An AI image recognition-based intelligent nondestructive testing method and system for carbon fiber wires

Abstract

The application relates to an AI image recognition-based intelligent nondestructive testing method and system for carbon fiber wires. The method comprises the following steps: acquiring nondestructive testing image data of the carbon fiber wire, and extracting micro feature data based on the data; based on the micro feature data, defect recognition is performed through a preset carbon fiber wire defect-feature database to obtain a wire defect recognition result; according to the recognition result, potential defect correlation prediction is performed based on the characteristics of the carbon fiber wire to obtain a potential defect prediction result; based on the prediction result, defect verification is performed by using a pre-trained machine learning model to obtain a potential defect verification result; and the recognition result, the prediction result and the verification result are integrated to generate a carbon fiber wire nondestructive testing report. The method can realize intelligent nondestructive testing of the carbon fiber wire, improve the accuracy of defect recognition and the reliability of potential risk prediction, and guarantee the safe operation of the wire.

Description

Technical Field

[0001] This invention belongs to the field of nondestructive testing, and in particular relates to an intelligent nondestructive testing method and system for carbon fiber conductors based on AI image recognition. Background Technology

[0002] With the rapid development of technologies in high-end equipment fields such as power transmission and aerospace, carbon fiber conductors, with their core characteristics of high strength, low weight, corrosion resistance, high conductivity, and excellent fatigue resistance, have become a key material to replace traditional metal conductors. To avoid safety accidents such as breakage and conductivity failure of carbon fiber conductors due to defects, non-destructive testing (NDT) technology has received widespread attention as a core means to ensure their operational reliability. This technology can detect internal and surface defects in the material without damaging the structural integrity of the conductor, and is an important component of the engineering safety assurance system.

[0003] Currently, non-destructive testing of carbon fiber conductors mainly relies on traditional testing methods, including visual inspection, ultrasonic testing, radiographic testing, and eddy current testing. Visual inspection involves inspectors using simple tools such as magnifying glasses and endoscopes to observe the surface condition of the conductor and, based on experience, determine the presence of defects such as surface damage and exposed fibers. Ultrasonic testing utilizes the propagation and reflection characteristics of ultrasonic waves within the carbon fiber conductor, analyzing echo signals to identify internal defects such as porosity and delamination. Radiographic testing uses X-rays and gamma rays to penetrate the conductor, determining the location and approximate shape of defects based on differences in image grayscale. Eddy current testing utilizes the principle of electromagnetic induction, detecting changes in eddy currents to identify areas of abnormal conductivity on and near the surface of the conductor, indirectly determining the extent of defects.

[0004] However, existing traditional testing methods have significant technical shortcomings: limited testing accuracy, manual visual inspection cannot identify microscopic defects (such as microfiber fractures, nanoscale pore anomalies, etc.), and is highly subjective, easily affected by personnel experience and fatigue, leading to a high rate of missed and false detections; ultrasonic testing and radiographic testing are insufficient in identifying the microscopic features of carbon fiber conductors (such as fiber alignment direction, texture density, etc.), making it difficult to achieve accurate quantitative analysis of defects; low testing efficiency, traditional testing methods rely heavily on manual operation, the testing process is cumbersome, and cannot adapt to the needs of large-scale, continuous conductor testing; lack of potential defect prediction capability, existing methods can only identify existing explicit defects, failing to establish defect correlations based on the material properties of carbon fiber conductors, and cannot predict the damage propagation trend and potential failure risk of defects; insufficient defect information integration, traditional testing results are mostly single-dimensional defect identifications, failing to achieve a comprehensive presentation of defect type, location, propagation trend, and failure probability, making it difficult to provide comprehensive support for subsequent maintenance decisions, thus restricting the safe and stable application of carbon fiber conductors in key engineering fields. Summary of the Invention

[0005] Therefore, it is necessary to provide an intelligent non-destructive testing method and system for carbon fiber conductors based on AI image recognition that can solve the above problems.

[0006] In a first aspect, this application provides an intelligent non-destructive testing method for carbon fiber conductors based on AI image recognition, including: Obtain non-destructive testing image data of carbon fiber conductors, and extract microscopic feature data based on the non-destructive testing image data; Based on microscopic feature data, defect identification is performed through a pre-set carbon fiber conductor defect-feature database to obtain conductor defect identification results; Based on the conductor defect identification results, and taking into account the characteristics of carbon fiber conductors, potential defect correlation prediction is performed to obtain potential defect prediction results. Based on the potential defect prediction results, a pre-trained machine learning model is used to verify the defects and obtain the potential defect verification results. By integrating the results of conductor defect identification, potential defect prediction, and potential defect verification, a non-destructive testing report for carbon fiber conductors is generated.

[0007] In one embodiment, microscopic feature data is extracted based on nondestructive testing image data, including: Based on non-destructive testing image data, a pre-trained YOLO model is used to determine the image region to be detected; Based on the region of the image to be detected, edge contours are extracted to obtain contour feature data; Based on the contour feature data, grayscale feature data is obtained by calculating the grayscale value distribution; Based on grayscale feature data, texture density is calculated to obtain texture feature data; Based on texture feature data, porosity feature data is obtained through porosity detection; Based on pore feature data, fiber alignment direction is identified to obtain alignment feature data; Microscopic feature data is extracted by integrating contour feature data, grayscale feature data, texture feature data, pore feature data, and arrangement feature data.

[0008] In one embodiment, the carbon fiber conductor defect-feature database includes several defect types, as well as reference feature data ranges and confidence calculation rules corresponding to the defect types. The defect types include fiber breakage, pore abnormality, disordered arrangement, surface damage, uneven texture, and grayscale abrupt change. Based on microscopic feature data, defect identification is performed using a pre-set carbon fiber conductor defect-feature database to obtain conductor defect identification results, including: Based on microscopic feature data, the wire defect type is determined according to the reference feature data range corresponding to each defect type; For each type of conductor defect, based on microscopic feature data and according to the confidence calculation rules, the confidence level of the defect type is calculated. Based on the contour feature data in the microscopic feature data, defect area data and defect location data are extracted; By integrating conductor defect type, defect type confidence level, defect area data, and defect location data, the conductor defect identification results are obtained.

[0009] In one embodiment, the characteristics of the carbon fiber conductor include material fatigue characteristic curves and conductor structure topology; Based on the conductor defect identification results, and taking into account the characteristics of carbon fiber conductors, potential defect correlation prediction is performed to obtain the potential defect prediction results, including: Based on the defect location data in the conductor defect identification results, combined with the conductor structure topology, the associated area around the defect is determined; For the area surrounding the defect, based on the wire defect type and defect area data, combined with the material fatigue characteristic curve, the defect damage propagation trend value is calculated. Based on a preset trend threshold, the associated areas around defects with a defect damage expansion trend value greater than the trend threshold are selected to obtain the potential defect prediction results.

[0010] In one embodiment, the machine learning model is a long short-term memory network; Based on the potential defect prediction results, a pre-trained machine learning model is used for defect verification to obtain the potential defect verification results, including: A three-dimensional verification feature matrix is ​​constructed by using the defect-related region, defect damage propagation trend value, and conductor structure topology from the potential defect prediction results. Damage propagation features are extracted using a long short-term memory network based on a three-dimensional verification feature matrix. For each defect and its associated region, the cumulative damage probability density is calculated based on the material fatigue characteristic curve and damage propagation characteristics. The damage evolution path is obtained by simulating the damage accumulation probability density using the Monte Carlo method. Based on the damage evolution path, the failure probability confidence interval is calculated through Bayesian network inference; By integrating the damage evolution path and the failure probability confidence interval, the potential defect verification results are obtained.

[0011] In one embodiment, for the area surrounding the defect, based on the wire defect type and defect area data, and combined with the material fatigue characteristic curve, the defect damage propagation trend value is calculated using the following formula: in, This represents the defect damage propagation trend value. Here, A is the defect type correction factor, and A is the integral range of the associated region surrounding the defect. The fatigue characteristic curve of the material at the critical temperature Stress corrosion rate under the following conditions The stress on the material is N, and the number of strain cycles is N. For the defect morphology topology factor, This represents the surface roughness entropy value of the defect. is the fiber alignment direction correction factor, Q is the defect volume energy release rate, and L is the minimum feature size of the defect.

[0012] In one embodiment, the results of conductor defect identification, potential defect prediction, and potential defect verification are integrated to generate a non-destructive testing report for carbon fiber conductors, including: Based on the defect type, defect area data, and defect location data in the conductor defect identification results, a visualization map framework is constructed. For the defect-related area in the potential defect prediction results, the defect damage expansion trend value and the damage evolution path in the potential defect verification results are mapped to generate a damage development trajectory line. The confidence levels of defect types and failure probability confidence intervals are converted into heatmap overlays. By integrating a visualization framework, damage development trajectory lines, and a heat map overlay, a non-destructive testing report for carbon fiber conductors is obtained.

[0013] Secondly, this application also provides an intelligent non-destructive testing system for carbon fiber conductors based on AI image recognition, comprising: The image feature extraction module is used to acquire non-destructive testing image data of carbon fiber conductors and extract microscopic feature data based on the non-destructive testing image data. The feature defect recognition module is used to identify defects based on microscopic feature data and through a preset carbon fiber conductor defect-feature database, and obtain conductor defect recognition results. The defect association prediction module is used to perform potential defect association prediction based on the characteristics of carbon fiber conductors, according to the conductor defect identification results, and obtain potential defect prediction results. The potential defect verification module is used to verify defects based on the potential defect prediction results using a pre-trained machine learning model, and obtain the potential defect verification results. The test result integration module is used to integrate the wire defect identification results, potential defect prediction results, and potential defect verification results to generate a non-destructive testing report for carbon fiber wires.

[0014] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described intelligent non-destructive testing method for carbon fiber conductors based on AI image recognition.

[0015] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described intelligent non-destructive testing method for carbon fiber conductors based on AI image recognition.

[0016] The aforementioned intelligent non-destructive testing method and system for carbon fiber conductors based on AI image recognition acquires non-destructive testing image data of carbon fiber conductors and extracts microscopic feature data. AI image recognition technology is used to automatically complete feature extraction, avoiding the influence of human subjectivity. Based on the microscopic feature data, a pre-set carbon fiber conductor defect-feature database is used for defect identification, enabling the identification of defect types such as fiber fracture and abnormal porosity, and calculating confidence levels, thus improving the accuracy and efficiency of defect identification. Based on the defect identification results, combined with the material fatigue characteristic curve and structural topology of the carbon fiber conductor, potential defect correlation prediction is performed, calculating the defect damage propagation trend value, achieving prediction of damage propagation trends, and compensating for the lack of potential defect prediction capabilities. A pre-trained machine learning model, such as a long short-term memory network, is used for defect verification. By simulating damage evolution paths and calculating failure probability confidence intervals, the reliability of risk prediction is enhanced. The identification, prediction, and verification results are integrated to generate a comprehensive non-destructive testing report, presenting defect information in a visualized graph framework, solving the problem of insufficient information integration in traditional methods, and providing comprehensive decision support for the safe operation of conductors. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an intelligent non-destructive testing method for carbon fiber conductors based on AI image recognition according to the present invention. Figure 2 This is a structural diagram of an intelligent non-destructive testing system for carbon fiber conductors based on AI image recognition, according to the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] In one embodiment, such as Figure 1 As shown, an intelligent non-destructive testing method for carbon fiber conductors based on AI image recognition is provided. This embodiment illustrates the application of this method to a terminal, but it is understood that the method can also be applied to a server, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In the implementation environment, this method can be deployed on a terminal device (such as an inspection instrument equipped with a camera) to acquire images for non-destructive testing of carbon fiber conductors. Simultaneously, a server (which stores a database of carbon fiber conductor defects and features and a pre-trained machine learning model) is used to calculate the detection results. Application scenarios include: when efficient and accurate non-destructive testing of conductors is required in the power transmission or aerospace fields, the terminal device acquires image data in real time and transmits it to the server via the network. The server performs microscopic feature extraction, defect identification, potential defect prediction and verification based on the received data, and feeds the results back to the terminal. The terminal integrates and generates a visualized inspection report, realizing an intelligent inspection process with end-to-end cloud collaboration.

[0021] In this embodiment, the method includes the following steps: S01, acquire non-destructive testing image data of carbon fiber conductors, and extract microscopic feature data based on the non-destructive testing image data.

[0022] Non-destructive testing image data refers to image information of the interior and surface of carbon fiber conductors acquired through non-destructive testing techniques such as ultrasonic testing, X-ray testing, eddy current testing, or optical imaging. The aim is to capture defect features without damaging the material structure. Based on this image data, microscopic feature data is extracted. This data includes contour feature data (extracted after determining the area to be detected through edge detection algorithms such as the Canny operator or pre-trained models such as YOLO), grayscale feature data (through grayscale value distribution calculations such as histogram analysis), texture feature data (through texture density calculations such as the gray-level co-occurrence matrix GLCM), porosity feature data (through porosity detection algorithms), and alignment feature data (through fiber alignment direction identification techniques such as gradient direction histograms), etc., to achieve a comprehensive quantification of the microscopic state of the conductor. S02, based on microscopic feature data, defect identification is performed through a pre-set carbon fiber conductor defect-feature database to obtain conductor defect identification results.

[0023] The defect-feature database is a pre-defined knowledge base containing common defect types of carbon fiber conductors (such as fiber breakage, abnormal porosity, disordered arrangement, surface damage, uneven texture, and abrupt grayscale changes) and their corresponding reference feature data intervals (based on historical experimental data or standard specifications) and confidence calculation rules (such as probability models, similarity matching, or statistical thresholds). In implementation, microscopic feature data can be matched with reference intervals in the database, and the conductor defect type can be determined using algorithms (such as pattern recognition, machine learning classifiers, or rule engines). For each defect type, the defect type confidence can be calculated based on confidence calculation rules (such as Bayesian inference or fuzzy logic). Defect area data (through pixel counting or geometric modeling) and defect location data (through coordinate mapping or spatial reference systems) can be extracted from the contour feature data in the microscopic feature data. Integrating defect type, confidence, area, and location data generates conductor defect identification results, achieving efficient and accurate defect quantification analysis.

[0024] S03. Based on the wire defect identification results, and considering the characteristics of carbon fiber wires, perform potential defect correlation prediction to obtain potential defect prediction results.

[0025] The characteristics of carbon fiber conductors include material fatigue characteristic curves (describing the stress-life relationship of the material under cyclic loading, established based on experimental data) and conductor structure topology (geometric layout obtained through 3D scanning or modeling). In implementation, based on defect location data and conductor structure topology, spatial analysis (such as neighborhood search or graph theory methods) can be used to determine the associated region around the defect (i.e., the region that may be affected by stress concentration). For this region, based on defect type (adjusted by type correction coefficient) and defect area data, combined with material fatigue characteristic curves, a computational model (such as integral calculation, numerical simulation, or machine learning regression) is used to calculate the defect damage propagation trend value (quantifying the defect propagation risk). High-risk areas are screened according to a preset trend threshold (based on safety standards) to obtain the potential defect prediction results.

[0026] S04. Based on the potential defect prediction results, a pre-trained machine learning model is used to verify the defects and obtain the potential defect verification results.

[0027] The pre-trained machine learning model is an intelligent model trained based on historical defect data, such as a Long Short-Term Memory Network (LSTM), a Convolutional Neural Network (CNN), or a Support Vector Machine (SVM). In implementation, a multi-dimensional feature matrix (such as a three-dimensional verification feature matrix) can be constructed from the defect's surrounding associated region, defect damage propagation trend value, and conductor structure topology in the prediction results to capture the spatiotemporal characteristics of the defect. A machine learning model is used to extract damage propagation features (reflecting the defect propagation pattern). Based on these features, a probabilistic method (such as Monte Carlo simulation) is used to simulate the damage evolution path and predict the defect's changes over time. Simultaneously, inference algorithms (such as Bayesian networks) can be used to calculate the failure probability confidence interval to quantify the failure risk. Finally, the evolution path and probability interval are integrated to obtain the potential defect verification results.

[0028] S05 integrates the results of conductor defect identification, potential defect prediction, and potential defect verification to generate a non-destructive testing report for carbon fiber conductors.

[0029] In practice, data fusion technology can be used to integrate the results of conductor defect identification, potential defect prediction, and potential defect verification into a structured output. For example, a visualization framework (such as a two-dimensional or three-dimensional map) can be constructed based on defect type, area, and location data. Damage expansion trend values ​​and evolution paths can be mapped to generate damage development trajectory lines (reflecting the dynamic changes of defects). Confidence levels can be converted into heat map overlays (with color gradients representing risk levels). These elements can then be integrated to generate a comprehensive report, thereby achieving a comprehensive and intuitive presentation of the defect status of carbon fiber conductors and providing reliable support for subsequent maintenance decisions.

[0030] In one embodiment, microscopic feature data is extracted based on nondestructive testing image data, including: S11. Based on non-destructive testing image data, a pre-trained YOLO model is used to determine the image region to be detected; S12, Based on the region of the image to be detected, perform edge contour extraction to obtain contour feature data; S13, Based on contour feature data, grayscale feature data is obtained by calculating grayscale value distribution; S14, Based on the grayscale feature data, perform texture density calculation to obtain texture feature data; S15, based on texture feature data, porosity detection is used to obtain porosity feature data; S16. Based on the pore feature data, fiber arrangement direction is identified to obtain arrangement feature data; S17 integrates contour feature data, grayscale feature data, texture feature data, pore feature data, and arrangement feature data to extract microscopic feature data.

[0031] For example, the acquired non-destructive testing images of carbon fiber conductors can be input into a pre-trained YOLO model. This model, trained and optimized with a large number of carbon fiber conductor defect and normal image samples, can quickly and accurately locate the areas to be detected in the image that may contain defects, and remove meaningless background areas. For the determined image areas to be detected, the Canny edge detection algorithm can be used to extract edge contours, capturing the contour morphology of the conductor structure and potential defects within the area, forming contour feature data. Based on the contour feature data, the gray-level distribution of the corresponding area is calculated using the histogram statistical method to obtain key parameters such as gray-level mean and variance, resulting in gray-level feature data. Based on the gray-level feature data, the gray-level co-occurrence matrix (GLCM) method can be used to calculate the texture density and distribution pattern of the image area, generating texture feature data. A threshold segmentation algorithm is used to identify pores in the areas corresponding to the texture feature data, statistically analyzing the proportion and size distribution of pores, obtaining pore feature data. The gradient orientation histogram (HOG) algorithm can be used to analyze the image areas corresponding to the pore feature data, identifying the gradient direction distribution of the fibers and determining the arrangement feature data. According to the preset data format, five types of feature data, namely contour, grayscale, texture, porosity and arrangement, are integrated to form micro-feature data, which provides a comprehensive quantitative basis for subsequent defect identification.

[0032] In one embodiment, the carbon fiber conductor defect-feature database includes several defect types, as well as reference feature data ranges and confidence calculation rules corresponding to the defect types. The defect types include fiber breakage, pore abnormality, disordered arrangement, surface damage, uneven texture, and grayscale abrupt change. Based on microscopic feature data, defect identification is performed using a pre-set carbon fiber conductor defect-feature database to obtain conductor defect identification results, including: S21. Based on microscopic feature data, determine the wire defect type according to the reference feature data range corresponding to each defect type; S22, For each type of conductor defect, based on microscopic feature data and according to the confidence calculation rules, calculate the confidence level of the defect type; S23, based on the contour feature data in the microscopic feature data, extract the defect area data and defect location data; S24 integrates conductor defect type, defect type confidence level, defect area data, and defect location data to obtain conductor defect identification results.

[0033] Specifically, the pre-defined carbon fiber conductor defect-feature database stores six defect types, including fiber breakage and porosity anomalies. Each type corresponds to an experimentally calibrated reference feature data interval and a confidence calculation rule based on Bayesian inference. Microscopic feature data is compared with the reference feature data intervals for each defect type in the database. If a certain microscopic feature (such as porosity feature data) falls within the reference interval of the corresponding defect (such as porosity anomaly), the conductor defect type is determined. For the determined defect type, indices such as the degree of fit and feature parameter dispersion can be calculated based on the microscopic feature data and the reference interval. The confidence level of the defect type is quantified according to the confidence calculation rule, characterizing the reliability of the defect judgment. Based on the contour feature data in the microscopic feature data, the actual defect area data can be obtained by converting pixel counts with an image scale. The two-dimensional or three-dimensional coordinates of the defect in the conductor can be located based on the image coordinate system to obtain defect location data. According to the preset data structure, the defect type, defect type confidence, defect area data and defect location data are integrated to form the wire defect identification result containing the core defect information, providing basic data support for subsequent potential defect prediction.

[0034] In one embodiment, the characteristics of the carbon fiber conductor include material fatigue characteristic curves and conductor structure topology; S31, Based on the conductor defect identification results and the characteristics of carbon fiber conductors, perform potential defect correlation prediction to obtain potential defect prediction results, including: S32, Based on the defect location data in the conductor defect identification results, combined with the conductor structure topology, determine the associated area around the defect; S33, for the area associated with the defect, based on the wire defect type and defect area data, combined with the material fatigue characteristic curve, the defect damage propagation trend value is calculated; S34. Based on the preset trend threshold, filter out the surrounding associated areas of defects whose defect damage expansion trend value is greater than the trend threshold, and obtain the potential defect prediction results.

[0035] For example, based on the defect location data in the conductor defect identification results, combined with the conductor structure topology, a spatial neighborhood search algorithm is used to delineate a preset radius range as the defect's surrounding associated region, centered on the defect location coordinates and according to the conductor's geometric layout and stress transmission path. For this associated region, based on the determined conductor defect type (e.g., a defect type correction coefficient of 1.1 for fiber fracture and 0.9 for porosity anomaly), combined with the converted actual defect area data, the stress corrosion rate at the critical temperature is retrieved from the material fatigue characteristic curve and substituted into a preset formula. Simultaneously, parameters such as the defect morphology topology factor (proportional to the defect perimeter / area ratio), surface roughness entropy value (calculated through a gray-level co-occurrence matrix), and fiber alignment direction correction factor (inversely proportional to the cosine of the fiber orientation angle) can be entered to calculate the defect damage propagation trend value. A preset trend threshold (e.g., 0.7 based on engineering safety standards and extensive experimental data) is used to filter out defect-related regions where the defect damage propagation trend value is greater than the trend threshold. This region is considered to have potential failure risk. Integrating the coordinates and trend values ​​of these regions forms the potential defect prediction result.

[0036] In one embodiment, the machine learning model is a long short-term memory network; Based on the potential defect prediction results, a pre-trained machine learning model is used for defect verification to obtain the potential defect verification results, including: S41, construct a three-dimensional verification feature matrix from the defect surrounding area, defect damage expansion trend value and conductor structure topology in the potential defect prediction results; S42, based on the three-dimensional verification feature matrix, uses a long short-term memory network to extract damage propagation features; S43, For each defect-related region, calculate the cumulative damage probability density based on the material fatigue characteristic curve and damage propagation characteristics; S44, the damage evolution path is obtained by simulating the damage evolution path using the Monte Carlo method based on the damage accumulation probability density; S45. Based on the damage evolution path, the failure probability confidence interval is calculated through Bayesian network inference. S46 integrates the damage evolution path and the failure probability confidence interval to obtain the potential defect verification results.

[0037] Specifically, a verification feature matrix can be constructed based on the spatial coordinates of the associated region around the defect and the defect damage propagation trend value from the potential defect prediction results, combined with the geometric parameters of the conductor structure topology, according to a three-dimensional structure of "spatial dimension - trend value dimension - topological dimension". The spatial dimension is quantified by the boundary coordinates of the associated region, and the topological dimension is characterized by the conductor fiber arrangement density and hierarchical structure parameters. This three-dimensional verification feature matrix is ​​input into a pre-trained long short-term memory network, which has been trained and optimized with a large amount of carbon fiber conductor damage evolution time-series data. This network can automatically extract damage propagation features reflecting the time-varying law of defect propagation. For each associated region around the defect, stress-life relationship data can be retrieved from the material fatigue characteristic curve. Combined with the damage propagation features, the damage accumulation probability density is calculated using the Weibull distribution probability density function. Based on this probability density, multiple sets of damage development parameters can be randomly generated using the Monte Carlo method with 10,000 simulation iterations to simulate the damage evolution path under different working conditions. The damage evolution path is used as the observation data input into a pre-defined Bayesian network. This network has been calibrated with prior probabilities based on historical failure data. The failure probability confidence interval at a 95% confidence level is obtained through network inference calculation. The damage evolution path and the failure probability confidence interval are integrated to form the potential defect verification result.

[0038] In one embodiment, S51, for the area surrounding the defect, based on the wire defect type and defect area data, and combined with the material fatigue characteristic curve, the defect damage propagation trend value is calculated using the following formula: in, This represents the defect damage propagation trend value. Here, A is the defect type correction factor, and A is the integral range of the associated region surrounding the defect. The fatigue characteristic curve of the material at the critical temperature Stress corrosion rate under the following conditions The stress on the material is N, and the number of strain cycles is N. For the defect morphology topology factor, This represents the surface roughness entropy value of the defect. is the fiber alignment direction correction factor, Q is the defect volume energy release rate, and L is the minimum feature size of the defect.

[0039] For example, The calibration is based on the type of conductor defect, such as 1.2 for fiber breakage and 0.9 for porosity anomalies; A represents the geometrical spatial range of the region surrounding the defect, centered on the defect location coordinates, and the integration boundary (e.g., the three-dimensional spatial range of a 5mm radius region) is determined according to the conductor structure topology; the critical temperature is retrieved from the material fatigue characteristic curve. (For example, the stress corrosion rate corresponding to 80℃ is commonly used for carbon fiber conductors) ,in Take the actual working stress of the conductor and N as the design strain cycle number; The value is determined by image contour analysis based on the defect morphology (e.g., 1.1 for crack-like defects and 0.8 for pore-like defects). The entropy value of the pixel gray-level distribution on the defect surface is obtained by calculating the gray-level co-occurrence matrix; The angle between the fiber arrangement and the stress direction is set (e.g., 0.7 for the longitudinal direction and 1.3 for the perpendicular direction); Q is calculated using the fracture mechanics energy release rate formula based on defect area data and material elastic modulus; L is extracted from the profile feature data to determine the minimum geometric dimensions of the defect (e.g., minimum crack length, minimum hole diameter). These are then substituted into the formula to calculate the exponential term. ,Will and Multiply the integral result of A (where the integral variable dA covers the geometric boundary of the associated region) to calculate and The product of the two results, plus the product of the two parts, yields the defect damage propagation trend value. .

[0040] In one embodiment, the results of conductor defect identification, potential defect prediction, and potential defect verification are integrated to generate a non-destructive testing report for carbon fiber conductors, including: S61. Based on the defect type, defect area data and defect location data in the conductor defect identification results, a visualization map framework is constructed. S62, For the defect-related area in the potential defect prediction results, map the defect damage expansion trend value and the damage evolution path in the potential defect verification results to generate a damage development trajectory line. S63, converts the defect type confidence level and failure probability confidence interval into a heatmap overlay layer; S64 integrates a visualization framework, damage development trajectory lines, and a heat map overlay to obtain a non-destructive testing report for carbon fiber conductors.

[0041] Specifically, based on the defect type, defect area data, and defect location data from the conductor defect identification results, a visual atlas framework can be constructed using a three-dimensional coordinate system. The conductor entity is mapped to a digital model at a 1:1 scale, and the model accurately labels the type identifier, actual area value, and three-dimensional spatial coordinates of each defect, presenting the core information of visible defects. For the defect-related areas in the potential defect prediction results, using the time axis as the horizontal axis and the spatial coordinates as the vertical axis, the defect damage expansion trend value is mapped to the damage evolution path in the potential defect verification results. A dynamic damage development trajectory line is generated through line segment fitting, demonstrating the possible expansion direction and rate of the defect. The defect type confidence level (0-1 interval) and the failure probability confidence interval (e.g., the probability range at a 95% confidence level) are converted into a heatmap overlay layer, using a red-yellow-green color gradient to represent the risk level, with red corresponding to high confidence / high failure probability and green corresponding to low confidence / low failure probability. Finally, by using layer overlay technology, the visualization framework, damage development trajectory line, and heat map overlay are integrated, and textual information such as key defect parameters and risk level judgment criteria are added to form a non-destructive testing report for carbon fiber conductors that includes graphical display and quantitative data, providing an intuitive and comprehensive basis for maintenance decisions.

[0042] The aforementioned intelligent non-destructive testing method for carbon fiber conductors based on AI image recognition acquires non-destructive testing image data of carbon fiber conductors, uses a pre-trained YOLO model to locate the area to be tested and extracts microscopic feature data such as contour, grayscale, texture, porosity and fiber arrangement. Relying on a pre-set defect-feature database containing multiple defect types, reference feature data intervals and confidence calculation rules, it accurately matches defect types, calculates confidence, and obtains defect area and location information. Combining the fatigue characteristic curves and structural topology of carbon fiber conductor materials, it determines the associated areas around the defects based on the defect identification results and calculates the damage propagation trend value to screen high-risk potential defects. It uses a pre-trained long short-term memory network to construct a three-dimensional verification feature matrix, combines the Monte Carlo method to simulate the damage evolution path and Bayesian network inference to calculate the failure probability confidence interval, and realizes potential defect verification. It integrates the defect identification, potential defect prediction and verification results, and generates a comprehensive inspection report through a visualization map framework, damage development trajectory line and heat map overlay. This solution effectively addresses the technical problems of limited accuracy, high false negative and false positive rates, low efficiency, lack of potential defect prediction, and insufficient information integration in traditional testing methods. It improves the accuracy of defect identification and the reliability of potential risk prediction, adapts to the needs of large-scale continuous testing, and provides comprehensive and accurate decision support for the safe and stable operation of carbon fiber conductors.

[0043] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0044] Based on the same inventive concept, this application also provides an AI image recognition-based intelligent non-destructive testing system for carbon fiber conductors to implement the aforementioned AI image recognition-based intelligent non-destructive testing method for carbon fiber conductors. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the AI ​​image recognition-based intelligent non-destructive testing system for carbon fiber conductors provided below can be found in the above-described limitations of the AI ​​image recognition-based intelligent non-destructive testing method for carbon fiber conductors, and will not be repeated here.

[0045] In one exemplary embodiment, such as Figure 2 As shown, an intelligent non-destructive testing system for carbon fiber conductors based on AI image recognition is provided, comprising: The image feature extraction module 101 is used to acquire non-destructive testing image data of carbon fiber conductors and extract microscopic feature data based on the non-destructive testing image data. The feature defect recognition module 102 is used to identify defects based on microscopic feature data and through a preset carbon fiber conductor defect-feature database to obtain conductor defect recognition results. The defect association prediction module 103 is used to perform potential defect association prediction based on the characteristics of carbon fiber conductors and the conductor defect identification results to obtain potential defect prediction results. The potential defect verification module 104 is used to perform defect verification based on the potential defect prediction results using a pre-trained machine learning model, and obtain the potential defect verification results. The test result integration module 105 is used to integrate the wire defect identification results, potential defect prediction results and potential defect verification results to generate a non-destructive test report for carbon fiber wires.

[0046] In one embodiment, the image feature extraction module 101 is further configured to: Based on non-destructive testing image data, a pre-trained YOLO model is used to determine the image region to be detected; Based on the region of the image to be detected, edge contours are extracted to obtain contour feature data; Based on the contour feature data, grayscale feature data is obtained by calculating the grayscale value distribution; Based on grayscale feature data, texture density is calculated to obtain texture feature data; Based on texture feature data, porosity feature data is obtained through porosity detection; Based on pore feature data, fiber alignment direction is identified to obtain alignment feature data; Microscopic feature data is extracted by integrating contour feature data, grayscale feature data, texture feature data, pore feature data, and arrangement feature data.

[0047] In one embodiment, the carbon fiber conductor defect-feature database in the feature defect identification module 102 includes several defect types, as well as reference feature data ranges and confidence calculation rules corresponding to the defect types. The defect types include fiber breakage, abnormal pores, disordered arrangement, surface damage, uneven texture, and grayscale abrupt change. Based on microscopic feature data, defect identification is performed using a pre-set carbon fiber conductor defect-feature database to obtain conductor defect identification results, including: Based on microscopic feature data, the wire defect type is determined according to the reference feature data range corresponding to each defect type; For each type of conductor defect, based on microscopic feature data and according to the confidence calculation rules, the confidence level of the defect type is calculated. Based on the contour feature data in the microscopic feature data, defect area data and defect location data are extracted; By integrating conductor defect type, defect type confidence level, defect area data, and defect location data, the conductor defect identification results are obtained.

[0048] In one embodiment, the carbon fiber conductor characteristics in the defect correlation prediction module 103 include material fatigue characteristic curves and conductor structure topology. Based on the conductor defect identification results, and taking into account the characteristics of carbon fiber conductors, potential defect correlation prediction is performed to obtain the potential defect prediction results, including: Based on the defect location data in the conductor defect identification results, combined with the conductor structure topology, the associated area around the defect is determined; For the area surrounding the defect, based on the wire defect type and defect area data, combined with the material fatigue characteristic curve, the defect damage propagation trend value is calculated. Based on a preset trend threshold, the associated areas around defects with a defect damage expansion trend value greater than the trend threshold are selected to obtain the potential defect prediction results.

[0049] In one embodiment, the machine learning model in the potential defect verification module 104 is a long short-term memory network; Based on the potential defect prediction results, a pre-trained machine learning model is used for defect verification to obtain the potential defect verification results, including: A three-dimensional verification feature matrix is ​​constructed by using the defect-related region, defect damage propagation trend value, and conductor structure topology from the potential defect prediction results. Damage propagation features are extracted using a long short-term memory network based on a three-dimensional verification feature matrix. For each defect and its associated region, the cumulative damage probability density is calculated based on the material fatigue characteristic curve and damage propagation characteristics. The damage evolution path is obtained by simulating the damage accumulation probability density using the Monte Carlo method. Based on the damage evolution path, the failure probability confidence interval is calculated through Bayesian network inference; By integrating the damage evolution path and the failure probability confidence interval, the potential defect verification results are obtained.

[0050] In one embodiment, the defect association prediction module 103 is further configured to calculate the defect damage propagation trend value for the associated region surrounding the defect using the following formula: based on the wire defect type and defect area data, combined with the material fatigue characteristic curve. in, This represents the defect damage propagation trend value. Here, A is the defect type correction factor, and A is the integral range of the associated region surrounding the defect. The fatigue characteristic curve of the material at the critical temperature Stress corrosion rate under the following conditions The stress on the material is N, and the number of strain cycles is N. For the defect morphology topology factor, This represents the surface roughness entropy value of the defect. is the fiber alignment direction correction factor, Q is the defect volume energy release rate, and L is the minimum feature size of the defect.

[0051] In one embodiment, the detection result integration module 105 is further configured to: Based on the defect type, defect area data, and defect location data in the conductor defect identification results, a visualization map framework is constructed. For the defect-related area in the potential defect prediction results, the defect damage expansion trend value and the damage evolution path in the potential defect verification results are mapped to generate a damage development trajectory line. The confidence levels of defect types and failure probability confidence intervals are converted into heatmap overlays. By integrating a visualization framework, damage development trajectory lines, and a heat map overlay, a non-destructive testing report for carbon fiber conductors is obtained.

[0052] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the intelligent non-destructive testing method for carbon fiber conductors based on AI image recognition as described above.

[0053] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0054] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0055] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligent non-destructive testing of carbon fiber conductors based on AI image recognition, characterized in that, The method includes: Obtain non-destructive testing image data of carbon fiber conductors, and extract microscopic feature data based on the non-destructive testing image data; Based on the aforementioned microscopic feature data, defect identification is performed using a pre-set carbon fiber conductor defect-feature database to obtain conductor defect identification results. Based on the defect identification results of the conductor, and based on the characteristics of the carbon fiber conductor, potential defect correlation prediction is performed to obtain potential defect prediction results; Based on the predicted potential defects, a pre-trained machine learning model is used to verify the defects, and the potential defect verification results are obtained. By integrating the conductor defect identification results, the potential defect prediction results, and the potential defect verification results, a non-destructive testing report for carbon fiber conductors is generated.

2. The method according to claim 1, characterized in that, The extraction of microscopic feature data based on the non-destructive testing image data includes: Based on non-destructive testing image data, a pre-trained YOLO model is used to determine the image region to be detected; Based on the image region to be detected, edge contour extraction is performed to obtain contour feature data; Based on the contour feature data, grayscale feature data is obtained by calculating the grayscale value distribution; Based on the grayscale feature data, texture density is calculated to obtain texture feature data; Based on the texture feature data, porosity feature data is obtained through porosity detection; Based on the pore feature data, fiber arrangement direction is identified to obtain arrangement feature data; By integrating the contour feature data, the grayscale feature data, the texture feature data, the pore feature data, and the arrangement feature data, microscopic feature data is extracted.

3. The method according to claim 2, characterized in that, The carbon fiber conductor defect-feature database includes several defect types, as well as reference feature data intervals and confidence calculation rules corresponding to the defect types. The defect types include fiber breakage, abnormal pores, disordered arrangement, surface damage, uneven texture, and abrupt grayscale changes. Based on the microscopic feature data, defect identification is performed through a preset carbon fiber conductor defect-feature database to obtain conductor defect identification results, including: Based on the microscopic feature data, the wire defect type is determined according to the reference feature data range corresponding to each defect type; For each of the aforementioned conductor defect types, based on the microscopic feature data and according to the confidence calculation rules, the defect type confidence score is calculated. Based on the contour feature data in the microscopic feature data, defect area data and defect location data are extracted; By integrating the conductor defect type, the defect type confidence level, the defect area data, and the defect location data, the conductor defect identification result is obtained.

4. The method according to claim 3, characterized in that, The properties of the carbon fiber conductor include material fatigue characteristic curves and conductor structure topology; The step of performing potential defect correlation prediction based on the conductor defect identification results and the characteristics of carbon fiber conductors to obtain potential defect prediction results includes: Based on the defect location data in the conductor defect identification results, and combined with the conductor structure topology, the associated area around the defect is determined; For the area surrounding the defect, based on the wire defect type and the defect area data, and combined with the material fatigue characteristic curve, the defect damage propagation trend value is calculated. Based on a preset trend threshold, the associated regions around defects whose defect damage expansion trend value is greater than the trend threshold are selected to obtain the potential defect prediction result.

5. The method according to claim 4, characterized in that, The machine learning model is a long short-term memory network; The step of using a pre-trained machine learning model to verify defects based on the potential defect prediction results, and obtaining potential defect verification results, includes: A three-dimensional verification feature matrix is ​​constructed using the defect surrounding region, defect damage propagation trend value, and conductor structure topology from the potential defect prediction results. Based on the three-dimensional verification feature matrix, the damage propagation features are extracted using the long short-term memory network. For each of the aforementioned defect-related regions, the cumulative damage probability density is calculated based on the material fatigue characteristic curve and the damage propagation characteristics. Based on the aforementioned damage accumulation probability density, the damage evolution path is simulated using the Monte Carlo method. Based on the damage evolution path, the failure probability confidence interval is calculated through Bayesian network inference. By integrating the damage evolution path and the failure probability confidence interval, the verification result of the potential defect is obtained.

6. The method according to claim 4, characterized in that, For the area surrounding the defect, based on the wire defect type and the defect area data, and combined with the material fatigue characteristic curve, the defect damage propagation trend value is calculated using the following formula: in, This represents the defect damage propagation trend value. Here, A is the defect type correction factor, and A is the integral range of the associated region surrounding the defect. The fatigue characteristic curve of the material at the critical temperature Stress corrosion rate under the following conditions The stress on the material is N, and the number of strain cycles is N. For the defect morphology topology factor, This represents the surface roughness entropy value of the defect. is the fiber alignment direction correction factor, Q is the defect volume energy release rate, and L is the minimum feature size of the defect.

7. The method according to claim 5, characterized in that, The process integrates the conductor defect identification results, the potential defect prediction results, and the potential defect verification results to generate a non-destructive testing report for carbon fiber conductors, including: Based on the defect type, defect area data, and defect location data in the wire defect identification results, a visualization map framework is constructed. For the defect-related region in the potential defect prediction results, the defect damage expansion trend value and the damage evolution path in the potential defect verification results are mapped to generate a damage development trajectory line. The confidence levels of the defect types and the confidence intervals of the failure probabilities are converted into a heatmap overlay layer; By integrating the visualization framework, the damage development trajectory line, and the overlay of the heat map, a non-destructive testing report for the carbon fiber conductor is obtained.

8. An intelligent non-destructive testing system for carbon fiber conductors based on AI image recognition, characterized in that, The system includes: The image feature extraction module is used to acquire non-destructive testing image data of carbon fiber conductors and extract microscopic feature data based on the non-destructive testing image data. The feature defect identification module is used to identify defects based on the microscopic feature data and through a preset carbon fiber conductor defect-feature database to obtain conductor defect identification results. The defect association prediction module is used to perform potential defect association prediction based on the characteristics of carbon fiber conductors, according to the conductor defect identification results, and obtain potential defect prediction results. The potential defect verification module is used to perform defect verification using a pre-trained machine learning model based on the potential defect prediction results, and obtain the potential defect verification results. The test result integration module is used to integrate the conductor defect identification results, the potential defect prediction results, and the potential defect verification results to generate a non-destructive testing report for carbon fiber conductors.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images