An AI algorithm-based intelligent defect identification and detection method for steel-plastic composite pipes

Abstract

The application discloses a kind of steel plastic composite pipe defect intelligent identification detection method based on AI algorithm, comprising the following steps: S1: the multi-view image of steel plastic composite pipe outer wall, end face, fusion zone is synchronously collected, and pipe production parameters are synchronously bound, to obtain initial data set;S2: texture feature extraction is carried out to initial data set, to obtain texture feature data;S3: AI model is constructed, and hybrid loss function is used to iteratively train model, to complete defect feature automatic learning;S4: texture feature data is input into the inference of AI model trained, identifies defect position, size and type, and combines preset threshold to divide defect risk level.The present application relies on amplitude and depth dual-mode image fusion, three-dimensional point cloud coordinate matching multiple technologies, synchronously identifies pipe surface layer and internal hidden defect and quantifies damage expansion risk, avoids traditional single visual detection missed detection and misdiagnosis problem, realizes steel plastic composite pipe automatic high-precision grading quality inspection.

Description

Technical Field

[0001] This invention relates to the field of intelligent defect identification technology, and in particular to an intelligent defect identification and detection method for steel-plastic composite pipes based on AI algorithms. Background Technology

[0002] Steel-plastic composite pipes are widely used in municipal water supply and drainage, industrial fluid transportation, fire water supply and other engineering scenarios. During the extrusion molding, end cutting and multi-layer welding and sealing process, defects such as exposed steel wires, weld gaps, dents and bulges in the pipe wall, uneven wall thickness and surface scratches are very likely to occur. These defects will greatly reduce the pressure-bearing and seepage-proof performance of the pipe, and after long-term water flow, leakage, pipe burst and other safety accidents are likely to occur.

[0003] Currently, the mainstream quality inspection methods in the industry are divided into two categories: manual visual inspection and single two-dimensional visual inspection. Manual inspection relies on workers' visual judgment, which is inefficient and has a very high rate of missed detection for hidden defects such as internal dents and tiny weld gaps in pipes. Moreover, the standards of manual judgment are affected by subjective conditions, resulting in poor consistency in quality inspection. Single two-dimensional visual inspection only relies on planar amplitude images to identify defects, and cannot obtain three-dimensional depth information of the pipe, making it difficult to capture three-dimensional defects such as wall thickness changes and internal collapses. At the same time, the steel wire mesh on the surface of the pipe is prone to reflective spots, which will cause a large number of false detections. Existing AI visual inspection models only use single-modal image training and do not combine the three-dimensional spatial information of the pipe and the prior mechanical laws. They can only achieve static single-frame defect recognition, and cannot quantify the defect expansion trend, assess the long-term pressure damage risk, or achieve integrated management of defect classification and automated sorting on the production line. Therefore, an intelligent identification and detection method for steel-plastic composite pipe defects based on AI algorithms is proposed to solve the above problems. Summary of the Invention

[0004] Based on the technical problems existing in the background technology, this invention proposes an intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms, in order to solve the above problems.

[0005] This invention proposes an intelligent defect identification and detection method for steel-plastic composite pipes based on AI algorithms, comprising the following steps:

[0006] S1: Simultaneously acquire multi-view images of the outer wall, end face, and welding area of ​​the steel-plastic composite pipe and simultaneously bind pipe production parameters to obtain the initial dataset;

[0007] S2: Extract texture features from the initial dataset to obtain texture feature data;

[0008] S3: Build an AI model, use a hybrid loss function to iteratively train the model, and complete the automatic learning of defect features;

[0009] S4: Input texture feature data into the trained AI model for inference, identify the location, size and type of defects, and classify the defect risk level based on preset thresholds.

[0010] In a preferred embodiment, in step S1, the initial dataset includes at least two types of raw data: amplitude pixel data of pipe surface imaging and depth data output from ranging sensors. Based on the two types of data, amplitude imaging maps and depth imaging maps are generated respectively, and combined to form a dual-modal image sample set of the pipe.

[0011] In a preferred embodiment, in step S2, the corresponding geometric regions where the pipe defects are located are delineated in the amplitude imaging map and the depth imaging map, respectively. The intersection operation of the two regions is performed to delineate a unified fusion defect analysis area, which will not overlap with the two original defect regions.

[0012] In a preferred embodiment, in step S2, based on the defined fusion defect analysis area, the amplitude image pixel features and depth image spatial features within the area are extracted respectively.

[0013] The correlation strength between the two types of features and the identification of various defects in pipes is calculated. Weighting coefficients are assigned according to the correlation strength, and a unified fusion feature vector is generated by weighted summation.

[0014] In a preferred embodiment, in step S3, a laser scanning device is used to collect three-dimensional point information of the pipe and build a three-dimensional point cloud model dedicated to the pipe.

[0015] The visible light image of the pipe and the 3D point cloud are spatially aligned to establish a mapping and binding relationship between pixel positions and 3D spatial coordinates.

[0016] In a preferred embodiment, in step S3, a regularization term for pipe mechanics prior physical constraints is added to the loss term of the convolutional network, and the regularization term is incorporated into the segmentation base loss to synthesize the overall optimization objective.

[0017] Based on the labeled pipe defect samples, the network parameters are iteratively updated, and after outputting the defect probability distribution map, threshold cutting is performed to calculate the geometric contour of the defect area and the extension main direction parameters.

[0018] In a preferred embodiment, in step S4, the spatiotemporal sequence data of defects are continuously extracted according to a fixed sampling period, the displacement and movement rate of the defect points are calculated segment by segment, and the acceleration time series array corresponding to the extension of the defect area is further derived. The classification of pipe defect evolution patterns is completed based on the short-time time series network.

[0019] In a preferred embodiment, in step S4, the matching relationship between the main direction of defect extension and the orientation of the pipe bearing pressure is compared to assess the risk of structural damage caused by the continuous expansion of the defect.

[0020] Based on the defect classification results and risk rating, output quality inspection and handling instructions.

[0021] The beneficial effects of this invention are:

[0022] This invention eliminates false detection interference caused by steel wire reflection and imaging noise by dual-modal defect region intersection screening and correlation weighted fusion features, while taking into account both surface appearance defects and three-dimensional hidden defects; combined with three-dimensional point cloud registration to map two-dimensional pixels to the real space coordinates of the pipe, and introduces pipe mechanical prior regularization terms to optimize the training of convolutional network, the accuracy of defect segmentation boundary is greatly improved.

[0023] This invention predicts the expansion and evolution trend of defects by using time-series data of defect displacement, velocity, and acceleration, and quantifies the damage risk level by combining the direction of pipe pressure stress. It is not limited to single-frame static defect identification and can distinguish between stable minor defects and continuously expanding high-risk defects. The entire process is compatible with fully automated production lines and can automatically output sorting and disposal instructions and simultaneously archive all quality inspection data, significantly reducing the cost of manual quality inspection. Detailed Implementation

[0024] The present invention will be further explained below with reference to specific embodiments.

[0025] This embodiment proposes an intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms, including the following steps:

[0026] S1: Simultaneously acquire multi-view images of the outer wall, end face, and welding area of ​​the steel-plastic composite pipe and simultaneously bind pipe production parameters to obtain the initial dataset;

[0027] S2: Extract texture features from the initial dataset to obtain texture feature data;

[0028] S3: Build an AI model, use a hybrid loss function to iteratively train the model, and complete the automatic learning of defect features;

[0029] S4: Input texture feature data into the trained AI model for inference, identify the location, size and type of defects, and classify the defect risk level based on preset thresholds.

[0030] In step S1, the initial dataset contains at least two types of raw data: amplitude pixel data of pipe surface imaging and depth data output from ranging sensors. Based on these two types of data, amplitude imaging maps and depth imaging maps are generated respectively, and combined to form a dual-modal image sample set of the pipe.

[0031] Single-plane imaging is easily affected by reflections from the steel wire mesh on the surface of steel-plastic composite pipes and color differences in the PE substrate. Three-dimensional hidden defects such as shallow internal depressions and localized thinning of the wall are difficult for ordinary cameras to capture. Amplitude images, relying on pixel brightness differences, can accurately distinguish two-dimensional surface defects such as exposed steel wires, surface scratches, weld joint cracks, and surface color differences and stains. Depth images, relying on spatial distance information, can quantify the degree of unevenness in the pipe wall and detect three-dimensional structural defects such as hidden localized bulges, uneven wall thickness, and internal collapse of the sealing layer. The data from these two modalities work together to reduce the probability of model misjudgments caused by texture reflections, thus reducing the risk of leakage in engineering projects caused by missed inspections at the pipe factory.

[0032] At the quality inspection station of the DN200 steel wire mesh reinforced composite pipe production line, a high frame rate area array industrial camera, paired with a ring-shaped soft light source, captures images of the pipe end face and outer wall. Pixel brightness information is processed into amplitude imaging, capable of distinguishing minute surface scratches as small as 0.2mm and scattered steel wires protruding from the sealing position. Simultaneously deployed line laser rangefinders scan along the circumference of the pipe, outputting relative distance values ​​for each point, which are converted into depth imaging, capable of identifying minor defects such as dents on the inner wall of the pipe even when the surface appearance is normal. Two images corresponding to the same section of pipe are batch-numbered and bound, and grouped into a unified dual-modal image sample set. A portion of the samples are labeled with defects and used for model training to optimize weights, while the remaining online-acquired samples are directly fed into the model to complete real-time defect detection on the production line.

[0033] In step S2, the corresponding geometric regions where the pipe defects are located are delineated in the amplitude imaging map and the depth imaging map, respectively. The intersection operation of the two regions is performed to delineate a unified fusion defect analysis area, which will not overlap with the two original defect regions.

[0034] In step S2, based on the defined fusion defect analysis area, the amplitude image pixel features and depth image spatial features within the area are extracted respectively.

[0035] The correlation strength between the two types of features and the identification of various defects in pipes is calculated. Weighting coefficients are assigned according to the correlation strength, and a unified fusion feature vector is generated by weighted summation.

[0036] By separately delineating defect areas in two types of images and then taking their intersection to define the analysis range, false defect points caused by reflections, light spots, and ranging noise in single-modal imaging can be filtered out, and invalid interference areas inherent in the single modality can be eliminated. Weighting coefficients are dynamically assigned according to the correlation between features and defect identification, giving higher weights to features with stronger identifiability. This amplifies the effective representation of pipe defects such as scratches and internal dents, while reducing negative interference caused by wire mesh texture and light and shadow fluctuations. The final output fused feature vector integrates both appearance texture and three-dimensional structure information, resulting in stronger feature identification.

[0037] Taking the quality inspection of the sealing part of a DN150 steel wire mesh composite pipe as an example, a false defect spot is generated in the amplitude imaging image due to the reflection of the steel wire. The depth imaging image does not have this abnormal point. After finding the intersection of the two defect areas, the reflective interference area is directly removed. In the remaining locked fused defect area, the grayscale features of the weld seam pixels at the amplitude level and the depression space drop features at the depth level are extracted respectively. After quantitative evaluation, the depth drop feature has a stronger correlation with the discrimination of internal hidden collapse defects, and is assigned a weight of 0.6. The grayscale texture feature is assigned a weight of 0.4. The weighted sum is used to obtain the integrated one-dimensional fused feature vector, which is fed into the subsequent AI model to determine the defect category. This can avoid the reflective spot being misjudged as a crack defect and accurately identify the collapse defects on the inner side of the seal that are not easily observed by the naked eye.

[0038] In step S3, a laser scanning device is used to collect the three-dimensional point information of the pipe and build a three-dimensional point cloud model of the pipe.

[0039] The visible light image of the pipe and the 3D point cloud are spatially aligned to establish a mapping and binding relationship between pixel positions and 3D spatial coordinates.

[0040] In step S3, a regularization term is constructed by adding prior physical constraints of pipe mechanics to the loss term of the convolutional network, and the regularization term is incorporated into the segmentation base loss to synthesize the overall optimization objective.

[0041] Based on the labeled pipe defect samples, the network parameters are iteratively updated, and after outputting the defect probability distribution map, threshold cutting is performed to calculate the geometric contour of the defect area and the extension main direction parameters.

[0042] By constructing a 3D point cloud model of the pipe and completing image-point cloud spatial registration, the planar defect pixels of the 2D image can be mapped to the actual 3D spatial location of the pipe. A regularized loss term is constructed by introducing prior physical constraints related to pipe mechanics, and the pressure-bearing and stress-deformation laws of the welded layer of the steel-plastic composite pipe are integrated into the model training process. This ensures that the network's defect segmentation logic aligns with the actual failure mechanism of the pipe, avoiding erroneous segmentation results that deviate from real-world conditions based solely on pixel grayscale differences. After training, the model can simultaneously output two types of geometric parameters: defect contour and the main direction of defect extension, improving the accuracy of defect segmentation boundaries.

[0043] For the inspection of the fusion sealing station of DN250 steel wire mesh composite pipe, a LiDAR system is used to scan the entire circumference of the pipe, collecting the three-dimensional coordinates of all points on the pipe wall to generate a complete three-dimensional point cloud model of the pipe. A calibration plate is used to match the spatial coordinates of the visible light image with the point cloud, ensuring that each pixel in the fusion seam corresponds to a unique three-dimensional spatial coordinate of the pipe wall. When building the defect segmentation convolutional network, a regular constraint term is designed based on the mechanical laws of stress cracking in the fusion layer of the steel-plastic composite pipe. This term is combined with the basic segmentation loss to form the total model loss. The model is iteratively trained using pipe samples labeled with exposed steel wire, internal depressions, and fusion seams. After the model outputs a defect probability map, a fixed threshold is set for segmentation to accurately extract the complete contour of the defect at the sealing point. Simultaneously, the principal direction parameters of the defect's radial extension along the pipe wall are calculated to determine whether there is a risk of continuous expansion and breakage of the fusion seam.

[0044] In step S4, the spatiotemporal sequence data of defects are continuously extracted according to a fixed sampling period, the displacement and motion rate of the defect points are calculated segment by segment, and the acceleration time series array corresponding to the extension of the defect area is further derived. Based on the short-time time series network, the pipe defect evolution mode is classified.

[0045] In step S4, the matching relationship between the main direction of defect extension and the bearing stress direction of the pipe is compared to assess the risk of structural damage caused by the continuous expansion of the defect.

[0046] Based on the defect classification results and risk rating, output quality inspection and handling instructions.

[0047] By continuously collecting multi-cycle spatiotemporal sequences of defects, the system quantifies the evolution of defects as production conditions progress through displacement, velocity, and acceleration time-series data. It automatically distinguishes between two evolution modes—stable minor defects and continuously expanding high-risk damage—using short-time time-series networks, thus overcoming the limitation that single-frame static images can only identify instantaneous defects. Simultaneously, by combining the pipe's pressure-bearing mechanical logic and comparing the coincidence of the defect extension direction with the principal stress direction, the system quantifies the risk of damage and ultimately outputs standardized sorting and disposal instructions. This allows for the early identification of pipes with long-term leakage or pipe burst risks.

[0048] The production line continuously collects multi-frame spatiotemporal data on sealing defects of DN300 steel wire mesh composite pipes. It calculates the pipe wall displacement and extension velocity at the center point of the defect area cycle by cycle, deriving the time-series sequence of defect expansion acceleration. This sequence is input into a short-time memory time-series network to distinguish defect evolution patterns. After extracting the main direction of defect extension, it compares it with the direction of the principal stress on the pipe wall generated by water pressure. If the angle between the two is less than 30°, the weld defect is judged to have a high risk of continuous expansion and subsequent leakage; if the angle is greater than 70°, it is judged as a low-risk minor defect. Combining the "welded through gap" defect category and high-risk rating identified by AI, the system automatically outputs a scrapping sorting instruction, pushing it to the production line sorting mechanism to divert the pipes to the scrapping area. If it is a low-risk surface scratch, a rework and re-inspection instruction is output, and the defect evolution time-series and risk level data are simultaneously stored in the quality database for traceability and archiving.

[0049] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0050] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0051] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing embodiments, and will not be repeated here.

[0052] In the several embodiments provided in this application, it should be understood that the disclosed systems and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0053] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0054] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0055] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0056] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0057] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent identification and detection of defects in steel-plastic composite pipes based on AI algorithms, characterized in that, Includes the following steps: S1: Simultaneously acquire multi-view images of the outer wall, end face, and welding area of ​​the steel-plastic composite pipe and simultaneously bind pipe production parameters to obtain the initial dataset; S2: Extract texture features from the initial dataset to obtain texture feature data; S3: Build an AI model, use a hybrid loss function to iteratively train the model, and complete the automatic learning of defect features; S4: Input texture feature data into the trained AI model for inference, identify the location, size and type of defects, and classify the defect risk level based on preset thresholds.

2. The intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms according to claim 1, characterized in that: In step S1, the initial dataset contains at least two types of raw data: amplitude pixel data of pipe surface imaging and depth data output from ranging sensors. Based on these two types of data, amplitude imaging maps and depth imaging maps are generated respectively, and combined to form a dual-modal image sample set of the pipe.

3. The intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms according to claim 2, characterized in that: In step S2, the corresponding geometric regions where the pipe defects are located are delineated in the amplitude imaging map and the depth imaging map, respectively. The intersection operation of the two regions is performed to delineate a unified fusion defect analysis area, which will not overlap with the two original defect regions.

4. The intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms according to claim 3, characterized in that: In step S2, based on the defined fusion defect analysis area, the amplitude image pixel features and depth image spatial features within the area are extracted respectively. The correlation strength between the two types of features and the identification of various defects in pipes is calculated. Weighting coefficients are assigned according to the correlation strength, and a unified fusion feature vector is generated by weighted summation.

5. The intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms according to claim 4, characterized in that: In step S3, a laser scanning device is used to collect the three-dimensional point information of the pipe and build a three-dimensional point cloud model of the pipe. The visible light image of the pipe and the 3D point cloud are spatially aligned to establish a mapping and binding relationship between pixel positions and 3D spatial coordinates.

6. The intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms according to claim 5, characterized in that: In step S3, a regularization term is constructed by adding prior physical constraints of pipe mechanics to the loss term of the convolutional network, and the regularization term is incorporated into the segmentation base loss to synthesize the overall optimization objective. Based on the labeled pipe defect samples, the network parameters are iteratively updated, and after outputting the defect probability distribution map, threshold cutting is performed to calculate the geometric contour of the defect area and the extension main direction parameters.

7. The intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms according to claim 6, characterized in that: In step S4, the spatiotemporal sequence data of defects are continuously extracted according to a fixed sampling period, the displacement and motion rate of the defect points are calculated segment by segment, and the acceleration time series array corresponding to the extension of the defect area is further derived. Based on the short-time time series network, the pipe defect evolution mode is classified.

8. The intelligent identification and detection method for defects in steel-plastic composite pipes based on AI algorithms according to claim 7, characterized in that: In step S4, the matching relationship between the main direction of defect extension and the bearing stress direction of the pipe is compared to assess the risk of structural damage caused by the continuous expansion of the defect. Based on the defect classification results and risk rating, output quality inspection and handling instructions.