A pipeline material fatigue risk intelligent evaluation method based on multi-source data fusion

Abstract

The application provides a pipeline material fatigue risk intelligent evaluation method based on multi-source data fusion, comprising: acquiring material composition data, collecting element composition ratio from the sample surface through a spectrum analyzer to obtain a composition vector representation, wherein each component of the vector corresponds to the percentage of the content of an element; through matching degree calculation of the environment vector and the durability value, the cosine similarity function is used to quantify the correlation strength of the two, and a matching degree score is obtained, wherein the score range is limited between zero and one; if the matching degree score is lower than a preset threshold, a fatigue-prone element subset is extracted from the composition vector to obtain a subset list, wherein the list contains element names with a proportion higher than the average value; through superimposed fusion of the defect probability distribution graph and the environment vector, a pixel-level weighted average operation is used to generate a risk heat map to obtain a heat map, wherein the brightness of the heat map corresponds to the risk intensity after fusion.

Description

Technical Field

[0001] This invention relates to the field of pipeline inspection technology, and in particular to an intelligent assessment method for pipeline material fatigue risk based on multi-source data fusion. Background Technology

[0002] Nondestructive testing (NDT) of metallic materials is a crucial field for ensuring industrial safety and product quality, bearing a vital mission. Especially in industries such as aerospace, energy, and manufacturing, the accuracy and reliability of testing directly impact the safety and stability of equipment operation. Even the smallest defect can trigger a major accident; therefore, improving testing efficiency and accuracy through technological means has become an indispensable strategic issue in this field.

[0003] However, current mainstream testing methods often reveal significant shortcomings when faced with complex working conditions. Many solutions rely heavily on the experience and judgment of testing personnel, lacking the ability to systematically predict potential risks to materials. This is especially true when dealing with large quantities and diverse materials, making it difficult to achieve comprehensive coverage and accurate location of problem areas. This limitation makes the testing process time-consuming, costly, and prone to missing critical hidden dangers, necessitating a more intelligent and forward-looking approach to overcome this shortcoming.

[0004] A deeper technological challenge lies in effectively integrating multi-source information and achieving early risk prediction. The formation of defects in metallic materials is influenced by various factors. For example, the material's composition determines its durability and corrosion resistance, and the differences in the performance of different compositions under specific operating environments further exacerbate the unpredictability of defect occurrence. This complex correlation between material properties and environmental adaptability makes it difficult to accurately determine which areas are more likely to experience problems before inspection. For instance, pipeline materials operating under high temperature and pressure conditions may preferentially develop fatigue cracks in certain areas due to compositional inhomogeneity, as they endure extreme conditions for extended periods. However, current technologies cannot identify these high-risk points before inspection, often leading to uneven allocation of inspection resources and the neglect of key areas.

[0005] Therefore, accurately identifying high-risk areas of potential defects based on material properties and service conditions before inspection has become a key issue in improving inspection efficiency and reliability. Solving this problem requires not only overcoming technological barriers but also fundamentally changing the way inspection resources are allocated to provide a more robust guarantee for industrial safety. Summary of the Invention

[0006] This invention provides an intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion, mainly including: Material composition data is obtained by collecting elemental composition ratios from the sample surface using a spectrometer, resulting in a composition vector representation where each component of the vector corresponds to the percentage content of a single element. The durability index is calculated based on the component vector. The content of each element is weighted and summed using a linear combination formula to obtain the durability value. The weights are based on the contribution rate of the corresponding elements in a preset material database. The service environment parameters are obtained by monitoring temperature, pressure and corrosive medium concentration outside the pipeline through a sensor array, and an environmental vector representation is obtained, in which the vector components are temperature value, pressure value and concentration value in turn. The matching degree between the environment vector and the durability value is calculated, and the cosine similarity function is used to quantify the correlation strength between the two to obtain the matching degree score, where the score range is limited to between zero and one. If the matching score is lower than the preset threshold, a subset of easily fatigued elements is extracted from the component vector to obtain a subset list, which contains the names of elements with a proportion higher than the average. A defect formation model is constructed based on a subset list. A convolutional neural network is used to process the historical defect image dataset to train the model parameters. The model output is a defect probability distribution map, where the pixel values ​​of the distribution map represent the local occurrence probability. By superimposing and fusing the defect probability distribution map with the environmental vector, a risk heat map is generated using a pixel-level weighted average operation. The brightness of the heat map corresponds to the risk intensity after fusion. To identify connected regions in the heatmap whose brightness exceeds a threshold, a region growing algorithm is used to expand the boundary from the seed point to determine the coordinate set of the high-risk area boundary, where the coordinate set consists of continuous pixel positions.

[0007] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an intelligent assessment method for pipeline material fatigue risk based on multi-source data fusion. This method addresses the unique business scenario where pipelines are prone to localized fatigue defects due to mismatches between material composition and environmental parameters under complex service environments. It deeply correlates the component vector obtained from material spectral analysis with environmental vectors collected in real-time by sensors, quickly determining the matching degree through cosine similarity. If the matching degree is below a threshold, a subset of fatigue-prone elements is precisely extracted. A defect formation model trained by a convolutional neural network is then used to generate a defect probability distribution map, which is further fused with the environmental vector at the pixel level to form a risk heatmap. Finally, a region growing algorithm is used to delineate the boundaries of high-risk areas, achieving visualized intelligent early warning from microscopic material composition to macroscopic risk, significantly improving the prediction accuracy and prevention efficiency of pipeline fatigue failure. Attached Figure Description

[0008] Figure 1This is a flowchart of an intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion, according to the present invention.

[0009] Figure 2 This is a schematic diagram of an intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to the present invention.

[0010] Figure 3 This is another schematic diagram of an intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0012] like Figures 1-3 This embodiment of a method for intelligent assessment of pipeline material fatigue risk based on multi-source data fusion may specifically include: S101. Obtain material composition data. Collect the elemental composition ratio from the sample surface using a spectrometer to obtain a composition vector representation, where each component of the vector corresponds to the percentage content of one element.

[0013] The sample surface is scanned using a spectroscopic instrument to acquire raw elemental composition signal data, obtaining preliminary elemental distribution information. Based on this preliminary information, signal processing techniques are used to denoise the acquired signal data, resulting in a clear elemental composition signal. For this clear signal, the relative intensity value of each element is calculated to determine its percentage content. Using this percentage data, a corresponding component proportion vector is constructed to obtain the sample's component vector representation. If the percentage content of certain elements in the component vector representation is below a preset threshold, secondary signal enhancement processing is performed on that portion of the data to determine whether re-acquiring of the relevant elemental signals is necessary. Based on the data after secondary signal enhancement, the component proportion vector is updated to determine the final elemental composition representation of the sample. Finally, a support vector machine algorithm is used to classify the component vector, obtaining the material category information of the sample.

[0014] Specifically, in the process of acquiring material composition data, the sample surface is first scanned by a spectrometer to collect the elemental composition ratio data. Assuming that an X-ray fluorescence spectrometer (XRF) is used, its working principle is to use X-rays to excite the atoms on the sample surface to generate characteristic fluorescence signals, and the energy peaks of different elements are recorded by a detector.

[0015] For example, when scanning a metal sample, the instrument detected an energy peak of 6.4 keV and an intensity of 5000 counts for iron (Fe), 8.0 keV and an intensity of 2000 counts for copper (Cu), and 8.6 keV and an intensity of 1000 counts for zinc (Zn). The system then converts these intensity data into content percentages using a built-in calibration curve. The calculation formula is: Content Percentage = (Element Intensity / Total Intensity) × Calibration Coefficient, where the calibration coefficient is determined by the standard sample. Assuming the coefficient is 1.2 for iron, 1.1 for copper, and 1.0 for zinc, with a total intensity of 8000 counts, the iron content is (5000 / 8000) × 1.2 = 75%, copper is (2000 / 8000) × 1.1 = 27.5%, and zinc is (1000 / 8000) × 1.0 = 12.5%. Since the sum exceeds 100%, the system automatically normalizes the data, resulting in an iron content of approximately 68.2%, a copper content of 25.0%, and a zinc content of 11.4%. Next, these percentage data are organized into a component vector representation, in the form [68.2, 25.0, 11.4], corresponding to the content proportions of iron, copper, and zinc, respectively. The system further stores this vector in a database and compares it with a known materials database using a clustering algorithm (such as K-means). It calculates the Euclidean distance to find the nearest material category. Assuming a minimum distance of 0.5 corresponds to low-carbon steel, the system determines that the sample likely belongs to this category. Finally, the system generates an analysis report, including elemental proportions, vector representations, and category inference results, which is automatically uploaded to a cloud platform for subsequent process optimization or quality control, forming a complete closed loop from data acquisition to result output.

[0016] S102. Calculate the durability index based on the component vector, and use a linear combination formula to weight and sum the contents of each element to obtain the durability value, where the weights are based on the contribution rate of the corresponding elements in the preset material database.

[0017] Based on the percentage content of each element in the component vector, the durability contribution weight coefficients of the corresponding elements are extracted from a pre-established material durability database. A linear weighted summation formula is used to calculate the basic durability score: D = Σ(w_i × c_i), where w_i is the element's durability contribution weight coefficient and c_i is the element's percentage content. Based on this basic durability score, the environmental type of the sample is matched, and the corresponding coefficient is obtained from the environmental correction coefficient table. The environmentally corrected durability value is obtained by multiplying the basic durability score by the environmental correction coefficient. The cosine similarity between the component vector and historical reference samples is calculated, and the top-ranked reference samples are selected. A nearest neighbor regression model is used, with cosine similarity as the weight, to perform a weighted average of the actual lifespan records of the reference samples, resulting in a lifespan prediction adjustment coefficient. The final durability index is obtained by multiplying the environmentally corrected durability value by the lifespan prediction adjustment coefficient. Specifically, in calculating the material durability index, the system first extracts the stored component vector data from the database. Assuming the vector for a metal sample is [60.5, 30.2, 9.3], representing the percentage content of nickel, chromium, and molybdenum, respectively. Next, the system calls the preset material database to obtain the contribution weight of each element to durability. Assuming the weight of nickel is 0.4, chromium is 0.35, and molybdenum is 0.25, these weights are based on historical experimental data and the analysis of the elements' roles in corrosion and wear resistance. Subsequently, the system uses a linear combination formula to calculate the durability value: Durability Index = Nickel Content × Nickel Weight + Chromium Content × Chromium Weight + Molybdenum Content × Molybdenum Weight. Substituting the data, the calculation is 60.5 × 0.4 + 30.2 × 0.35 + 9.3 × 0.25 = 24.2 + 10.57 + 2.325 = 37.095. Next, the system compares the value with the durability rating standards in the database. Assuming the standard defines a durability index between 30 and 40 as a medium durability level, the system determines that the sample belongs to the medium durability level and stores the result in the analysis module. Simultaneously, the system automatically connects to the materials application scenario database to analyze whether the durability level meets the requirements of specific industrial applications.

[0018] For example, if used in a high-temperature environment, the system will further extract relevant environmental parameters and calculate the durability correction value. Assuming the correction coefficient is 0.9, the corrected durability is 37.095 × 0.9 = 33.3855, which is still within the medium durability range. Finally, a comprehensive report containing durability indicators, grade determination, and application scenario adaptability is generated and automatically saved to the cloud system for subsequent material screening, forming a complete logical chain from component analysis to application evaluation.

[0019] S103. Obtain service environment parameters. Monitor temperature, pressure and corrosive medium concentration outside the pipeline through a sensor array to obtain an environmental vector representation, where the vector components are temperature value, pressure value and concentration value in sequence.

[0020] A sensor array continuously collects data from the outside of the pipeline, acquiring temperature, pressure, and concentration values ​​in the service environment to construct an initial environmental vector. Based on this initial environmental vector, a pre-established environmental classification standard is used to divide the vector components into intervals to determine the service environment category. If the service environment category belongs to a high-risk interval, the corresponding alarm parameters are retrieved from a preset threshold table to determine the abnormal state of the environmental vector. For environmental vectors in abnormal states, historical data records are compared to obtain historical environmental vectors with similar components, leading to processing strategies for similar environments. Based on the processing strategies for similar environments, a temporary adjustment plan is generated for the current service environment, and it is determined whether to trigger the protection mechanism. The execution results of the temporary adjustment plan are used to update the dynamic records of environmental monitoring, determining the focus of subsequent data acquisition. If the focus of subsequent data acquisition changes, the monitoring frequency of the sensor array is adjusted to obtain more accurate environmental vector data.

[0021] Specifically, in acquiring pipeline service environment parameters, the system collects key data in real time through a sensor array deployed outside the pipeline, forming an environmental vector representation. First, the system uses a high-precision temperature sensor to monitor the external ambient temperature of the pipeline, assuming the collected temperature value is 85.3 degrees Celsius. Simultaneously, a pressure sensor records the pressure data outside the pipeline, assuming a measured value of 2.7 MPa. Furthermore, a chemical sensor detects the concentration of surrounding corrosive media, assuming a hydrogen sulfide concentration of 0.12 mg / m³. This data is integrated into an environmental vector [85.3, 2.7, 0.12] and automatically uploaded to the environmental analysis module. Next, the system invokes a preset environmental impact assessment algorithm to compare each component of the vector with standard thresholds. Assuming a temperature threshold of 80 degrees Celsius, a pressure threshold of 3.0 MPa, and a concentration threshold of 0.15 mg / m³, the system calculates the deviations (temperature deviation = 85.3 - 80 = 5.3, pressure deviation = 3.0 - 2.7 = 0.3, concentration deviation = 0.15 - 0.12 = 0.03). The system further uses a weighted summation formula to assess the environmental severity, with weights of 0.5, 0.3, and 0.2. The calculation result is 5.3 × 0.5 + 0.3 × 0.3 + 0.03 × 0.2 = 2.65 + 0.09 + 0.006 = 2.746. The system compares this value with environmental level standards in the database. Assuming a severity level between 2.0 and 3.0 indicates a moderate risk environment, the system determines the current environment to be of moderate risk. Subsequently, the system automatically connects to the pipeline material database, extracts the potential corrosion rate model of the material in this environment, assumes that the model predicts a corrosion rate increase factor of 1.15, stores the results in the risk assessment module, generates an environmental parameter and material compatibility analysis report, uploads it to the cloud for backup, and forms a complete logical chain from environmental monitoring to risk early warning.

[0022] S104. By calculating the matching degree between the environment vector and the durability value, the cosine similarity function is used to quantify the correlation strength between the two to obtain the matching degree score, where the score range is limited to between zero and one.

[0023] Initial data collection of environmental vectors and durability values ​​is conducted by retrieving corresponding vector data and numerical records from a pre-established data repository to determine the basic comparison dataset. For this dataset, a cosine similarity function is used to calculate the association strength between environmental vectors and durability values, yielding a matching score. Based on the matching score, if the score is below a preset threshold, the environmental vector data is stratified to determine if outliers exist. By identifying outliers, similar environmental vector fragments are retrieved from historical records to pinpoint potential sources of bias. For these potential sources, a data correction tool is used to preprocess the environmental vectors, resulting in a corrected vector dataset. Based on this corrected dataset, the association strength with durability values ​​is calculated again using the cosine similarity function to determine the final matching score. If the final matching score is still below a preset threshold, the result is recorded in the dynamic monitoring log for future analysis.

[0024] Specifically, in the durability assessment of the pipeline service environment, the system constructs a matching degree analysis between environmental vectors and durability values, and uses a cosine similarity function to quantify the correlation strength between the two to generate a matching degree score, with the score range limited to 0 to 1. The specific implementation method is as follows: the system first extracts the current environmental vector from the database, assumed to be [72.5, 1.8, 0.08], representing a temperature of 72.5 degrees Celsius, a pressure of 1.8 MPa, and a corrosive medium concentration of 0.08 mg / m³, respectively. At the same time, it obtains the corresponding pipeline material durability reference vector, assumed to be [75.0, 2.0, 0.1]. Next, the system calls the cosine similarity algorithm to calculate the cosine of the angle between the two vectors. The formula is: Similarity = (72.5 × 75.0 + 1.8 × 2.0 + 0.08 × 0.1) / (sqrt(72.5^2 + 1.8^2 + 0.08^2) × sqrt(75.0^2 + 2.0^2 + 0.1^2)). The numerator is calculated to be 5437.5 + 3.6 + 0.008 = 5441.108, and the denominator is sqrt(5256.25 + 3.24 + 0.0064) × sqrt(5625 + 4 + 0.01) ≈ 72.52 × 75.03 ≈ 5441.175. Therefore, the similarity is approximately 5441.108 / 5441.175 ≈ The score is 0.999. The system compares this score with a preset threshold of 0.95, determining a very high match, indicating that the current environment has little impact on material durability. Subsequently, the system automatically stores this match score in the analysis module and associates it with the material fatigue life prediction model. Assuming the prediction result shows a life decay coefficient of 0.98, it indicates good environmental adaptability. Finally, the system generates a match analysis report and automatically uploads it to the cloud database, forming a complete logical chain from environmental vector calculation to durability assessment.

[0025] S105. If the matching score is lower than the preset threshold, extract the fatigue-prone element subset from the component vector to obtain a subset list, where the list contains the names of elements with a proportion higher than the average.

[0026] Based on the comparison between the secondary matching score and a preset threshold, it is determined whether to enter the high-fluctuation element separation process. The fluctuation amplitude of each element component is extracted from the corrected component vector to obtain a fluctuation amplitude dataset. By comparing the fluctuation amplitude dataset with a preset fluctuation amplitude standard, element components exceeding the standard are identified. These element components are clustered to form a high-fluctuation element set. A vector retrieval method is used to match the high-fluctuation element set in the environmental interference library to obtain a list of similar interference patterns. The item with the highest matching degree is selected from the list of similar interference patterns to determine the dominant interference pattern. The high-fluctuation element set is then reverse-corrected using the compensation coefficient corresponding to the dominant interference pattern to obtain a compensated element sub-vector. The compensated element sub-vector is combined with the remaining stable components using a vector concatenation method to obtain a stable component vector. Specifically, in the pipeline material durability assessment system, when the matching score calculated by cosine similarity is lower than the preset threshold of 0.90, the system immediately triggers the composition sensitivity compensation process. First, it retrieves the complete composition vector of the current pipeline steel grade from the material database, for example, [Fe: 96.82%, C: 0.18%, Mn: 1.25%, Si: 0.35%, Cr: 0.42%, Ni: 0.68%, Mo: 0.22%, V: 0.08%]. The system automatically calculates the global average value of each element content and marks the deviation. Elements whose actual content is more than 1.2 times higher than the average value are defined as fatigue-sensitive elements. A subset list [Cr: 0.42%, Ni: 0.68%, Mo: 0.22%] is extracted through a traversal algorithm. The system then constructs a weighted sensitivity coefficient vector [0.42×2.8, 0.68×3.5, 0.22×4.1] for this subset, where the sensitivity coefficients of each element are derived from the corrosion fatigue test database, resulting in a sensitivity intensity vector [1.176, 2.380, 0.902]. The system further calculates the Euclidean distance between this vector and the current environmental vector [68.0, 2.3, 0.15], obtaining 3.74. Using the sigmoid normalization function f(x) = 1 / (1 + e^(-2(x-3.5))), the system calculates the environment-sensitive element coupling risk factor of 0.89. This risk factor is then multiplied and fused with the original matching score of 0.87 to obtain a comprehensive durability correction coefficient of 0.776. This coefficient is automatically written into the correction term of the real-time life prediction model, ultimately generating a compensation analysis record containing the fatigue-prone element subset, the coupling risk factor, and the correction coefficient. This record is then pushed to the maintenance decision module to complete the fully automated risk quantification closed loop.

[0027] S106. Construct a defect formation model based on the subset list, use a convolutional neural network to process the historical defect image dataset to train the model parameters, and obtain the model output as a defect probability distribution map, where the pixel value of the distribution map represents the local occurrence probability.

[0028] The system extracts key element combinations related to defect prediction from a subset list, classifies and labels each element within the combination, resulting in a classified element set. Using this classified element set, a convolutional neural network is used to extract features from historical image data, yielding a feature map image. Based on the feature map image, high-density regions are segmented to determine a set of segmented region blocks. If the pixel density of a block in the region block set exceeds a preset threshold, a depth scan is performed on that block to obtain local defect distribution information. Based on this local defect distribution information, probability calculations are performed on each region block to obtain a defect occurrence probability value. High-probability regions are prioritized according to their defect occurrence probability values ​​to determine a sorted region sequence. Preceding regions are extracted from the sorted region sequence, and combined with relevant historical data, a dynamic monitoring scheme for defect prediction is constructed.

[0029] Specifically, in the pipeline material defect assessment system, the system first automatically constructs a defect formation model based on a subset list obtained from the material analysis module, such as a set containing specific elements [Cu: 0.35%, Ti: 0.12%, Nb: 0.09%]. It then calls upon a historical defect image dataset stored in the cloud, containing over 5000 high-resolution images of pipeline surface defects, each with a resolution of 1024x1024 pixels. Next, the system trains the model using a convolutional neural network algorithm. The network structure consists of 5 convolutional layers and 3 fully connected layers, with a 3x3 kernel size, a stride of 1, a ReLU activation function, and the Adam algorithm as the optimizer. The learning rate is 0.001, the batch size is 32, and the training run is 50 epochs. The system automatically adjusts the model parameters by extracting features from the image dataset. During training, cross-entropy loss is used as the loss function, and training stops once the validation set accuracy reaches 0.92. After training, the system inputs test images into the model and outputs a defect probability distribution map. Each pixel value in the distribution map ranges from 0 to 1, representing the probability of a local defect. For example, a pixel value of 0.85 in the center of a certain area indicates a high probability of defect occurrence in that area. The system then performs post-processing on the distribution map, using a Gaussian smoothing algorithm (with a standard deviation of 2.5) to reduce noise and extracting areas with probability values ​​higher than 0.7 as high-risk defect points, generating a coordinate list, such as [(x:256, y:312), (x:458, y:129)]. These coordinates are then mapped to the pipeline geometry model, automatically associating them with specific pipeline segment numbers, such as segment P-2023-017, establishing a correspondence between defect locations and pipeline structures. Finally, the system stores the probability distribution map and high-risk point coordinate data in a defect database and pushes it to the pipeline health monitoring module for subsequent defect trend analysis and early warning mechanisms, constructing a complete logical chain from image processing to location mapping to ensure the accuracy and automation of defect assessment.

[0030] S107. By superimposing and fusing the defect probability distribution map with the environmental vector, a risk heat map is generated using a pixel-level weighted average operation. The brightness of the heat map corresponds to the risk intensity after fusion.

[0031] High-brightness pixel sets are extracted from the risk heatmap to identify high-risk pixel regions. For each high-risk pixel region, the pixel brightness gradient amplitude is calculated to obtain a gradient distribution map. The peak position of the gradient amplitude is detected using the gradient distribution map to determine the risk boundary contour. The heatmap is then segmented using the risk boundary contour to obtain a set of independent risk sub-regions. The mean pixel brightness of each sub-region is calculated to determine the risk level sequence of the sub-regions. If the mean brightness of a sub-region in the risk level sequence exceeds a specified multiple of the overall mean brightness of the heatmap, this sub-region is marked as a key monitoring area, and a set of coordinates for the key monitoring areas is obtained. The pixel values ​​of the key monitoring area coordinates are overlaid on the original defect probability distribution map to obtain an enhanced risk heatmap.

[0032] Specifically, in the pipeline risk assessment system, the system first acquires a defect probability distribution map. This map has a resolution of 512x512 pixels, with each pixel value ranging from 0 to 1, representing the probability of a local defect. For example, a pixel value of 0.75 is found in a certain area. Simultaneously, the system extracts environmental vector data from the environmental monitoring module, including parameters such as temperature, humidity, and corrosion factor. For instance, in the temperature distribution map, a certain area has a value of 35.2 degrees Celsius, humidity of 78.3%, and a corrosion factor of 0.62. This data is also stored in a 512x512 pixel matrix. Next, the system normalizes the defect probability distribution map and environmental vectors, mapping the environmental parameter values ​​to a range of 0 to 1. For example, the temperature of 35.2 degrees Celsius is linearly normalized to 0.58. Subsequently, the system employs a pixel-level weighted average algorithm for fusion, setting the defect probability weight to 0.6 and the environment vector weight to 0.4. The calculation formula is: Fusion value = 0.6 * Defect probability + 0.4 * Environment parameter value. For example, if a pixel has a defect probability of 0.75 and an environment parameter value of 0.58, the fused value is 0.6 * 0.75 + 0.4 * 0.58 = 0.682. A risk heatmap is generated after fusion, maintaining a resolution of 512x512 pixels. The heatmap brightness is positively correlated with the fusion value; for example, a fusion value of 0.682 corresponds to a brightness of 68.2% (brightness range 0 to 100%). Finally, the system performs a hierarchical analysis of the heat map, classifying the brightness values ​​into three risk levels: a brightness value greater than 70% is considered high risk, 50% to 70% is considered medium risk, and less than 50% is considered low risk. The system also generates a risk area statistical report, for example, showing that high-risk areas account for 12.5% ​​and medium-risk areas account for 28.7%. The data is automatically transmitted to the pipeline maintenance and scheduling module for subsequent priority ranking of risk areas, forming a complete logical chain from data fusion to risk classification.

[0033] S108. Determine the connected regions in the heatmap whose brightness exceeds the threshold, and use the region growing algorithm to expand the boundary from the seed point to determine the coordinate set of the high-risk area boundary, where the coordinate set consists of continuous pixel positions.

[0034] Based on the brightness distribution of the heatmap, an initial seed point set is determined. A region growing algorithm is used to expand the seed point set outwards in an eight-neighbor direction. It is determined whether the brightness of adjacent pixels exceeds the same threshold; if so, they are included in the current connected region, resulting in multiple independent high-risk connected regions. For each high-risk connected region, boundary tracking is performed to obtain the closed boundary coordinate sequence, forming a high-risk area boundary set. A masking operation is performed on the heatmap using the boundary coordinate set to extract pixel sub-images within each high-risk connected region. A statistical histogram of pixel brightness within each high-risk connected region is calculated to determine the brightness peak position. Based on the offset between the brightness peak position and the region center coordinates, the location of the core area of ​​the risk concentration is determined, obtaining a core area coordinate subset. Local magnification processing is performed on the original heatmap using the core area coordinate subset to generate a high-resolution risk detail map. Specifically, in the pipeline risk assessment system, the system sets a high-risk threshold of 0.75 for the generated 512x512 pixel risk heatmap, meaning that pixels with a brightness exceeding 75% are considered potentially high-risk points. First, a global scan identifies all points with pixel values ​​≥ 0.75 as the initial seed point set. For example, a total of 127 seed points are found, such as a pixel value of 0.83 at coordinates (187, 312) and a pixel value of 0.79 at coordinates (245, 178). Subsequently, an 8-connected region growth algorithm was used to expand each seed point. The growth rule was that adjacent pixel values ​​≥ 0.70 could be merged into the same connected region. For example, starting from the seed point (187,312), the growth was carried out in eight directions: up, down, left, right and diagonal. When the pixel (188,313) with a value of 0.72 was encountered, it was included in the boundary. The growth continued until the pixel (190,315) with a value of 0.68 was reached and then stopped. Finally, an irregular high-risk connected region containing 483 pixels was formed. The boundary coordinate set was recorded as a clockwise closed polygon vertex sequence [(175,300), (202,305), (208,328), (189,340), (170,333)], etc., 15 vertices. After growing all 127 seed points, the system obtained 42 independent high-risk connected regions. The largest connected region contained 1267 pixels, with an approximately elliptical shape, a major axis of about 38 pixels, a minor axis of about 26 pixels, and a center coordinate of (256, 289). Finally, the system calculated the centroid, circumscribed rectangle, perimeter, and area of ​​each connected region. For example, the centroid of the largest connected region was (255.7, 290.3), and the range of the circumscribed rectangle was x∈[228, 283], y∈[271, 311]. The system automatically wrote the boundary coordinates and area data of all high-risk regions into the GIS vector layer, which can be directly called by the subsequent UAV precision inspection path planning module, realizing a complete automated process from heat map to precise spatial positioning of high-risk areas.

[0035] The above description of the embodiments is only for the purpose of helping to understand the technical solutions and core ideas of this application; those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent assessment of fatigue risk of pipeline materials based on multi-source data fusion, characterized in that, The method includes: Material composition data is obtained by collecting elemental composition ratios from the sample surface using a spectrometer, resulting in a composition vector representation where each component of the vector corresponds to the percentage content of a single element. The durability index is calculated based on the component vector. The content of each element is weighted and summed using a linear combination formula to obtain the durability value. The weights are based on the contribution rate of the corresponding elements in a preset material database. The service environment parameters are obtained by monitoring temperature, pressure and corrosive medium concentration outside the pipeline through a sensor array, and an environmental vector representation is obtained, in which the vector components are temperature value, pressure value and concentration value in turn. The matching degree between the environment vector and the durability value is calculated, and the cosine similarity function is used to quantify the correlation strength between the two to obtain the matching degree score, where the score range is limited to between zero and one. If the matching score is lower than the preset threshold, a subset of easily fatigued elements is extracted from the component vector to obtain a subset list, which contains the names of elements with a proportion higher than the average. A defect formation model is constructed based on a subset list. A convolutional neural network is used to process the historical defect image dataset to train the model parameters. The model output is a defect probability distribution map, where the pixel values ​​of the distribution map represent the local occurrence probability. By superimposing and fusing the defect probability distribution map with the environmental vector, a risk heat map is generated using a pixel-level weighted average operation. The brightness of the heat map corresponds to the risk intensity after fusion. To identify connected regions in the heatmap whose brightness exceeds a threshold, a region growing algorithm is used to expand the boundary from the seed point to determine the coordinate set of the high-risk area boundary, where the coordinate set consists of continuous pixel positions.

2. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, The material composition data is obtained by collecting elemental composition ratios from the sample surface using a spectrometer, resulting in a composition vector representation. Each component of the vector corresponds to the percentage content of one element, including: The sample surface is scanned by a spectroscopic analysis instrument to obtain the original elemental composition signal data and obtain preliminary elemental distribution information; Based on the preliminary element distribution information, signal processing techniques are used to denoise the collected signal data to obtain a clear element composition signal. For a clear elemental composition signal, calculate the relative intensity value of each element and determine the percentage content data of each element; By using the percentage content data, construct the corresponding component ratio vector to obtain the component vector representation of the sample; If the percentage of certain elements in the component vector representation is lower than a preset threshold, then secondary signal enhancement processing is performed on that part of the data to determine whether the relevant element signals need to be re-acquired. Based on the data after secondary signal enhancement processing, the component ratio vector is updated to determine the final elemental composition representation of the sample. For the final elemental composition representation of the sample, a support vector machine algorithm is used to classify the component vectors to obtain the material category information of the sample.

3. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, The durability index is calculated based on the component vector by using a linear combination formula to weighted summation of the content of each element to obtain the durability value. The weights are based on the contribution rate of the corresponding element in a pre-defined material database, including: Based on the percentage content of each element in the component vector, the durability contribution weight coefficients of the corresponding elements are extracted from a pre-established material durability database. A linear weighted summation formula is used to calculate the basic durability score: D=Σ(w_i×c_i), where w_i is the element's durability contribution weight coefficient and c_i is the element's percentage content. Based on this basic durability score, the environmental type of the sample is matched, and the corresponding coefficient is obtained from the environmental correction coefficient table. The environmentally corrected durability value is obtained by multiplying the basic durability score by the environmental correction coefficient. The cosine similarity between the component vector and historical reference samples is calculated, and the top-ranked reference samples are selected. A nearest neighbor regression model is used, with cosine similarity as the weight, to perform a weighted average of the actual lifespan records of the reference samples, resulting in a lifespan prediction adjustment coefficient. The final durability index is obtained by multiplying the environmentally corrected durability value by the lifespan prediction adjustment coefficient.

4. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, The process of acquiring service environment parameters involves monitoring temperature, pressure, and corrosive medium concentration outside the pipeline using a sensor array to obtain an environmental vector representation. The vector components are, in order, temperature, pressure, and concentration values, including: By continuously collecting data outside the pipeline using a sensor array, temperature, pressure, and concentration values ​​in the service environment are obtained, and an initial environmental vector is constructed. Based on the initial environment vector, the vector components are divided into intervals using a pre-established environment classification standard to determine the category of the service environment; If the service environment category is in the high-risk range, the corresponding alarm parameters are obtained from the preset threshold table to determine the abnormal state of the environment vector. For environmental vectors in abnormal states, historical data records are compared to obtain historical environmental vectors that are similar to the current vector components, and processing strategies for similar environments are obtained. Based on the processing strategies under similar environments, generate a temporary adjustment plan for the current service environment and determine whether the protection mechanism is triggered; By temporarily adjusting the implementation results of the plan, updating the dynamic records of environmental monitoring, and determining the key focus areas for subsequent data collection; If the focus of subsequent data acquisition changes, the monitoring frequency of the sensor array will be adjusted to obtain more accurate environmental vector data.

5. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, The matching degree calculation between the environmental vector and the durability value is performed, and the cosine similarity function is used to quantify the correlation strength between the two to obtain a matching degree score. The score range is limited to between zero and one, including: By collecting initial data on environmental vectors and durability values, the corresponding vector data and numerical records are obtained from a pre-established data repository to determine the basic comparison dataset; For the basic comparison dataset, the cosine similarity function is used to calculate the association strength between the environment vector and the durability value to obtain the matching score result; Based on the matching score, if the score is lower than the preset threshold range, the environmental vector data will be processed in layers to determine whether there are any data anomalies. By identifying data anomalies, environmental vector fragments similar to the anomalies are obtained from historical records to determine potential sources of deviation. To identify potential sources of bias, a data correction tool is used to preprocess the environmental vectors, resulting in a corrected vector dataset. Based on the corrected vector dataset, the correlation strength with the durability value is calculated again using the cosine similarity function to determine the final matching score; If the final matching score is still lower than the preset threshold, the result will be recorded in the dynamic monitoring log to obtain a reference for subsequent analysis.

6. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, If the matching score is lower than a preset threshold, a subset of easily fatigued elements is extracted from the component vector to obtain a subset list, which contains the names of elements with a proportion higher than the average, including: Based on the comparison between the secondary matching score and the preset threshold, it is determined whether to enter the high-fluctuation element separation process. The fluctuation amplitude of each element component is extracted from the modified component vector to obtain the fluctuation amplitude dataset. By comparing the fluctuation amplitude dataset with the preset fluctuation amplitude standard, the element components that exceed the standard are identified. The element components that exceed the standard are clustered to form a high-fluctuation element set. The high-fluctuation element set is matched in the environmental interference library using a vector retrieval method to obtain a list of similar interference patterns. The item with the highest matching degree is selected from the list of similar interference patterns to determine the dominant interference pattern. The high-fluctuation element set is reverse-corrected using the compensation coefficient corresponding to the dominant interference pattern to obtain the compensated element sub-vector. The compensated element sub-vector is combined with the remaining stable components using a vector concatenation method to obtain the stable component vector.

7. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, The defect formation model is constructed based on a subset list. A convolutional neural network is used to process the historical defect image dataset to train the model parameters, resulting in a defect probability distribution map as the model output. The pixel values ​​in the distribution map represent the local probability of occurrence, including: Extract key element combinations related to defect prediction from the subset list, classify and label each element in the combination, and obtain the classified element set. Using the classified set of elements, a convolutional neural network is used to extract features from image data in historical records to obtain a feature mapping image; Based on the feature map image, high-density regions are segmented to determine the set of segmented region blocks; If the pixel density of a certain block in the set of region blocks is higher than a preset threshold, then a depth scan is performed on that region block to obtain the distribution information of local defects; By using the distribution information of local defects, the probability of defect occurrence in each region block is calculated to obtain the probability value of defect occurrence in each region block; Based on the probability values ​​of defect occurrence, high-probability areas are prioritized to determine the sorted area sequence. Extracting preceding regions from the sorted region sequence and combining them with relevant data from historical records, a dynamic monitoring scheme for defect prediction is constructed.

8. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, The process involves overlaying and fusing the defect probability distribution map with the environmental vector, and then generating a risk heatmap using a pixel-level weighted averaging operation. The resulting heatmap's brightness corresponds to the risk intensity after fusion, and includes: High-brightness pixel sets are extracted from the risk heatmap to determine high-risk pixel regions; Calculate the pixel brightness gradient magnitude for high-risk pixel areas and obtain the gradient distribution map; The location of the gradient amplitude peak is detected by gradient distribution map to determine the risk boundary outline; The heatmap is segmented using risk boundary contours to obtain a set of independent risk sub-regions. For each set of independent risk sub-regions, the average pixel brightness of each sub-region is calculated to determine the risk level sequence of the sub-regions; If the average brightness of a certain sub-region in the risk level sequence exceeds a specified multiple of the average brightness of the entire heat map, then the sub-region is marked as a key monitoring area, and the coordinate set of the key monitoring area is obtained. An enhanced risk heat map is obtained by overlaying pixel values ​​of the key monitoring area coordinate set and the original defect probability distribution map.

9. The intelligent assessment method for fatigue risk of pipeline materials based on multi-source data fusion according to claim 1, characterized in that, The process of identifying connected regions in the heatmap whose brightness exceeds a threshold involves using a region growing algorithm to expand the boundary from the seed point and determine the coordinate set of the high-risk area boundary. This coordinate set consists of continuous pixel positions, including: Based on the brightness distribution of the heatmap, an initial seed point set is determined. A region growing algorithm is used to expand the seed point set towards the eight-neighbor direction. It is determined whether the brightness of adjacent pixels exceeds the same threshold. If it does, it is included in the current connected region, resulting in multiple independent high-risk connected regions. For each high-risk connected region, boundary tracking is performed to obtain the closed boundary coordinate sequence, forming a high-risk area boundary set. A mask operation is performed on the heatmap using the boundary coordinate set to extract the pixel sub-images inside each high-risk connected region. The brightness statistical histogram of pixels inside each high-risk connected region is calculated to determine the brightness peak position. Based on the offset between the brightness peak position and the coordinates of the region center, the position of the core area of ​​the risk concentration is determined, and a core area coordinate subset is obtained. For the core area coordinate subset, local magnification processing is performed on the original heatmap to generate a high-resolution risk detail map.

Images