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 4, characterized in that, The first time delay value, the second time delay value, the first scaling factor, and the second scaling factor are arranged in a fixed order to form a four-dimensional correlation feature vector of the pixel region. The four-dimensional correlation feature vectors of all pixel regions are normalized to map the numerical range of each dimension to the same interval. The Euclidean distance between the normalized feature vectors of any two pixel regions is calculated. The reciprocal or negative correlation transformation value of the Euclidean distance is used as the similarity measure of the correlation features between the two pixel regions. According to the total number of pixel regions covered by the thermal dynamic response subset and the preset feature region division granularity requirements, a third preset threshold is set. Pixel regions with similarity measures greater than or equal to the third preset threshold are judged to have the same correlation feature attributes and are classified into the same initial cluster. Pixel regions that are adjacent in the image space and belong to the same initial cluster are marked and merged to form the first candidate region with continuous spatial boundaries. Isolated pixel regions that are not adjacent in the image space but belong to the same initial cluster and have the same correlation feature attributes as the first candidate region are merged into the nearest first candidate region to form the final feature region. The boundary contours, spatial positions, and mutual topological relationships of the feature regions in the coordinate system of the reference image are digitally encoded to generate a marker image with the same pixel size as the reference image. Each pixel in the marker image is assigned a unique identifier for its respective feature region. The marked image is identified as a structural feature map of the device.

6. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 5, characterized in that, In step S3, the thermal behavior image of the device is constructed by fusing the structural feature map and the infrared thermal image sequence, and the theoretical baseline thermal state of the device is inferred based on the thermal behavior image and the current environmental parameters. The spatial position correspondence between the infrared thermal image sequence and the visible light image sequence is read from the multi-source synchronous data stream and recorded as a list of pixel coordinates of feature points in the two images. At the same time, the spatial position correspondence between the first frame of the visible light image sequence and the structural feature map is read from the auxiliary information of the structural feature map. The correspondence is formed by drawing the first frame of the visible light image sequence as the base map when generating the structural feature map, so that each pixel position of the structural feature map corresponds one-to-one with the pixel position of the first frame of the visible light image sequence. Based on the correspondence between feature points of infrared thermal images and visible light images, and the one-to-one correspondence between pixel positions of the first frame of the visible light image and the structural feature map, a position mapping table is established from the pixel position of each frame of infrared thermal image to the pixel position of the structural feature map. The position mapping table records the target row coordinates and target column coordinates of each pixel in the infrared thermal image on the structural feature map. According to the position mapping table, the temperature values ​​of each pixel in each frame of the infrared thermal image sequence are moved to the target pixel position corresponding to the structural feature map. For empty pixels that do not have temperature values ​​after moving, the temperature values ​​of the four neighboring pixels around the pixel are taken and weighted averaged to fill the missing pixels, thereby obtaining a spatially registered infrared thermal image sequence with the same number of pixel rows and pixel columns as the structural feature map.

7. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 6, characterized in that, Using the acquisition time of each frame of the spatially registered infrared thermal image sequence as the time index, a time sequence list of the sequence is established, and the partition boundary of the structural feature map is kept unchanged at each moment of the time sequence list. Thus, the spatiotemporal alignment of the structural feature map with the spatially registered infrared thermal image sequence is completed. For each feature region in the structural feature map, based on the temperature performance of the feature region in the spatially registered infrared thermal image sequence, and associated with the correlation features recorded by each pixel region in the feature region, the thermal behavior parameter set of the feature region is calculated. The temperature values ​​of all pixels covered by the feature region in each frame of thermal image are extracted from the spatially registered infrared thermal image sequence. Invalid fill values ​​caused by position movement are removed. The average temperature of the remaining valid pixels in the frame of thermal image is calculated. The average value is used as the regional representative temperature of the feature region in the frame of thermal image. According to the time sequence of each frame of thermal image, the regional representative temperatures are arranged sequentially to form the regional temperature change curve of the feature region in the first diagnostic cycle. At the same time, the curves of the first type of environmental parameter values ​​and the curves of the second type of environmental parameter values ​​corresponding to the time of the first diagnostic cycle are read from the multi-source synchronous data stream. From the generated and saved pixel region association feature record table, all pixel region identifiers belonging to the current feature region are selected. The first time delay value of each of these pixel regions is read. If the feature region contains multiple pixel regions, the average value of the first time delay value of all pixel regions is calculated and used as the basic thermal inertia parameter of the feature region. If the feature region contains only one pixel region, the first time delay value of the pixel region is directly used as the basic thermal inertia parameter. From the pixel region association feature record table, all pixel regions belonging to the current feature region are selected. The first proportional coefficient of each of these pixel regions is read and its average value is calculated and used as the first environmental response sensitivity parameter of the feature region. The second proportional coefficient of each of these pixel regions is read and its average value is calculated and used as the second environmental response sensitivity parameter of the feature region. The complete numerical sequences of the first and second environmental parameters continuously recorded during the first diagnostic cycle are read from the multi-source synchronous data stream. Simultaneously, the equipment operation logs and fault alarm records contained in the multi-source synchronous data stream are read. The thermal time constant of the component with the largest thermal time constant among the main components of the target power equipment is multiplied by three, and the resulting value is set as the fourth preset threshold. This value is used to determine the shortest continuous duration required to achieve steady state. The standard deviations of the fluctuations of the first and second environmental parameters during the stable period without environmental disturbances are calculated respectively. Twice the standard deviations of the fluctuations are set as the fifth preset thresholds for the first and second environmental parameters, respectively. Set a threshold, iterate through each moment of the first diagnostic cycle, find all time periods where the continuous duration exceeds the fourth preset threshold, and during this period, the fluctuation range of the instantaneous value of the first type of environmental parameter relative to the average value of the time period is always less than the fifth preset threshold of the first type of environmental parameter, and at the same time, the fluctuation range of the instantaneous value of the second type of environmental parameter relative to the average value of the time period is always less than the fifth preset threshold of the second type of environmental parameter. List these time periods as candidate steady-state time periods. Remove any time period from the candidate steady-state time periods that overlaps with the fault occurrence time interval recorded in the equipment operation log or fault alarm record. Determine the time period with the longest duration among the remaining candidate steady-state time periods as the steady-state time period. For each feature region, the region representative temperature of the feature region is extracted from the spatially registered infrared thermal image sequence on all frames of thermal images during the steady-state period. The average value of these region representative temperatures is calculated and used as the steady-state operating temperature of the feature region. The instantaneous values ​​of ambient temperature collected at the same time as the thermal image frames during the steady-state period are extracted from the multi-source synchronous data stream. The average value of these instantaneous values ​​of ambient temperature is calculated and used as the ambient reference temperature for the steady-state period. The steady-state operating temperature of the feature region is subtracted from the ambient reference temperature, and the difference is determined as the steady-state thermal balance level parameter of the feature region. The basic thermal inertia parameters, the first environmental response sensitivity parameters, the second environmental response sensitivity parameters, and the steady-state thermal equilibrium level parameters are compiled into a set of data as the thermal behavior parameter set of the characteristic region. Using the structural feature map as the basis for spatial division, a unique number is assigned to each feature region in the structural feature map, and the set of thermal behavior parameters of the feature region is associated with the unique number and stored to form a lookup table of feature region numbers and thermal behavior parameter sets. The lookup table and the image data of the structural feature map are saved together to generate the thermal behavior image of the device. When extrapolating the theoretical baseline thermal state of the device based on thermal behavior images and current environmental parameters, the theoretical temperature value of each feature region at the current moment is first determined. The instantaneous values ​​of the first type of environmental parameter and the instantaneous values ​​of the second type of environmental parameter at the current moment are read from the multi-source synchronous data stream. The environmental baseline temperature corresponding to the steady-state period is read from the lookup table of the thermal behavior images. The environmental baseline temperature is used as the baseline value of the first type of environmental parameter. The average value of the second type of environmental parameter corresponding to the steady-state period is read from the lookup table of the thermal behavior images. The average value is used as the baseline value of the second type of environmental parameter. Subtracting the baseline value of the first type of environmental parameter from the instantaneous value of the first type of environmental parameter at the current moment yields the change in the first type of environmental parameter. Subtracting the baseline value of the second type of environmental parameter from the instantaneous value of the second type of environmental parameter at the current moment yields the change in the second type of environmental parameter. For each characteristic region, using the unique number of the characteristic region as an index, the steady-state thermal equilibrium level parameter, the first environmental response sensitivity parameter, and the second environmental response sensitivity parameter of the characteristic region are searched and read from the lookup table of the thermal behavior image. The change in the first type of environmental parameter is multiplied by the first environmental response sensitivity parameter, and the result is used as the first environmental contribution temperature increment of the characteristic region.

8. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 7, characterized in that, Multiply the current change in the second type of environmental parameter by the second environmental response sensitivity parameter, and use the result as the second environmental contribution temperature increment of the characteristic region. Add the steady-state thermal equilibrium level parameter, the first environmental contribution temperature increment, and the second environmental contribution temperature increment of the characteristic region, and the sum is determined as the theoretical temperature value of the characteristic region at the current moment. Integrate the theoretical temperature values ​​of all feature regions to generate the theoretical baseline thermal state of the device at the current moment. Based on the number of pixel rows and columns of the structural feature map, create a blank image. Each pixel position in the blank image corresponds one-to-one with the pixel position in the structural feature map. Set the initial value of all pixels to zero. Scan each pixel position in the blank image line by line to check the unique feature region number to which the pixel belongs in the corresponding position in the structural feature map. Based on the unique number, the theoretical temperature value of each feature region at the current moment is obtained from the calculated theoretical temperature values ​​of each feature region, and the theoretical temperature value is directly written to the pixel position. For the background pixel position in the structural feature map that is not included in any feature region, the instantaneous value of the ambient temperature at the current moment is read from the multi-source synchronous data stream, and the instantaneous value of the ambient temperature is written to the pixel position. The acquisition time mark and temperature unit mark are added to the image after all pixel assignments are completed, and the image is determined as the theoretical reference thermal state.

9. The infrared fault diagnosis method for power equipment combined with image recognition according to claim 1, characterized in that, In step S4, the real-time infrared thermal image aligned in time and space is subjected to pixel-by-pixel subtraction with the theoretical reference thermal state. The real-time temperature value of each pixel position is subtracted from the theoretical temperature value of the corresponding pixel position, and the resulting difference image is determined as the global thermal deviation field. Using the structural feature map as the spatial division benchmark, the distribution of the global thermal deviation field in each feature region is statistically and morphologically analyzed. For each feature region, the average value, maximum value and the percentage of pixels with thermal deviation exceeding the sixth preset threshold are statistically analyzed. Pixels exceeding the sixth preset threshold are marked as connected components, and at least one thermal deviation cluster region is identified. The geometric center coordinates and spatial range of each thermal deviation cluster region are recorded. Based on the thermal behavior parameter set of each feature region in the thermal behavior image, thermal conduction logic verification is performed on the thermal deviation cluster region. Deviation regions that cannot be explained by known thermal conduction relationships are marked. For each thermal deviation cluster region, the basic thermal inertia parameters and environmental response sensitivity parameters of the feature region where the thermal deviation cluster region is located and all its spatially adjacent feature regions are read from the thermal behavior image. The expected time delay order of each feature region's response to thermal conduction is determined according to the magnitude of the basic thermal inertia parameters. The thermal deviation evolution curves of the thermal deviation cluster region and its adjacent regions within at least one complete environmental disturbance cycle before the real-time acquisition time are extracted from the continuous time series thermal images. The actual time order of the thermal deviation peaks of the two is compared. Simultaneously, the steady-state thermal balance level parameters and environmental response sensitivity parameters of the characteristic region where the thermal deviation accumulation area is located are read from the thermal behavior image. The theoretical environmental contribution temperature of the region at the current moment is calculated based on the change in environmental parameters at the current moment. The theoretical environmental contribution temperature and steady-state thermal balance level parameters are subtracted from the actual thermal deviation peak value of the thermal deviation accumulation area to obtain the residual thermal deviation. If the peak value of thermal deviation in the thermal deviation accumulation area appears earlier than that in its adjacent areas, or if the absolute value of the residual thermal deviation exceeds the seventh preset threshold, then the thermal deviation accumulation area is determined to be inexplicable by the heat conduction of the adjacent areas or its own historical thermal behavior, and the complete feature area where the thermal deviation accumulation area is located is marked as a deviation area that cannot be explained by the known heat conduction relationship. Within the marked deviation region, at least one pixel-level heat accumulation core is determined by extreme value search and clustering of the spatial distribution of thermal deviation, serving as an endogenous thermal anomaly source. For each marked deviation region, the local maximum value of thermal deviation at each pixel position is calculated, and the pixel with the largest value in the local neighborhood is retained as a candidate core point. Spatial proximity clustering is performed on the candidate core points, and adjacent candidate core points with an Euclidean distance less than the eighth preset threshold are merged into the same heat accumulation core. The arithmetic mean of the pixel coordinates of all merged core points is used as the final positioning coordinates of the core. For each heat accumulation core, verify whether the residual thermal deviation at the core location exceeds the seventh preset threshold and the actual thermal deviation at the core location, after deducting the residual conduction deviation contributed by the expected conduction from the adjacent feature regions, exceeds the ninth preset threshold. Heat accumulation cores that simultaneously meet the above conditions are identified as endogenous thermal anomaly sources, and the pixel coordinates, thermal deviation peak value, and feature region identifier of the anomaly source are recorded. Endogenous thermal anomaly sources are output as anomalies.

Images