Infrared thermal imaging full-dimension monitoring method for corrosion of pipeline under thermal insulation layer

Abstract

The invention relates to the technical field of pipeline detection, in particular to an infrared thermal imaging full-dimensional monitoring method for corrosion of a pipeline under a thermal insulation layer, which comprises the following steps of: acquiring temperature distribution under the thermal insulation layer and screening measuring points with consistent directions, extracting a boundary continuous region to calculate thermal probability, identifying abnormal measuring points and extracting time sequence characteristics, and predicting and marking response delay measuring points based on a model. And generating a CUI risk region division result. According to the method, a temperature difference direction consistency identification mechanism is constructed through space sequence subdivision and heat flow direction trend judgment of pipeline surface temperature distribution data, a measurement point set with clear space heat conduction characteristics is obtained through continuous screening, and a gradient change rate local extreme value is further taken as an identification basis; the method comprises the following steps: constructing a thermal anomaly probability evaluation parameter by combining spatial density and change intensity to realize preliminary identification of an abnormal point, then introducing a time sequence temperature parameter to construct a multi-dimensional dynamic feature set, and completing multi-step temperature prediction through normalized combination calculation after model training.

Description

Technical Field

[0001] The present invention relates to the field of pipeline detection technology, and in particular to a full-dimensional infrared thermal imaging monitoring method for pipeline corrosion under an insulation layer. Background Art

[0002] The field of pipeline inspection technology involves the inspection and evaluation of the structural status and operational safety of industrial transmission pipelines during their service life, including the determination of defects, corrosion, cracks, sediments, etc. on the outer and inner walls of the pipelines, and the monitoring and prevention of pipeline safety under long-term operation in combination with different environmental conditions.

[0003] Among them, the full-dimensional monitoring method of pipeline corrosion under the insulation layer by infrared thermal imaging refers to the use of infrared thermal imaging equipment to collect temperature distribution information on the pipeline surface under the working conditions where the pipeline is covered with an insulation layer in order to detect possible metal corrosion under the insulation layer. The heat flow conduction characteristics formed by external heating or natural temperature difference conditions are combined with multi-angle and multi-directional thermal imaging observation methods to obtain comprehensive thermal distribution data on the outer surface of the pipeline.

[0004] The infrared thermal imaging information of the pipeline collected without removing the insulation layer only reflects the external distribution of heat conduction and lacks the ability to dynamically analyze the internal thermal change trend. The existing technology relies on observing thermal distribution images at different angles. Although it can cover the surface area, it cannot accurately distinguish which locations have continuous thermal conduction anomalies. It lacks quantitative judgment of the time series temperature rise trajectory and does not distinguish the essential difference between abnormal temperature response and systematic temperature fluctuation. This leads to the risk of false alarms in thermal interference environments. For example, in areas with short-term sunlight exposure or air duct interference, the temperature distribution image may appear uneven, but no corrosion actually occurs. Due to the lack of temperature change trend comparison and spatial aggregation structure analysis, the existing method is difficult to accurately identify corrosion hazards and is prone to missing areas with strong thermal conductivity delay under the insulation layer, which in turn causes the problem of local corrosion development not being discovered in time. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a full-dimensional monitoring method for pipeline corrosion under the insulation layer using infrared thermal imaging.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a full-dimensional infrared thermal imaging monitoring method for pipeline corrosion under insulation layer, comprising the following steps: S1: Obtain the surface temperature distribution data of the pipe under the insulation layer, divide the surface temperature distribution data into equally spaced temperature measurement points, determine the consistency of the temperature change direction corresponding to each temperature measurement point, and select temperature measurement points with consistent directions; S2: obtaining spatial position information and gradient change rate peaks of the temperature measurement points with consistent directions, identifying boundary continuous temperature measurement points where the gradient changes between adjacent temperature measurement points are continuous, and calculating boundary thermal probabilities of the boundary continuous temperature measurement points; S3: The outer wall of the pipe under the insulation layer screens abnormal response temperature measurement points whose boundary thermal probability exceeds the abnormal probability threshold, and extracts an abnormal temperature time series feature data set of the abnormal response temperature measurement points; S4: Inputting the abnormal temperature time series feature dataset into the long short-term memory network (LSTM) model to perform multi-step temperature prediction of abnormal response temperature measurement points and mark the response delay temperature measurement points; S5: Extract a measurement point cluster area on the outer wall of the pipeline below the insulation layer based on the spatial distribution of the response delayed temperature measurement points, and generate a CUI risk area division result based on the ratio of the number of response delayed temperature measurement points in the measurement point cluster area to the total number of temperature measurement points.

[0007] As a further solution of the present invention, the steps of obtaining the temperature measurement points in the same direction are specifically as follows: S111: Obtaining temperature distribution data on the surface of the pipe under the insulation layer, evenly dividing the data into equally spaced temperature measurement points based on spatial coordinates, establishing a corresponding temperature value recording point at each temperature measurement point, and constructing a temperature measurement point distribution array; S112: For each temperature measurement point in the temperature measurement point distribution array, calling the temperature value corresponding to the temperature measurement point and the temperature value of the adjacent temperature measurement point, calculating the direction of change of the temperature difference in each direction, and determining whether the change trends in adjacent directions are the same, marking the temperature measurement points with consistent directions as meeting the direction consistency condition, and obtaining a direction consistency identification mark set; S113: Identify a chain of temperature measurement points that are adjacent and continuous and meet direction consistency in the direction consistency identification mark set, remove discrete measurement points with discontinuous direction change trends, and obtain direction consistent temperature measurement points.

[0008] As a further solution of the present invention, the step of obtaining the boundary thermal probability is specifically as follows: S211: obtaining spatial position information and the corresponding gradient change rate of each temperature measurement point among the temperature measurement points with consistent directions, extracting the position where the gradient change rate reaches a local maximum value as a representative value of the gradient mutation point, and obtaining a gradient change rate peak set; S212: identifying, based on the gradient change rate peak set, adjacent direction-consistent temperature measurement points whose gradient change rates are greater than a boundary gradient threshold and whose positions are continuous, and marking the identified adjacent direction-consistent temperature measurement points as boundary-continuous temperature measurement points to generate a boundary-continuous temperature measurement point set; S213: Calculate the proportional relationship between the density and the gradient change rate peak of each boundary continuous temperature measuring point in the boundary continuous temperature measuring point set on the outer wall of the pipeline under the heat preservation layer in the circumferential direction, and obtain the boundary heat probability of the boundary continuous temperature measuring point.

[0009] As a further scheme of the present application, the step of obtaining the abnormal temperature time sequence feature data set is specifically: S311: Compare the boundary heat probability corresponding to all boundary continuous temperature measuring points with the preset abnormal probability threshold point by point, filter the temperature measuring points whose boundary heat probability exceeds the abnormal probability threshold, and mark them as abnormal response temperature measuring points to generate an abnormal response temperature measuring point identification result; S312: According to the abnormal response temperature measuring point identification result, extract the temperature value sequence of each abnormal response temperature measuring point in the continuous time frame, calculate the temperature change between adjacent time frames, and obtain the total temperature rise value corresponding to the abnormal response temperature measuring point, to obtain an abnormal response temperature measuring point temperature change parameter group. S313: Based on the abnormal response temperature measuring point temperature change parameter group, combine and archive the temperature change, temperature change rate and cumulative temperature rise value corresponding to each abnormal response temperature measuring point, and obtain an abnormal temperature time sequence feature data set.

[0010] As a further scheme of the present application, the step of obtaining the response delay temperature measuring point is specifically: S411: Input the abnormal temperature time sequence feature data set into a long short-term memory network (LSTM) model, call the temperature change, temperature change rate and cumulative temperature rise value corresponding to each abnormal response temperature measuring point, and perform multi-step temperature prediction operation to obtain an abnormal response temperature measuring point predicted temperature sequence; S412: Call the abnormal response temperature measuring point predicted temperature sequence respectively, call the actual temperature value sequence in the historical observation record which is the same as the prediction time period, and calculate the corresponding difference value between the predicted value of each abnormal response temperature measuring point and the actual value in the historical record according to the time step to obtain an abnormal response temperature measuring point prediction residual sequence; S413: Count the number of time steps of continuous positive bias or continuous negative bias in each abnormal response temperature measuring point in the abnormal response temperature measuring point prediction residual sequence, and determine whether the time step duration exceeds the response delay determination threshold, if the condition is met, mark the corresponding abnormal response temperature measuring point as a response delay state, and obtain a response delay temperature measuring point.

[0011] As a further scheme of the present application, the step of obtaining the CUI risk area division result is specifically: S511: Locating the distribution area on the outer wall of the pipe below the insulation layer based on the spatial coordinate information of the response delay temperature measurement points, identifying the spatial clustering trend between adjacent response delay temperature measurement points, extracting interconnected measurement point areas in combination with a position continuity rule, and generating a clustered position group of response delay temperature measurement points; S512: Extracting spatially connected measurement point cluster regions within the outer wall of the pipe below the insulation layer based on the response delay temperature measurement point cluster position group, and calculating the ratio of the number of response delay temperature measurement points within each cluster region to the number of all temperature measurement points to obtain a response delay temperature measurement point ratio. S513: Based on the proportion of response-delay temperature measurement points and the location boundary and area information of the corresponding spatial clustering area, a spatial risk distribution structure covering the outer wall of the pipeline under the insulation layer is constructed. Areas with densely distributed response-delay temperature measurement points are identified as CUI risk areas, and the CUI risk area division results are obtained.

[0012] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, a temperature difference direction consistency identification mechanism is constructed by subdividing the spatial sequence of pipeline surface temperature distribution data and judging the heat flow direction trend. A set of measuring points with clear spatial thermal conductivity characteristics is obtained through continuity screening. The local extreme value of the gradient change rate is further used as the identification basis. The spatial density and change intensity are combined to construct thermal anomaly probability assessment parameters to achieve preliminary identification of anomalies. Time series temperature parameters are then introduced to construct a multidimensional dynamic feature set. Multi-step temperature prediction is completed through normalized combination calculation after model training. The prediction results are compared with historical data to form a residual sequence. The thermal response hysteresis behavior is judged by combining deviation persistence. Connectivity is identified based on the spatial clustering trend of this behavior, and delayed clustering area extraction is completed. The risk distribution level is constructed by calculating the proportion of response measuring points in the clustering area. Finally, the corrosion potential area under the insulation layer is marked, effectively realizing the continuous conversion of thermal distribution data into corrosion risk area judgment, improving the spatial accuracy and temporal resolution of corrosion hazard identification, enhancing the ability to identify high hysteresis heat conduction areas, reducing the misjudgment rate and omission risk, and improving the pertinence and applicability of the assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the main steps of the present invention;

[0014] Figure 2 This is a flow chart of step S1 of the present invention;

[0015] Figure 3 This is a flow chart of step S2 of the present invention;

[0016] Figure 4 This is a flow chart of step S3 of the present invention;

[0017] Figure 5 This is a flow chart of step S4 of the present invention;

[0018] Figure 6 This is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] See also Figure 1 The present invention provides a technical solution: a full-dimensional infrared thermal imaging monitoring method for pipeline corrosion under an insulation layer, comprising the following steps: S1: Obtain the surface temperature distribution data of the pipe under the insulation layer, divide the surface temperature distribution data into equally spaced temperature measurement points, determine the consistency of the temperature change direction corresponding to each temperature measurement point, and select temperature measurement points with consistent directions; S2: Obtain the spatial position information and gradient change rate peak of the temperature measurement points with consistent directions, identify the boundary continuous temperature measurement points with continuous gradient changes between adjacent temperature measurement points, and calculate the boundary thermal probability of the boundary continuous temperature measurement points; S3: The outer wall of the pipe under the insulation layer screens abnormal response temperature measurement points whose boundary thermal probability exceeds the abnormal probability threshold, and extracts the abnormal temperature time series feature data set of the abnormal response temperature measurement points; S4: Input the abnormal temperature time series feature dataset into the long short-term memory network (LSTM) model to perform multi-step temperature prediction of abnormal response temperature measurement points and mark the response delay temperature measurement points; S5: Based on the spatial distribution of the response-delayed temperature measurement points, a measurement point cluster area is extracted on the outer wall of the pipe under the insulation layer. The CUI risk area division result is generated based on the ratio of the number of response-delayed temperature measurement points in the measurement point cluster area to the total number of temperature measurement points.

[0021] See also Figure 2 The specific steps for obtaining temperature measurement points with consistent directions are as follows: S111: Obtaining temperature distribution data on the surface of the pipe under the insulation layer, evenly dividing the data into equally spaced temperature measurement points based on spatial coordinates, establishing a corresponding temperature value recording point at each temperature measurement point, and constructing a temperature measurement point distribution array; To obtain temperature distribution data on the pipe surface beneath the insulation layer, the surface is divided into evenly spaced temperature measurement points based on spatial coordinates. A corresponding temperature recording point is established at each temperature measurement point, forming a temperature measurement point distribution array. To obtain temperature distribution data on the pipe surface beneath the insulation layer, the insulation layer must first be partially removed or a non-destructive thermal image is acquired using infrared imaging. The pipe's outer wall is scanned and measured using a thermal imaging device. A coordinate expansion method is used to map the three-dimensional outer wall of the pipe onto a two-dimensional coordinate plane. Circumferential positions correspond to the Y direction (i.e., the pipe's circumference), while axial positions correspond to the X direction (i.e., the pipe's length). To represent spatial hierarchy or thickness, the Z direction can be introduced to stack multiple time-series images or different imaging angles to form a complete data volume structure. The temperature measurement points are divided into a uniform grid with a standard spacing of 1 cm × 1 cm in both the X and Y directions. The cylindrical outer wall is flattened to form a regular two-dimensional measurement point array. Each intersection coordinate point represents the location of a temperature measurement point. By mapping the thermal image coordinates to the actual physical coordinates, the pixel coordinates extracted from the image are converted to corresponding spatial coordinates, establishing a correspondence between the measurement points and their actual locations. Each recording point corresponds to a specific physical location and its temperature value. The temperature value at the time of collection is recorded for each measurement point, and synchronized through time stamps, forming a multidimensional array structure that includes spatial location, time series, and temperature data. To improve temperature measurement accuracy, multiple measurements are taken during point collection, and the arithmetic average is calculated. Outliers with deviations exceeding a preset fluctuation range (e.g., ±0.5°C) are eliminated, ultimately forming an array of evenly distributed temperature measurement points with stable data.

[0022] S112: For each temperature measurement point in the temperature measurement point distribution array, call the temperature value corresponding to the temperature measurement point and the temperature value of the adjacent temperature measurement point, calculate the direction of change of the temperature difference in each direction, and determine whether the change trends in adjacent directions are the same, mark the temperature measurement points with consistent directions as meeting the direction consistency condition, and obtain a direction consistency identification mark set; For each temperature point in the distributed array, the corresponding temperature value at that point is retrieved from the temperature values ​​of adjacent temperature points. The direction of the temperature difference in each direction is calculated, and the trend in adjacent directions is determined to be the same. Temperature points with consistent trends are marked as meeting the directional consistency condition, resulting in a directional consistency identification tag set. Specifically, the temperature value of the central point is read first. The temperature values ​​of adjacent points in eight directions (up, down, left, right, and diagonal) are then retrieved. The temperature differences are calculated and determined to be positive or negative to determine the heat flow direction. The heat flow trend labels for all eight directions are recorded. If a point exhibits consistent temperature trends (e.g., decreasing or increasing temperature) in any three or more adjacent directions, and the temperature difference exceeds a set threshold (e.g., 0.8°C), the point is marked as meeting the directional consistency condition. In practice, complex structures may occur, with some directions showing positive trends and others showing negative trends. To address this, the system allows groups of adjacent directions with consistent trends to be considered a directional consistency subset. As long as any of these groups meets the judgment criteria, the central point is marked as a directional consistency point. If multiple directional consistency subsets exist for the same central measurement point, they will be recorded in parallel and merged in subsequent processing. Because each measurement point is referenced multiple times in the neighborhood analysis, some measurement points may appear in multiple directional consistency marker sets. To address this duplicate marking phenomenon, a deduplication and priority retention mechanism is used in the subsequent processing process. This includes prioritizing the most representative chain group path based on the directional continuity length and temperature difference change amplitude of the chain group in which the measurement point is located. At the same time, duplicate markers are merged during the clustering and chain building process to avoid duplicate counting or directional label conflicts, thereby ensuring that the final selected set of directional consistency measurement points is unique and consistent with the physical heat conduction logic.

[0023] S113: identifying a chain of temperature measurement points that are adjacent and continuous and satisfy direction consistency in the direction consistency identification mark set, eliminating discrete measurement points with discontinuous direction change trends, and obtaining direction-consistent temperature measurement points; After the construction of the direction consistency identification mark set is completed, the measurement points in the mark set need to be further identified in spatial structure to extract continuous measurement point chain groups with clear heat conduction path characteristics. In the specific processing process, each measurement point marked as direction consistency is taken as the starting point, and according to its direction label, recursive search is carried out in the same heat flow direction to find the next measurement point adjacent to it, with the same direction label and the temperature difference still meeting the threshold condition, to construct a continuous direction consistency path. This recursive process requires that the actual physical connectivity between measurement points be maintained in space, that is, any two adjacent measurement points in a chain must be within the eight-neighborhood range and the distance between them must not exceed twice the standard measurement point distance. In the recursive process, if the direction label changes or the ΔT value is lower than the set threshold, the chain connection is terminated. This path tracking operation is performed for each measurement point to identify direction consistency paths with a length of not less than 3 continuous measurement points, which are referred to as direction consistency measurement point chain groups.

[0024] At the same time, for the measurement points in the mark set which are marked as direction consistency but cannot form a continuous chain meeting the length requirement in any direction path, they are regarded as discrete measurement points and should be removed. The judgment standard for discrete points is that in any one direction, there are less than 3 continuous adjacent direction consistency measurement points, or there are direction breaks and heat flow trend reversals, which do not have spatial heat conduction continuity and cannot form effective heat conduction structure, so they are not retained in subsequent calculations. In addition, in order to prevent the repetition of measurement points participating in multiple path groups due to the intersection of multiple chain groups, conflict detection is performed on all formed chain groups, and the chain group path with longer continuous length or higher average temperature difference intensity is preferentially retained, and the remaining overlapping measurement points can be merged according to the direction consistency score strategy to ensure that each measurement point belongs to only one effective chain group.

[0025] Please refer to Figure 3 The specific steps for obtaining the boundary heat probability are as follows: S211: Obtain the spatial position information and corresponding gradient change rate of each temperature measurement point in the direction consistency temperature measurement point, extract the position of the local maximum value of the gradient change rate as the representative value of the gradient mutation point, and obtain the gradient change rate peak set; To obtain the spatial location information of consistent temperature measurement points, the mapping relationship between pixel coordinates extracted from the thermal image and the measurement point layout in physical space is combined to establish a two-dimensional coordinate matrix of the measurement points in the physical coordinate system, ensuring that each temperature measurement point has identifiable location attributes. The term "adjacent measurement points" here refers to the screened consistent temperature measurement points, not all original measurement points, to ensure the continuity of the gradient analysis and the physical rationality of thermal conductivity. Based on this, the gradient change rate of each measurement point is calculated. Specifically, the ratio of the temperature difference to the directional distance between the measurement point and its adjacent measurement points in the upper, lower, left, right, and four diagonal directions is calculated. The maximum value is selected as the representative gradient change rate of the measurement point, which is used to sensitively identify localized sharp temperature change trends. To further reduce the impact of measurement fluctuations, based on the average of multiple temperature data measurements, "iterative averaging" is introduced here to perform time-series averaging of the gradient change rates of the same measurement point over multiple time sections. This eliminates occasional interference and improves the stability of gradient abrupt change detection. After calculating the gradient rate of change, a local maximum is determined for all temperature measurement points with consistent orientations. If the gradient rate of change at a particular point is higher than that of all other measurement points in its eight neighboring regions and is at least 20% greater than the average gradient rate of change for its neighboring measurement points, the point is marked as a gradient mutation point and recorded as a local extreme value. The "20% threshold" set here is based on the extraction of variation characteristics between normal heat flux attenuation areas and mutation areas from field measurement data: In well-insulated areas, the temperature gradient is typically around 1.0°C / cm, while in areas of corrosion or thermal resistance mutation, the gradient often rises to 1.2°C / cm or higher. Therefore, using the neighborhood mean as the baseline and setting a 20% amplitude as the mutation judgment threshold can effectively distinguish between stable heat flux and abnormal mutation areas. Finally, all local extreme value points that meet the above conditions are aggregated to form a gradient rate of change peak set.

[0026] S212: identifying, based on the gradient change rate peak set, adjacent direction consistent temperature measurement points whose gradient change rate is greater than a boundary gradient threshold and whose positions are continuous, and marking the identified adjacent direction consistent temperature measurement points as boundary continuous temperature measurement points to generate a boundary continuous temperature measurement point set; Starting from each measurement point with a peak gradient rate, path extension is performed in eight directions. Path extension is performed only along the direction consistent with the heat flow direction of the starting point. This requires that all measurement points along the extended path maintain the same direction as the starting point in their directional labels and are themselves marked as directionally consistent temperature measurement points. On each directional path, each point is tested to see if the gradient rate of change between it and the previous point exceeds a set boundary gradient threshold. Furthermore, the spatial continuity between these measurement points is determined, meaning that the physical distance between these measurement points does not exceed twice the standard measurement point spacing. If three or more measurement points in a given direction meet both of these conditions, the path exhibits a stable high-gradient abrupt change trend. In this case, all directionally consistent temperature measurement points along the path are marked as boundary-continuous temperature measurement points. The directionality requirement involves three levels: first, the starting point must be a directionally consistent temperature measurement point; second, all measurement points traversed during the path extension must also be directionally consistent temperature measurement points; and third, the path direction must be consistent with the heat flow direction of the starting point. Directional jumps or reverse reversals are not permitted. The continuity condition refers to the dual requirements of gradient values ​​exceeding the threshold and spatial continuity in the aforementioned determination. Ensure that the extracted continuous boundary temperature measurement points form a heat conduction path in physical space that is guided by the same heat flow direction, has a significant thermal gradient mutation, and is continuous in position.

[0027] S213: Calculating the proportional relationship between the density of each boundary continuous temperature measurement point in the boundary continuous temperature measurement point set on the circumferential direction of the outer wall of the pipe under the insulation layer and the peak value of the gradient change rate by a normalized weighted calculation method, and obtaining the boundary thermal probability of the boundary continuous temperature measurement point; The boundary thermal probability of the boundary continuous temperature measuring points is obtained by normalized weighted calculation. First, two key variables need to be clarified: one is the density of the boundary continuous temperature measuring points on the circumference of the outer wall of the pipeline, and the other is the peak value of the gradient change rate corresponding to the boundary continuous temperature measuring points. In the implementation process, the density is obtained by counting the number of boundary continuous temperature measuring points marked per unit circumferential length. The unit length is set to 10 cm, and a cylindrical expansion map is constructed on the thermal image. The circumference is expanded into a one-dimensional linear coordinate, divided into segments in sequence, and the number of measuring points in each segment is recorded. The ratio of the number of measuring points in the segment to the maximum number of measuring points in the segment is taken as the normalized density weight; secondly, the peak value of the gradient change rate of all boundary measuring points in the corresponding segment is extracted, the average value is taken, and the value is divided by the maximum gradient change rate value of all boundary continuous temperature measuring points to obtain the normalized gradient intensity coefficient; finally, the boundary thermal probability of each measuring point or segment is calculated using the following weighted formula: ;in, :Indicates the Boundary thermal probability of the segment; :Indicates the The number of continuous temperature measurement points at the boundary of the segment; : represents the maximum number of continuous temperature measurement points at the boundaries of all segments; :Indicates the The peak value of the average gradient change rate of continuous temperature measurement points at the boundary of the segment; : Indicates the peak value of the maximum gradient change rate of all continuous boundary temperature measurement points.

[0028] In this formula, the two components normalize the distribution density of the measuring points and the gradient intensity respectively, so that different sections have a unified standard when compared. The product reflects the possibility that the area has both high density and high gradient jump. The final boundary thermal probability value is The value ranges from 0 to 1 and can be used to rank and grade the intensity of thermal anomaly distribution, serving as an important quantitative basis for identifying risk areas. In actual operations, data extraction relies on thermal imaging image analysis modules, coordinate expansion and conversion tools, and numerical processing tools such as Python scripts or nested functions in Excel for batch calculations. All parameters are derived from the collected coordinate-temperature dataset and its second-order calculation results, without external estimates or subjective settings. A comparison benchmark is only required during the normalization phase, with the current sample maximum value as the denominator to ensure the internal consistency of the proportional expression.

[0029] Suppose a section of pipe under insulation is measured by thermal imaging. It is expanded into cylindrical coordinates and divided into 5 equal length segments, each 10 cm long and with a total circumference of 50 cm. The analysis goal is to calculate the boundary thermal probability of the third segment. .

[0030] Get the number of continuous temperature measurement points on the boundary , after the image coordinates are expanded, the continuous temperature measurement points are marked on the boundaries: 8 in the first segment, 12 in the second segment, 20 in the third segment, 10 in the fourth segment, and 15 in the fifth segment, so: , ;

[0031] Get the average gradient change rate of continuous temperature measurement points on the boundary ,The change rates of all continuous boundary measurement points in the 3rd segment are as follows (in °C / cm): 1.25, 1.30, 1.20, 1.10, 1.40, 1.35, 1.28, 1.22, 1.18, 1.26 (a total of 10 sample points);

[0032] Calculate the average: ;

[0033] Assume that the maximum gradient change rate peak value of all continuous boundary temperature measurement points is: ;

[0034] Substitute into the formula to calculate , substitute the above values ​​into:

[0035] ;

[0036] The result indicates that the comprehensive evaluation of the third paragraph on the boundary thermal risk is 0.8957.

[0037] The core purpose of the whole calculation process is to identify the possible thermal abnormal area of the outer wall of the pipeline under the insulation layer in a quantitative way and to conduct risk level assessment. The process first divides the development area of the outer wall of the pipeline, counts the number of boundary continuous temperature measuring points identified in each section, and uses it to reflect the spatial density distribution of temperature mutation in the area. Secondly, the temperature gradient change rate of these measuring points is extracted, and the average value is calculated to measure the severity of the thermal mutation. In the calculation, by comparing the number of measuring points with the largest number in each section, the relative weight of the density is obtained, and by comparing the average value of the gradient change rate with the maximum gradient change rate in the whole section, the strength weight is obtained. The product of the two weight values is the boundary thermal probability of the region, which is used to represent the possibility of the region to have multiple measuring points with concentrated and severe temperature difference characteristics. The reason for using the product is that the density of measuring point distribution reflects the spatial scale of thermal anomaly, and the gradient change rate reflects the severity of thermal anomaly. The combination of the two can fully express the real characteristics of the thermal instability risk area. If only one of them is considered, it may appear to be a misjudgment of occasional strong gradient or large area of slight change.

[0038] Please refer to Figure 4 , the steps of obtaining the abnormal temperature time series feature data set are:

[0039] S311: Compare the boundary thermal probability corresponding to all boundary continuous temperature measuring points with the preset abnormal probability threshold point by point, filter the temperature measuring points with boundary thermal probability exceeding the abnormal probability threshold, and mark them as abnormal response temperature measuring points, and generate abnormal response temperature measuring point identification results;

[0040] When determining anomalies for all continuous boundary temperature measurement points, the boundary thermal probability value corresponding to each measurement point must be compared with a pre-set anomaly probability threshold. This threshold should be calibrated based on typical thermal anomaly cases in historical data. Typically, a lower probability limit that consistently reflects thermal anomaly regions is selected as a benchmark. For example, after multiple experiments and analysis of actual thermal failure cases, it has been found that regions with boundary thermal probabilities above the 75th percentile mostly correspond to areas with a high risk of thermal failure. Therefore, this 75th percentile value can be used as the anomaly probability threshold. All measurement points with thermal probability values ​​exceeding this threshold are considered to be thermal anomaly response points and are identified as abnormal response temperature measurement points. During the judgment process, each boundary thermal probability is retrieved and compared point by point, and the calculated value corresponding to each measurement point is called. If the value exceeds the set threshold, a marker is added and its spatial coordinates and timestamp are recorded. Finally, all measurement points that meet the conditions are aggregated to generate the abnormal response temperature measurement point identification results, forming a complete spatial distribution information set containing abnormal measurement points.

[0041] S312: Based on the abnormal response temperature measurement point identification result, extract the temperature value sequence of each abnormal response temperature measurement point in the continuous time frame, calculate the temperature change between adjacent time frames, and obtain the total temperature rise value of the time series corresponding to the abnormal response temperature measurement point to obtain the temperature change parameter group of the abnormal response temperature measurement point; After identifying temperature measurement points with abnormal responses, the time series temperature data for each marked measurement point during the continuous monitoring process must be extracted. This means obtaining a sequence of temperature values ​​corresponding to each measurement point at multiple consecutive sampling moments. The time interval is typically set between 1 and 5 seconds, depending on the rate of temperature change under actual operating conditions, to ensure that sufficiently detailed temperature trends are captured. Based on this, the temperature difference between adjacent time points is calculated: the current temperature minus the previous temperature, resulting in the temperature difference at each time step. The sign of the temperature difference reflects the direction of temperature change: a positive value indicates an increase in temperature, while a negative value indicates a decrease in temperature.

[0042] At the same time, the temperature differences at all time points (regardless of whether they are positive or negative) are accumulated sequentially to obtain the total temperature rise value for that point during the entire monitoring period. This value is used to determine whether the temperature is trending towards continuous growth or sudden changes. For example, if a certain abnormal point shows a temperature increase trend for five consecutive time periods, and the cumulative temperature rise exceeds 5 degrees Celsius, it indicates that there is a stable and significant thermal change at that point.

[0043] During the processing, it's also necessary to eliminate abnormal temperature fluctuations. For example, if the temperature rises abnormally for a short period of time but then quickly drops back down, this data should be smoothed using a sliding average or rate-of-change determination mechanism, or marked as invalid. Ultimately, a complete abnormal response temperature change parameter set is constructed, including the temperature change amount (including the direction of change), the time point, and the cumulative temperature rise value.

[0044] S313: Based on the temperature change parameter group of the abnormal response temperature measurement points, the temperature change amount corresponding to each abnormal response temperature measurement point, the temperature change rate of adjacent time frames, and the accumulated temperature rise value are combined and archived to obtain an abnormal temperature time series feature data set;

[0045] Based on the temperature change parameter group, the dynamic temperature characteristics of each abnormal response temperature measurement point need to be structured and combined to establish an abnormal temperature time series feature dataset for subsequent analysis. During the construction process, three core parameters corresponding to each measurement point need to be extracted separately, including the temperature change between each time period, the temperature change rate between each two time points, and the cumulative temperature rise value from the start of anomaly identification to the current time point. These parameters are archived and stored in chronological order, and the temperature change range, velocity gradient, and cumulative change trend within each time period are recorded in the data structure, forming a basic information set for abnormal behavior modeling and trend prediction. This feature dataset can be used to identify typical abnormal evolution patterns, such as continuous slow rise, periodic fluctuation, or sudden jump. It is structurally stored in a multi-field format divided by time segments. Each record corresponds to a time period and contains multiple indicators such as temperature difference, rate, and cumulative change.

[0046] See also Figure 5 ,The specific steps for obtaining the response delay temperature measurement point are:

[0047] S411: Input the abnormal temperature time series feature dataset into the long short-term memory network (LSTM) model, call the temperature change, temperature change rate, and cumulative temperature rise value corresponding to each abnormal response temperature measurement point, perform a multi-step temperature prediction operation, and obtain a predicted temperature sequence for the abnormal response temperature measurement point;

[0048] To accurately predict the future state of abnormally responsive temperature measurement points, this solution uses a long short-term memory (LSTM) model as a prediction tool. Combined with key input variables in the abnormal temperature time series feature dataset, this solution performs multi-step temperature predictions using a time-dependent weight structure fitted by the model. In actual engineering, the LSTM model captures short-term fluctuations and long-term trends in temperature series through memory cells and gating mechanisms, dynamically mapping input features to historical change structures. However, during deployment, especially during inference, the complex internal structure of the LSTM model can be approximated through formulaic expression, allowing the influence relationships between various input features to be presented in numerical form. This simplifies the model prediction into a specific, executable temperature calculation formula, as follows:

[0049] ;

[0050] This formula is based on the structural reasoning capability and time series feature recognition logic of the LSTM model. Specifically, the model output is approximated as a weighted prediction item based on the current temperature, which is composed of three normalized temperature features. In the formula, Indicates the predicted temperature value at the next moment in degrees Celsius, which is the final output result; Indicates the actual temperature value at the current moment, which is collected through on-site thermal imaging or temperature sensing; Represents the temperature change between the current moment and the previous moment, that is, , used to reflect short-term temperature change trends; and Respectively represent the mean and standard deviation of historical temperature changes, which are the reference benchmarks required for standard normalization of this feature. The data source is the historical observation records of the corresponding measuring point under long-term normal operation; is the temperature change rate at the current moment, which represents the temperature increment per unit time. Calculated by dividing by the sampling time interval, it reflects the intensity of instantaneous changes; its corresponding normalization parameter and are the mean and standard deviation of the historical rates, respectively; It represents the cumulative temperature rise value, that is, the sum of the temperature changes at all consecutive moments since a certain measuring point was identified as an abnormal response. It is used to reflect the medium- and long-term cumulative effect of temperature rise. Its corresponding normalization parameter is and , are the mean and standard deviation of the historical cumulative temperature rise values, which are obtained by counting the historical cumulative temperature rise values ​​of a specified period and calculating the mean and standard deviation respectively. 、 and These are the weighting coefficients corresponding to the three types of normalized features. These coefficients are not set manually, but are automatically adjusted and learned by the error minimization principle during the LSTM model training phase, reflecting the influence of each type of feature on the final temperature evolution.

[0051] The "current moment" and "previous moment" are defined based on the sampling interval. Typically, the sampling interval is set between 1 and 5 seconds, depending on the rate of temperature change. The time difference between each sampling interval is the "time step," which corresponds one-to-one to an actual time interval (such as every second or every minute). For example, if temperature data is collected every 1 second, each time step represents a 1-second interval. During the prediction process, the LSTM model makes predictions based on the current time step (i.e., the temperature value at the current sampling moment) and the temperature change at the previous moment.

[0052] This formula expresses the temperature at the next moment, output by the LSTM during the prediction phase, based on the model training results, by combining the three features mentioned above with the current temperature. This formula is suitable for step-by-step sliding prediction scenarios based on the current time. By iteratively applying this formula, multi-step temperature prediction can be achieved, using the current prediction result as the basis for the next step's input, and sequentially deducing a temperature sequence at successive time points.

[0053] Assume the current time point is the 10th second, and the prediction target is the temperature at the 11th second. , current temperature (10th second): ℃, sampling interval: 1 second; current temperature change: ℃; Current temperature change rate: ℃ / s; Current cumulative temperature rise (total temperature rise since the abnormal response started): ℃; historical average temperature change ℃; standard deviation of temperature change ℃; historical average of temperature change rate ℃ / s; standard deviation of temperature change rate ℃ / s; historical average of cumulative temperature rise ℃; Cumulative temperature rise standard deviation ℃; weighted coefficient obtained from model training , , .

[0054] Normalization of temperature change: ;

[0055] Normalized temperature change rate: ;

[0056] Normalized cumulative temperature rise: ;

[0057] Substitute the formula to calculate the predicted temperature:

[0058] ;

[0059] The predicted temperature at the 11th second is about 52.07°C, which is a significant increase compared to the current 48.0°C. This indicates that the abnormal response measuring point is continuing to heat up and has a strong thermal anomaly trend, which is consistent with the previous judgment logic.

[0060] S412: Recall the predicted temperature sequence of the abnormal response temperature measuring point, call the actual temperature value sequence in the historical observation record for the same period as the prediction, calculate the corresponding difference between the predicted value of each abnormal response temperature measuring point and the actual value in the historical record according to the time step, and obtain the abnormal response temperature measuring point prediction residual sequence; After the output of the predicted temperature sequence for each abnormal response temperature measuring point is completed, the measured temperature sequence strictly corresponding to the prediction time period needs to be extracted from the historical observation record, and the two need to be completely aligned in time to maintain the time step correspondence. On this basis, the difference between the predicted temperature value and the actual temperature value is calculated according to the time step, that is, the predicted value of each time step is subtracted by the measured value of the corresponding time, to obtain the prediction residual sequence of the measuring point in the time period. The historical observation record is obtained by the normal continuous monitoring stage of the abnormal response temperature measuring point before being identified as abnormal. This process automatically records the temperature value of the measuring point by the thermal imaging device or the temperature sensor according to the set sampling interval. The sampling interval can be set to once per second or once every 5 seconds according to the actual working condition, and the acquisition time range generally covers at least several minutes of data before the prediction starts, which is used to fully reflect the temperature change background of the measuring point. The record content includes the measuring point number, the acquisition time, the temperature value and the corresponding spatial coordinates, and is arranged in time sequence to form time sequence data as the basis for subsequent comparison with the predicted sequence. When calling the historical record, the measured temperature value of the corresponding time point is extracted according to the starting time of the prediction sequence backtracking for a corresponding time period, to ensure accurate matching of the time steps. The difference calculation process is executed point by point, and the predicted value and the measured value are compared according to the same time point to obtain the residual value of each step. In order to ensure the accuracy of the analysis, the missing time points or data format abnormal records in the historical data need to be removed, and only the continuous segment with complete time and valid data is retained for analysis. The final residual sequence can be used to reflect the difference between the prediction trend and the actual temperature change.

[0061] S413: Count the number of consecutive positive or consecutive negative time steps in each abnormal response temperature measuring point in the predicted residual sequence of the abnormal response temperature measuring point, and determine whether the time step duration exceeds the response delay determination threshold. If the condition is met, the corresponding abnormal response temperature measuring point is marked as a response delay state, and the response delay temperature measuring point is obtained. After obtaining the prediction residual sequence, statistical analysis is performed on time periods with consistent deviations to identify whether a measurement point exhibits response delay characteristics. The specific judgment rule is: for each abnormally responding temperature measurement point, the number of consecutive positive or negative time steps in the residual sequence is counted. If there are periods of consecutive positive deviations (i.e., predicted values ​​consistently higher than actual values) or consecutive negative deviations (i.e., predicted values ​​consistently lower than actual values) for each measurement point, and the duration of these periods exceeds the set response delay threshold, the measurement point is considered to have a lagging actual temperature response despite the predicted temperature increasing or decreasing, and is thus marked as a response delay temperature measurement point. The threshold setting principle should be based on the typical thermal inertia response time of the measurement point. For example, the response delay time calculated based on the pipe material, fluid temperature response characteristics, and heat conduction path is sufficient. If, under normal conditions, the actual change delay after a sudden temperature change does not exceed 5 seconds, the judgment threshold can be set to 5 consecutive time steps (with a sampling period of 1 second). In other words, a delayed response is only marked if the deviation direction occurs 5 or more times in a row. During the judgment process, the influence of short-term fluctuation errors must also be eliminated. For example, if there is a one-time directional interruption followed by a continuous deviation, a sliding window approach should be used to accommodate this. Finally, the measurement points that meet the response delay judgment criteria are summarized as the response delay temperature measurement point set.

[0062] See also Figure 6 , the specific steps for obtaining the CUI risk area division results are as follows: S511: Locate the distribution area on the outer wall of the pipe below the insulation layer based on the spatial coordinate information of the response delay temperature measurement points, identify the spatial clustering trend between adjacent response delay temperature measurement points, extract interconnected measurement point areas based on the position continuity rule, and generate a clustered position group of response delay temperature measurement points; First, the spatial coordinates of the delayed response temperature measurement points are mapped to their actual physical locations on the outer wall of the pipe beneath the insulation layer. This process is achieved by establishing a mapping relationship between image coordinates and physical coordinates, converting the temperature measurement points from image coordinates to the actual coordinates of the pipe's outer surface. Next, the spatial connectivity between adjacent measurement points is determined. The maximum allowable distance between measurement points is generally set to twice the standard sampling interval. If the distance between two measurement points does not exceed this maximum distance, they are considered potentially connected. In addition, the temperature change direction of two measurement points must be consistent to be considered spatially connected. Specifically, if the temperature difference between two measurement points changes in the same direction in eight possible directions (upward, downward, leftward, and four diagonal directions), the two measurement points are considered spatially connected. By recursively searching for all connectable measurement points, multiple delayed response measurement points are grouped into the same cluster. To exclude isolated points or pseudo-clustered areas, a minimum connectivity threshold is set: a region must contain at least three consecutive connected measurement points to be considered valid spatial clustering. Each valid clustered area is assigned a unique number, forming a clustered location group for delayed response temperature measurement points.

[0063] S512: Extracting spatially connected measurement point cluster regions within the outer wall of the pipe below the insulation layer based on the response delay temperature measurement point cluster position group, and calculating the ratio of the number of response delay temperature measurement points within each cluster region to the number of all temperature measurement points to obtain a response delay temperature measurement point ratio. To extract and analyze clustered locations of response-delay temperature measurement points, we first calculate the boundaries of the region within which each measurement point resides to determine the spatial footprint of the clustered region. Specifically, for all response-delay measurement points within each clustered region, the region's boundary envelope is determined based on their spatial coordinate information, generating a minimum closed polygon that accurately represents the region's physical spatial extent. For example, if a clustered region contains five response-delay measurement points, their spatial coordinates are used to calculate the region's minimum enclosing polygon, assuming the polygon has an area of ​​10 square meters. Next, within this boundary envelope, the total number of temperature measurement points and the number of response-delay temperature measurement points are counted. Assuming that within a specific clustered region, there are 100 total temperature measurement points, of which 20 are response-delay measurement points, the percentage of response-delay temperature measurement points is 20 / 100 = 0.2, meaning that the percentage of response-delay temperature measurement points within the clustered region is 20%. The calculation process of this proportion value is completed on the basis of standard grid mapping. By demarcating a grid in the aggregation area, the number of all measurement points in the area is used as the denominator, and the number of response delay temperature measurement points is used as the numerator, and finally the proportion of response delay temperature measurement points in the area is obtained.

[0064] Regarding the setting of thresholds, it is usually determined based on the statistical relationship between the cluster distribution and defect location in historical pipeline corrosion events. For example, historical data may show that when the proportion of response delay measurement points is greater than 50%, the corrosion risk of the pipeline area increases significantly. Therefore, a proportion threshold of 0.5 is set. That is, if the proportion of response delay measurement points in the area is greater than 50%, the area is considered to be a "high response delay cluster area". For example, in another cluster area, there are 60 response delay measurement points and the total number of measurement points is 100. The proportion is 60 / 100=0.6, indicating that the area belongs to a high response delay cluster area and may have a higher corrosion risk. Therefore, this method can effectively identify response delay areas of different severity.

[0065] S513: Based on the proportion of response-delay temperature measurement points and the location boundary and area information of the corresponding spatial clustering area, a spatial risk distribution structure covering the outer wall of the pipeline under the insulation layer is constructed. The area where the response-delay temperature measurement points are densely distributed is identified as a CUI risk area, and a CUI risk area division result is obtained. Based on the proportion of response delay temperature measurement points and the spatial range boundary information of each cluster area on the outer wall of the pipeline under the insulation layer, a spatial risk distribution structure for identifying potential corrosion risks needs to be constructed. This structure divides the outer wall area of ​​the pipeline into multiple sub-segments, embeds the corresponding proportion of the cluster area in each segment, and identifies different risk levels through color marking or graded labels. For cluster areas whose proportion values ​​exceed the set judgment threshold, they are marked as CUI risk areas, that is, areas where corrosion under the insulation layer may occur. During the construction process, the area of ​​the cluster area needs to be measured, that is, the area is calculated through the coordinate boundary of the enveloping polygon to obtain the specific coverage area of ​​each risk area, thereby establishing a risk level map covering the entire outer wall of the pipeline.

[0066] This method uses infrared thermal imaging technology to obtain temperature distribution data on the pipeline surface, combining spatial and time series analysis to accurately determine the corrosion situation under the insulation layer. First, by analyzing the consistency of the temperature change direction, temperature measurement points with stable heat flow characteristics are screened out. Their spatial position and gradient change information are further extracted to identify areas with sudden changes in thermal gradients. Next, by comparing the thermal probabilities of these areas, abnormal temperature response points are found. Multi-step temperature predictions are performed on them using a long short-term memory network, identifying measurement points with response delays. These measurement points reflect areas where corrosion problems may exist. Finally, by analyzing the spatial distribution of response delay measurement points, high-risk clusters are identified, and CUI (corrosion under insulation) risk areas are generated based on the degree of measurement point clustering.

[0067] CUI risk areas are areas of significant temperature anomalies and response delays identified through infrared thermal imaging monitoring of the pipeline surface beneath the insulation layer. Temperature changes in these areas reflect abnormal heat conduction behavior on the pipeline surface and may indicate corrosion issues within the pipeline. Specifically, CUI risk areas include the following levels of risk: Incipient Corrosion Risk: These areas may present mild temperature anomalies, perhaps manifesting as minor temperature changes but with a slight decrease in thermal conductivity. This indicates the beginnings of minor corrosion or thermal resistance accumulation on the pipeline surface. These areas may be signs of incipient corrosion and require regular monitoring and early intervention to prevent further deterioration.

[0068] Moderate Corrosion Risk: These areas experience significant temperature fluctuations, characterized by large temperature swings and significant differences from surrounding areas. This indicates the presence of more severe corrosion or uneven heat transfer caused by moisture accumulation beneath the insulation, indicating moderate corrosion progression within the pipe. These areas require further inspection and intervention to prevent further corrosion.

[0069] Severe corrosion risk: These areas experience significant temperature anomalies, often accompanied by significant thermal gradients or temperature increases, indicating significant heat flow disruption, most likely due to heat flow delays caused by severe corrosion within the pipe. Severe corrosion in these areas can lead to safety hazards such as weakened pipe strength or leaks, necessitating repair or replacement.

[0070] These CUI risk areas correspond to different degrees of corrosion risk. Through infrared thermal imaging technology and time series analysis, dynamic monitoring and precise positioning of pipeline corrosion status can be achieved.

[0071] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A full-dimensional infrared thermal imaging monitoring method for pipeline corrosion under insulation layer, characterized by: The following steps are involved: S1: Obtain the surface temperature distribution data of the pipe under the insulation layer, divide the surface temperature distribution data into equally spaced temperature measurement points, determine the consistency of the temperature change direction corresponding to each temperature measurement point, and select temperature measurement points with consistent directions; S2: obtaining spatial position information and gradient change rate peaks of the temperature measurement points with consistent directions, identifying boundary continuous temperature measurement points where the gradient changes between adjacent temperature measurement points are continuous, and calculating boundary thermal probabilities of the boundary continuous temperature measurement points; S3: The outer wall of the pipe under the insulation layer screens abnormal response temperature measurement points whose boundary thermal probability exceeds the abnormal probability threshold, and extracts an abnormal temperature time series feature data set of the abnormal response temperature measurement points; S4: Inputting the abnormal temperature time series feature dataset into the long short-term memory network (LSTM) model to perform multi-step temperature prediction of abnormal response temperature measurement points and mark the response delay temperature measurement points; S5: Extract a measurement point cluster area on the outer wall of the pipeline below the insulation layer based on the spatial distribution of the response delayed temperature measurement points, and generate a CUI risk area division result based on the ratio of the number of response delayed temperature measurement points in the measurement point cluster area to the total number of temperature measurement points.

2. The infrared thermal imaging full-dimensional monitoring method for pipeline corrosion under insulation layer according to claim 1 is characterized in that: The specific steps for obtaining the temperature measurement points with consistent directions are: S111: Obtaining temperature distribution data on the surface of the pipe under the insulation layer, evenly dividing the data into equally spaced temperature measurement points based on spatial coordinates, establishing a corresponding temperature value recording point at each temperature measurement point, and constructing a temperature measurement point distribution array; S112: For each temperature measurement point in the temperature measurement point distribution array, calling the temperature value corresponding to the temperature measurement point and the temperature value of the adjacent temperature measurement point, calculating the direction of change of the temperature difference in each direction, and determining whether the change trends in adjacent directions are the same, marking the temperature measurement points with consistent directions as meeting the direction consistency condition, and obtaining a direction consistency identification mark set; S113: Identify a chain of temperature measurement points that are adjacent and continuous and meet direction consistency in the direction consistency identification mark set, remove discrete measurement points with discontinuous direction change trends, and obtain direction consistent temperature measurement points.

3. The infrared thermal imaging full-dimensional monitoring method for pipeline corrosion under insulation layer according to claim 2 is characterized in that: The steps for obtaining the boundary thermal probability are specifically as follows: S211: obtaining spatial position information and the corresponding gradient change rate of each temperature measurement point among the temperature measurement points with consistent directions, extracting the position where the gradient change rate reaches a local maximum value as a representative value of the gradient mutation point, and obtaining a gradient change rate peak set; S212: identifying, based on the gradient change rate peak set, adjacent direction-consistent temperature measurement points whose gradient change rates are greater than a boundary gradient threshold and whose positions are continuous, and marking the identified adjacent direction-consistent temperature measurement points as boundary-continuous temperature measurement points to generate a boundary-continuous temperature measurement point set; S213: Calculate the proportional relationship between the density of each boundary continuous temperature measurement point in the boundary continuous temperature measurement point set on the circumferential direction of the outer wall of the pipeline under the insulation layer and the peak value of the gradient change rate by a normalized weighted calculation method to obtain the boundary thermal probability of the boundary continuous temperature measurement point.

4. The infrared thermal imaging full-dimensional monitoring method for pipeline corrosion under insulation layer according to claim 3 is characterized in that: The steps for obtaining the abnormal temperature time series feature dataset are as follows: S311: comparing the boundary thermal probabilities corresponding to all continuous boundary temperature measurement points with a preset abnormal probability threshold point by point, screening temperature measurement points whose boundary thermal probabilities exceed the abnormal probability threshold, marking them as abnormal response temperature measurement points, and generating abnormal response temperature measurement point identification results; S312: Extracting a temperature value sequence of each abnormal response temperature measurement point within a continuous time frame based on the abnormal response temperature measurement point identification result, calculating a temperature change between adjacent time frames, and obtaining a total temperature rise value of the time series corresponding to the abnormal response temperature measurement point to obtain a temperature change parameter group of the abnormal response temperature measurement point; S313: Based on the temperature change parameter group of the abnormal response temperature measurement points, the temperature change amount corresponding to each abnormal response temperature measurement point, the temperature change rate of adjacent time frames and the accumulated temperature rise value are combined and archived to obtain an abnormal temperature time series feature data set.

5. The infrared thermal imaging full-dimensional monitoring method for pipeline corrosion under insulation layer according to claim 4 is characterized in that: The specific steps for obtaining the response delay temperature measurement point are: S411: Inputting the abnormal temperature time series feature dataset into a long short-term memory network (LSTM) model, calling the temperature change, temperature change rate, and cumulative temperature rise value corresponding to each abnormal response temperature measurement point, performing a multi-step temperature prediction operation, and obtaining a predicted temperature sequence of the abnormal response temperature measurement point; S412: Recall the predicted temperature sequence of the abnormal response temperature measuring point, call the actual temperature value sequence in the historical observation record that is the same as the predicted time period, calculate the corresponding difference between the predicted value of each abnormal response temperature measuring point and the actual value in the historical record according to the time step, and obtain the abnormal response temperature measuring point prediction residual sequence; S413: Count the number of time steps with continuous positive deviation or continuous negative deviation in each abnormal response temperature measurement point in the abnormal response temperature measurement point prediction residual sequence, and determine whether the time step duration exceeds the response delay determination threshold. If the condition is met, mark the corresponding abnormal response temperature measurement point as a response delay state, and obtain the response delay temperature measurement point.

6. The infrared thermal imaging full-dimensional monitoring method for pipeline corrosion under insulation layer according to claim 5 is characterized in that: The steps for obtaining the CUI risk area division results are as follows: S511: Locating the distribution area on the outer wall of the pipe below the insulation layer based on the spatial coordinate information of the response delay temperature measurement points, identifying the spatial clustering trend between adjacent response delay temperature measurement points, extracting interconnected measurement point areas in combination with a position continuity rule, and generating a clustered position group of response delay temperature measurement points; S512: Extracting spatially connected measurement point cluster regions within the outer wall of the pipe below the insulation layer based on the response delay temperature measurement point cluster position group, and calculating the ratio of the number of response delay temperature measurement points within each cluster region to the number of all temperature measurement points to obtain a response delay temperature measurement point ratio. S513: Based on the proportion of response-delay temperature measurement points and the location boundary and area information of the corresponding spatial clustering area, a spatial risk distribution structure covering the outer wall of the pipeline under the insulation layer is constructed. Areas with densely distributed response-delay temperature measurement points are identified as CUI risk areas, and the CUI risk area division results are obtained.

Images