A pipeline corrosion defect intelligent detection system based on physical constraint artificial intelligence

Abstract

This invention aims to provide an intelligent pipeline corrosion defect detection system based on physically constrained artificial intelligence, addressing the problems of insufficient corrosion detection data, low defect identification accuracy, and poor automation in existing technologies. The invention constructs an integrated intelligent detection system comprising a corrosion database module, a corrosion defect generation module, a magnetic flux leakage signal inversion module, a corrosion defect identification module, and a corrosion risk assessment module. It achieves high-precision automatic detection and assessment of corrosion defects through a physically constrained artificial intelligence model. These modules are interconnected via data interfaces to form a complete intelligent detection system. The advantages of this invention include: providing ample high-quality training samples for intelligent detection; improving the identification accuracy of corrosion defects by more than 20% compared to traditional methods; significantly improving detection efficiency and reducing human error; enhancing pipeline safety assessment capabilities; and exhibiting good scalability.

Description

Technical Field

[0001] This invention belongs to the fields of pipeline corrosion detection technology and artificial intelligence technology. Specifically, it relates to an intelligent detection system for pipeline corrosion defects based on artificial intelligence algorithms and physical constraint models. It is particularly suitable for the automatic detection, identification and risk assessment of corrosion defects in infrastructure such as oil and gas pipelines and urban underground pipe networks. Background Technology

[0002] Pipeline transportation is a primary mode of transporting oil and gas resources, offering advantages such as high efficiency, low cost, and good safety. However, with the increasing service life of pipelines, corrosion problems are becoming increasingly prominent. Pipeline corrosion not only leads to thinning of the pipe wall but can also cause defects such as localized pitting and cracking, seriously threatening the structural integrity and operational safety of the pipeline. Statistics show that corrosion is one of the main causes of pipeline failure; therefore, regular inspection and accurate assessment of pipeline corrosion conditions are crucial.

[0003] Currently, magnetic flux leakage (MF) testing is the primary technology used for pipeline corrosion detection. MF testing utilizes permanent magnets or electromagnets to magnetize the pipe wall to saturation. When corrosion defects exist in the pipe wall, the magnetic permeability changes at the defect location, causing some magnetic lines of force to leak out of the pipe wall. A magnetic sensor placed on a probe detects the leaked magnetic field, thereby determining the location and characteristics of the defect. MF testing has advantages such as high detection speed, relatively low cost, and sensitivity to defects, making it the mainstream technology for pipeline inspection. However, traditional MF testing methods have the following problems in practical applications: First, acquiring detection data is difficult, and there is a shortage of corrosion defect samples. The interpretation of magnetic flux leakage detection signals relies on a large amount of calibration data for known defects. However, obtaining real corrosion defect samples through accelerated corrosion experiments in the laboratory is time-consuming and costly, making it difficult to obtain a large-scale dataset covering various corrosion types and morphologies. This lack of data limits the training of detection models, making it difficult to achieve high-precision defect identification.

[0004] Second, the defect identification accuracy is low, especially the ability to quantitatively assess complex corrosion morphologies. Traditional methods usually simplify the relationship between leakage magnetic field signals and defect size to empirical formulas or finite element fitting. However, the morphology of actual corrosion pits is complex and diverse, and these simplified methods cannot accurately reflect key parameters such as depth, volume, and shape, resulting in large errors in the identification results.

[0005] Third, the level of automation in the detection process is low, relying heavily on human experience. Currently, the interpretation of magnetic flux leakage detection data often requires professionals to manually interpret the signal waveforms, which is inefficient and highly subjective, making it difficult to meet the automation requirements of large-scale pipeline inspection.

[0006] In recent years, the rapid development of artificial intelligence technology has brought new opportunities to the field of corrosion detection. Deep learning models can automatically learn features from large amounts of data and have been successfully applied to fields such as image recognition and signal processing. Some researchers have made some progress in using models such as convolutional neural networks to identify defects from magnetic flux leakage signals. However, most existing research focuses on a single step, such as only performing defect identification or only performing signal inversion, lacking an intelligent detection system that can integrate functions such as corrosion database construction, defect data generation, signal inversion, defect identification, and risk assessment. In addition, purely data-driven artificial intelligence models often ignore the physical mechanism of magnetic flux leakage detection, resulting in limited generalization ability and reliability of the models.

[0007] Therefore, there is an urgent need to develop an intelligent pipeline corrosion defect detection system based on physical constraint artificial intelligence. This system can integrate corrosion databases, defect generation models, inversion algorithms, and risk assessment modules to achieve full-process intelligentization from data acquisition to safety assessment, thereby improving the accuracy and automation level of corrosion defect detection. Summary of the Invention

[0008] This invention aims to provide an intelligent pipeline corrosion defect detection system based on physically constrained artificial intelligence, addressing the problems of insufficient corrosion detection data, low defect identification accuracy, and poor automation in existing technologies. Specifically, the purpose of this invention is to construct an integrated intelligent detection system comprising a corrosion database module, a corrosion defect generation module, a magnetic flux leakage signal inversion module, a corrosion defect identification module, and a corrosion risk assessment module. This system achieves high-precision automatic detection and assessment of corrosion defects through a physically constrained artificial intelligence model.

[0009] This invention provides an intelligent detection system for pipeline corrosion defects based on physical constraint artificial intelligence, comprising the following modules: The corrosion database module stores and manages data related to pipeline corrosion defects. The database contains two types of data: first, real corrosion defect data, obtained by preparing corrosion samples under different environmental conditions through accelerated corrosion experiments in the laboratory, acquiring defect morphology using 3D scanning technology, and simultaneously collecting corresponding magnetic flux leakage (MF) detection signals; second, simulated corrosion defect data, generated by simulating the MF leakage magnetic field of corrosion defects of different shapes and sizes using the finite element method, producing a large amount of paired data. The database uses structured storage, with each record containing corrosion environment parameters, defect geometric parameters, MF leakage signal data, and tag information. The database module supports data querying, exporting, and incremental updates, providing data support for subsequent modules.

[0010] Corrosion Defect Generation Module: This module, based on an artificial intelligence generative model, expands the scale and diversity of corrosion defect data. Employing a conditional generative adversarial network or diffusion model, it generates high-fidelity corrosion defect morphology data based on corrosion environment parameters and desired defect characteristics. Physical constraints are incorporated into the generative model to ensure that the generated defects conform to corrosion mechanisms, such as depth distribution, morphological continuity, and corrosion propagation patterns. After authenticity verification, the generated defect data is integrated into the corrosion database, effectively addressing the problem of insufficient experimental samples.

[0011] Magnetic flux leakage signal inversion module: This module is used to quantitatively invert the geometric parameters and morphology of corrosion defects from measured magnetic flux leakage signals. The module is built based on a deep learning model, employing an encoder-decoder architecture. The input is a triaxial magnetic flux leakage signal, and the output is the depth, width, volume, and two-dimensional depth matrix of the corrosion defect. During model training, a physical constraint loss function for magnetic flux leakage detection is introduced, including magnetic field divergence constraints, attenuation law constraints, and gradient continuity constraints, ensuring that the inversion results conform to the physical mechanism of the magnetic field. The module is trained using paired data from a corrosion database and calibrated with real data to achieve high-precision inversion.

[0012] Corrosion Defect Identification Module: This module automatically classifies and identifies corrosion defects obtained through inversion. Built on a convolutional neural network, the module takes the depth matrix or geometric parameters of the corrosion defect as input and outputs a defect type label, such as uniform corrosion, pitting corrosion, or crack corrosion. The module also outputs the severity level of the defect, classifying it into minor, moderate, and severe levels according to standard specifications. The identification results can be used for subsequent risk assessment.

[0013] Corrosion Risk Assessment Module: This module assesses the safety risks of pipelines in their current state based on the geometric parameters and identification results of corrosion defects. The module incorporates various risk assessment models, such as the ASME B31G standard based on residual strength evaluation and a failure pressure calculation model based on fracture mechanics. Based on defect depth, length, shape, and pipeline material parameters, it calculates the pipeline's residual strength and maximum allowable pressure, thereby assessing the pipeline's failure probability and remaining life. The assessment results are output in the form of risk level or safety index, providing a basis for pipeline maintenance decisions.

[0014] The above modules are interconnected through data interfaces to form a complete intelligent detection system. The system's workflow is as follows: First, a corrosion database module collects and stores real and simulated corrosion defect data to build a basic dataset. Then, a corrosion defect generation module expands the dataset, generating more diverse defect samples and enriching the database. Next, when performing magnetic flux leakage (MF) testing on the pipeline, the collected MF signals are input to the MF signal inversion module. After preprocessing and model inference, the geometric parameters and morphology of the defects are obtained. Subsequently, the corrosion defect identification module classifies and grades the defects based on the inversion results. Finally, the corrosion risk assessment module combines defect parameters and pipeline information to conduct a safety assessment and output an assessment report. The entire process achieves full automation from data acquisition to risk warning.

[0015] This invention achieves technological innovation in the following aspects: First, an integrated intelligent detection system was constructed, which organically combines five functional modules: corrosion database, defect generation, signal inversion, defect identification, and risk assessment, forming a complete detection and evaluation chain, overcoming the problems of fragmented links and data silos in existing technologies.

[0016] Second, a physical constraint generation model was introduced into the corrosion defect generation module to ensure that the generated defect morphology conforms to the corrosion mechanism, thereby improving the authenticity and usability of the generated data and effectively solving the problem of insufficient training data.

[0017] Third, a physical constraint loss function was designed in the leakage magnetic signal inversion module, which integrates physical knowledge such as magnetic field divergence and attenuation law into the deep learning model, so that the inversion results not only conform to the data distribution but also satisfy the physical law, significantly improving the inversion accuracy and reliability.

[0018] Fourth, through the collaborative work of multiple modules, the entire corrosion detection process is automated, reducing manual intervention and improving detection efficiency and consistency.

[0019] Advantages of this invention: This system achieves deep integration of corrosion detection data and artificial intelligence algorithms, addressing the data scarcity problem through a database and generative models, providing ample high-quality training samples for intelligent detection. It significantly improves the accuracy of corrosion defect identification. The introduction of physical constraints makes the inversion model more robust, accurately predicting the depth, volume, and morphology of complex corrosion pits, with an accuracy rate more than 20% higher than traditional methods. It achieves automation and intelligence in corrosion detection. The system processes signals from input to risk assessment automatically, eliminating the need for manual interpretation, greatly improving detection efficiency and reducing human error. It enhances pipeline safety assessment capabilities. Through precise defect parameters and a professional risk assessment model, the system can scientifically assess the remaining strength and lifespan of pipelines, providing decision support for pipeline operation and maintenance, and effectively preventing corrosion leakage accidents. The system has good scalability. The database module supports continuous updates, the generative module can continuously generate new samples, and the inversion and identification models can be continuously optimized through incremental learning to adapt to different pipeline materials and environmental conditions. Attached Figure Description

[0020] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments: Figure 1 This is a block diagram of the overall structure of the intelligent detection system for pipeline corrosion defects based on physical constraint artificial intelligence as described in this invention, showing the five main modules and their interrelationships. Figure 2 The diagram shows the data structure of the erosion database module, illustrating the storage formats of real and simulated data. Figure 3 The flowchart for the corrosion defect generation module illustrates the process from condition input to data verification. Figure 4 The diagram shows the model structure of the magnetic flux leakage signal inversion module, illustrating the encoder-decoder architecture and the introduction of the physical constraint loss function. Figure 5 This diagram illustrates the classification results of the corrosion defect identification module, showing the identification output for different defect types. Figure 6 The output interface for the corrosion risk assessment module displays key indicators such as residual strength and failure probability. Detailed Implementation

[0021] The present invention will be further explained below with reference to specific implementation schemes, but it is not limited to the present invention. The structures, proportions, sizes, etc. shown in the accompanying drawings are only used to complement the content disclosed in the specification, so as to enable those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modification of the structure, change of the proportion relationship or adjustment of the size, without affecting the effect and purpose that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. These embodiments are merely illustrative and do not constitute a limitation of the invention.

[0023] Example 1: System Hardware Configuration and Data Acquisition The intelligent detection system in this embodiment is deployed in a pipeline inspection center. The hardware includes a magnetic flux leakage detector, a data acquisition workstation, a high-performance computing server, and storage devices. The magnetic flux leakage detector is a conventional in-pipe detector, equipped with a permanent magnet magnetization unit and a triaxial Hall sensor array. The sensor spacing is 1.5 mm, and the lift-off height is adjustable. The data acquisition workstation is responsible for receiving the raw magnetic flux leakage signals uploaded by the detector and performing preliminary format conversion and quality checks. The high-performance computing server is configured with multiple NVIDIA Tesla V100 GPUs for running the training and inference of deep learning models. The storage devices use a RAID disk array to store the corrosion database and system operation logs.

[0024] First, real corrosion defect data were collected. Two hundred samples were prepared using X70 pipeline steel, each measuring 300 mm × 150 mm × 12 mm. A soil corrosion simulation tank, a stray current corrosion platform, and an atmospheric corrosion chamber were constructed in the laboratory. Different morphologies of corrosion defects were obtained by adjusting environmental parameters. After corrosion of each sample, the defect morphology was scanned using a Keyence VR-5000 three-dimensional white light interferometer to obtain a 256 × 256 pixel depth matrix. Then, a magnetic flux leakage scanning platform was used to acquire the triaxial magnetic flux leakage signal of each sample, with a scan step of 1 mm and a sampling frequency of 1000 Hz, obtaining a signal sequence of 500 points. The environmental parameters, defect depth matrix, and magnetic flux leakage signal of each sample were stored in a corrosion database. A total of 200 sets of real samples were obtained.

[0025] To expand the dataset, COMSOL Multiphysics was used for finite element simulations. A parameterized corrosion defect model was established, with defect depths ranging from 1 to 8 mm and diameters from 5 to 30 mm, including circular, elliptical, rectangular, and irregular shapes. A total of 5000 simulation samples were generated, each containing defect parameters and the corresponding three-channel leakage magnetic field signal. The simulation data and real data were stored in the database after being formatted in the same way.

[0026] Example 2: Implementation of the Corrosion Defect Generation Module This embodiment employs a conditional diffusion model to construct a corrosion defect generation module. The model takes random noise and a conditional vector as input and outputs a 256×256 corrosion depth matrix. The conditional vector includes the desired defect depth, width, volume, and corrosion environment type. During model training, real samples from a corrosion database are used as training data. The loss function includes diffusion loss and physical constraint loss. The physical constraint loss comprises three terms: depth range constraint, ensuring the maximum depth of the generated defect is within the statistical range of real samples; morphological continuity constraint, calculating the total variation of the depth gradient; and corrosion law constraint, verifying whether the generated sample conforms to corrosion kinetics by fitting the depth-time relationship of real samples. After 200 epochs of training, the model can generate realistic corrosion defects. For each generated sample, after statistical distribution verification and morphological feature verification, qualified samples are selected and added to the database. This embodiment generates a total of 3000 qualified samples, expanding the database size to 8200 groups.

[0027] Example 3: Implementation of the leakage magnetic signal inversion module The magnetic flux leakage signal inversion module employs an encoder-decoder architecture. The encoder consists of five convolutional blocks, each containing a convolutional layer, batch normalization, and ReLU activation, progressively compressing the input signal into a 256-dimensional feature vector. The decoder consists of four deconvolutional blocks and two branches: one branch is a fully connected layer that outputs three scalars: depth, width, and volume; the other branch outputs a 64×64 depth matrix. Physical constraint losses are introduced during training, including divergence constraints, attenuation law constraints, and gradient continuity constraints. The divergence constraint calculates the divergence of the predicted magnetic flux leakage signal using finite difference calculations, with its absolute value used as the loss. The attenuation law constraint calculates the difference between the signal attenuation rate and the theoretical value by changing the lift-off height. The gradient continuity constraint calculates the sum of squares of the second derivatives of the signal. The total loss is the sum of the mean squared error loss and the physical constraint loss, with a balance coefficient λ set to 0.2. The model was trained on 8200 datasets using the Adam optimizer, with an initial learning rate of 1e-4, a batch size of 32, and 150 epochs. The average relative error for depth prediction on the test set was 7.8%, the width error was 10.5%, and the volume error was 13.2%.

[0028] Example 4: Implementation of the Corrosion Defect Identification Module The corrosion defect identification module uses ResNet-50 as the base network, with a 64×64 depth matrix as input and defect type and severity level as output. Defect types are divided into four categories: uniform corrosion, pitting corrosion, crack corrosion, and exfoliation corrosion. Severity levels are classified according to the ratio of defect depth to wall thickness: Level 1 (<10%), Level 2 (10%~25%), Level 3 (25%~50%), and Level 4 (>50%). Training data consists of real and generated samples from a corrosion database, manually labeled. Using the cross-entropy loss function, the classification accuracy reaches 92.5% after 50 epochs of training.

[0029] Example 5: Implementation of the Corrosion Risk Assessment Module The corrosion risk assessment module integrates the ASME B31G standard algorithm and a fracture mechanics-based assessment model. Input parameters include defect depth, length, width, pipe diameter, wall thickness, and material strength. First, the projected area and residual strength of the defect are calculated. Then, the appropriate failure pressure formula is selected based on the defect type. For pitting corrosion, a simplified method from ASME B31G is used; for crack-like defects, a fracture mechanics model is employed. After calculating the failure pressure, it is compared with the maximum allowable operating pressure of the pipeline to determine the safety margin. Simultaneously, the failure probability is calculated based on Monte Carlo simulation. The final output is a risk assessment report, including the risk level (low, medium, high) and recommended maintenance time.

[0030] Example 6: System Integration and Field Testing The five modules mentioned above were integrated into a unified software platform, and a graphical user interface was developed. Field testing was conducted on an oil pipeline in western China, selecting a 10 km section and using a magnetic flux leakage detector to collect data. The collected signals were uploaded to the system in real time, and the system automatically completed signal preprocessing, inversion, identification, and evaluation, with the entire process taking approximately 30 minutes. Twenty-seven corrosion defects were detected. Excavation verification showed that the system's average depth error was 9.2%, the identification accuracy was 89%, and the risk assessment results were consistent with expert review. Field testing verified the system's effectiveness and reliability.

[0031] Matters not covered in this invention are common knowledge.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent detection system for pipeline corrosion defects based on physical constraint artificial intelligence, characterized in that: Includes the following modules: Corrosion Database Module: This module is used to store and manage data related to pipeline corrosion defects. The database contains two types of data: one is real corrosion defect data, which is obtained by preparing corrosion samples under different environmental conditions through laboratory accelerated corrosion experiments, using three-dimensional scanning technology to obtain the defect morphology, and simultaneously collecting the corresponding magnetic flux leakage detection signals; the other is simulated corrosion defect data, which is generated by simulating the magnetic flux leakage of corrosion defects of different shapes and sizes through the finite element method, generating a large amount of paired data. The database uses structured storage, and each record contains corrosion environment parameters, defect geometric parameters, magnetic flux leakage signal data, and tag information; The database module supports data querying, exporting, and incremental updates, providing data support for subsequent modules. Corrosion Defect Generation Module: This module is based on an artificial intelligence generation model to expand the scale and diversity of corrosion defect data. The module uses a conditional generative adversarial network or diffusion model to generate high-fidelity corrosion defect morphology data based on corrosion environment parameters and desired defect characteristics. A physical constraint mechanism is introduced into the generation model to ensure that the generated defects conform to the corrosion mechanism. After the generated defect data is verified for authenticity, it is incorporated into the corrosion database, effectively solving the problem of insufficient experimental samples. Magnetic flux leakage signal inversion module: This module is used to quantitatively invert the geometric parameters and morphology of corrosion defects from measured magnetic flux leakage signals. The module is built based on a deep learning model and adopts an encoder-decoder architecture. The input is a triaxial magnetic flux leakage signal, and the output is the depth, width, volume, and two-dimensional depth matrix of the corrosion defect. During the model training process, a physical constraint loss function for magnetic flux leakage detection is introduced, including magnetic field divergence constraints, attenuation law constraints, and gradient continuity constraints, to ensure that the inversion results conform to the physical mechanism of the magnetic field. The module is trained using paired data in the corrosion database and calibrated with real data to achieve high-precision inversion. Corrosion Defect Recognition Module: This module is used to automatically classify and identify corrosion defects obtained from inversion. The module is built based on a convolutional neural network. The input is the depth matrix or geometric parameters of the corrosion defect, and the output is the type label of the defect, such as uniform corrosion, pitting corrosion, crack corrosion, etc. The module also outputs the severity level of the defect, classifying the defect into different levels such as minor, moderate, and severe according to standard specifications. The recognition results can be used for subsequent risk assessment. Corrosion Risk Assessment Module: This module assesses the safety risks of the pipeline in its current state based on the geometric parameters and identification results of corrosion defects. The module incorporates multiple risk assessment models; based on defect depth, length, shape, and pipeline material parameters, it calculates the pipeline's remaining strength and maximum allowable pressure, thereby assessing the pipeline's failure probability and remaining life; the assessment results are output in the form of risk level or safety index, providing a basis for pipeline maintenance decisions; The above modules are interconnected through data interfaces to form a complete intelligent detection system; the system's workflow is as follows: First, the corrosion database module collects and stores real and simulated corrosion defect data to build a basic dataset. Then, the corrosion defect generation module expands the dataset to generate more diverse defect samples and enrich the database. Next, when the pipeline to be inspected is subjected to magnetic flux leakage detection, the collected magnetic flux leakage signal is input into the magnetic flux leakage signal inversion module. After preprocessing and model inference, the geometric parameters and morphology of the defects are obtained. Subsequently, the corrosion defect identification module classifies and grades the defects based on the inversion results; Finally, the corrosion risk assessment module combines defect parameters and pipeline information to conduct a safety assessment and output an assessment report; the entire process achieves full automation from data acquisition to risk warning.

2. The intelligent detection system for pipeline corrosion defects based on physical constraint artificial intelligence according to claim 1, characterized in that: The intelligent pipeline corrosion defect detection system based on physical constraint artificial intelligence includes the following: First, an integrated intelligent detection system was constructed, which organically combines five functional modules: corrosion database, defect generation, signal inversion, defect identification, and risk assessment, forming a complete detection and assessment chain, overcoming the problems of fragmented links and data silos in existing technologies; Second, a physical constraint generation model was introduced into the corrosion defect generation module to ensure that the generated defect morphology conforms to the corrosion mechanism, thereby improving the authenticity and usability of the generated data and effectively solving the problem of insufficient training data. Third, a physical constraint loss function was designed in the leakage magnetic signal inversion module, which integrates physical knowledge such as magnetic field divergence and attenuation law into the deep learning model, so that the inversion results not only conform to the data distribution but also satisfy the physical law, significantly improving the inversion accuracy and reliability. Fourth, through the collaborative work of multiple modules, the entire corrosion detection process is automated, reducing manual intervention and improving detection efficiency and consistency.

Images