A petrochemical safety operation and maintenance digitalization intelligent management method and system

Abstract

This invention discloses a digital intelligent management method and system for safe operation and maintenance in the petrochemical industry. Addressing the industry pain points of traditional magnetic flux leakage (MF) detection, which relies on manual experience leading to low efficiency and accuracy, and the information silos formed by the disconnect between detection and maintenance, this invention intelligently converts MF signals into images and performs automatic model recognition. This achieves precise quantitative analysis of defect type, depth, and location information, significantly improving the accuracy and efficiency of defect identification. Simultaneously, this method constructs a closed-loop data linkage from defect identification and quantitative analysis to maintenance work order generation and feedback recording. This not only achieves transparent management and efficient collaboration in the operation and maintenance process but also provides data support for subsequent strategy optimization through the accumulation of maintenance logs, effectively eliminating safety hazards and greatly improving the intelligence level and safety of petrochemical pipeline operation and maintenance management.

Description

Technical Field

[0001] This invention belongs to the field of digital and intelligent operation and maintenance management technology in the petrochemical industry, specifically relating to a digital and intelligent management method and system for safe operation and maintenance in the petrochemical industry. Background Technology

[0002] Petrochemical pipelines operate under complex conditions of high pressure, high temperature, and corrosive media for extended periods, making them highly susceptible to defects such as corrosion and cracks. These defects are major hidden dangers that can lead to leaks, explosions, and other serious safety accidents, posing a serious threat to people's lives, property, and the ecological environment. Therefore, efficient and accurate non-destructive testing and intelligent operation and maintenance management of in-service pipelines have become the core of safety work in the petrochemical industry.

[0003] At present, magnetic flux leakage detection is a mature and widely used non-destructive testing technology. Its basic principle is to use an excitation source to magnetize the pipeline to a saturation state. When there is a defect in the pipeline, the magnetic permeability at the defect changes, causing magnetic field lines to overflow from the pipeline surface and form a magnetic leakage field. By capturing this signal through a sensor array, the defect can be detected. However, traditional magnetic flux leakage detection and subsequent operation and maintenance management mainly have the following technical bottlenecks: (1) Signal interpretation depends on human experience, which is highly subjective and inefficient. Traditional magnetic flux leakage detection mainly relies on manual analysis of one-dimensional time domain signal curves or by converting the signal into two dimensions. The images are then manually interpreted. This method is not only inefficient, but the interpretation results also heavily rely on the personal experience of experts. For minor defects or defects in complex structures (such as welds), it is easy to make misjudgments and miss detections. (2) The traditional detection process stops at issuing a detection report. The subsequent maintenance decision-making, work order dispatch, and result feedback are often independent of each other and lack effective data linkage. This information silo phenomenon leads to a long and opaque management chain from "discovering defects" to "completing maintenance". It is difficult to eliminate safety hazards in a timely manner and cannot form effective knowledge accumulation to guide future operation and maintenance strategies.

[0004] Therefore, based on the aforementioned shortcomings, there is an urgent need to provide a comprehensive method that can integrate intelligent identification, accurate quantitative analysis, and closed-loop operation and maintenance management to solve the problems of subjective experience-based identification, inaccurate defect identification, and low management efficiency in existing technologies. Summary of the Invention

[0005] The purpose of this invention is to provide a digital intelligent management method and system for safe operation and maintenance in the petrochemical industry, in order to solve the problems of existing technologies such as reliance on subjective experience for identification, inaccurate defect identification, and low management efficiency.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a digital and intelligent management method for safe operation and maintenance in the petrochemical industry is provided, including: The magnetic flux leakage detection signal of the petrochemical pipeline is acquired and converted into a magnetic flux leakage image; Using a pre-trained defect classification model, defects are identified in the magnetic flux leakage image to segment defect image blocks in the magnetic flux leakage image and to determine the defect type of the petrochemical pipeline. Based on the defect image block, the magnetic flux leakage defect signal at the defect location of the petrochemical pipeline is segmented from the magnetic flux leakage detection signal; Based on the magnetic flux leakage defect signal, a magnetic flux leakage signal curve at the defect center at the defect location is generated; Based on the magnetic flux leakage signal curve at the defect center, the defect depth and the location of the defect in the wall thickness direction of the petrochemical pipeline are determined. The defect type, defect depth, and defect location information of the petrochemical pipeline are pushed to the monitoring platform, and a corresponding operation and maintenance work order is generated and sent to the operation and maintenance terminal. The operation and maintenance work order contains the operation and maintenance strategy. It receives maintenance feedback information uploaded from the maintenance terminal and records maintenance work orders and maintenance feedback information as maintenance logs for storage, so as to complete the intelligent management of maintenance of petrochemical pipelines after storage.

[0007] Based on the above disclosure, this invention converts magnetic flux leakage detection signals into magnetic flux leakage images and uses a pre-trained defect classification model for automatic identification and segmentation. This allows for precise location of defect regions and determination of defect types from the images, overcoming the technical bottleneck of misjudgment and missed detection of minute defects and defects in complex structures when manually analyzing one-dimensional time-domain signals or two-dimensional images. This significantly improves the accuracy and efficiency of defect identification. Furthermore, after identifying the defect, this invention automatically generates a magnetic flux leakage signal curve at the defect center based on the magnetic flux leakage defect signal corresponding to the defect image block, and accurately quantifies the depth of the defect and its position in the wall thickness direction. The invention first identifies the location information, then pushes this key data along with the defect type to the monitoring platform, and automatically generates a work order containing the operation and maintenance strategy. In this way, the invention achieves full-chain data linkage of "identification-analysis-decision-execution". Finally, the invention receives operation and maintenance feedback and records it along with the operation and maintenance work order as an operation and maintenance log for storage. Based on this, it not only ensures the closed loop of hidden danger management, but also provides data support for subsequent knowledge accumulation and operation and maintenance strategy optimization, which greatly improves the transparency and efficiency of the entire operation and maintenance management chain, and ultimately effectively ensures the operational safety of petrochemical pipelines. Therefore, the invention is very suitable for large-scale application and promotion.

[0008] In one possible design, the leakage magnetic field detection signal is a triaxial signal, and the three axes are the axial direction, the circumferential direction and the radial direction, respectively; The process of generating a magnetic flux leakage signal curve at the defect center location based on the magnetic flux leakage defect signal includes: From the leakage magnetic defect signal, the leakage magnetic signal in the axial direction or the radial direction is extracted as the target signal; From the target signal, determine the signal point corresponding to the defect center, and use it as the defect center point; Along the axial direction, all signal points containing the defect center point are extracted from the target signal, and the defect center leakage magnetic signal curve is generated using all the extracted signal points.

[0009] In one possible design, based on the magnetic flux leakage signal curve at the defect center, the defect depth and the location of the defect in the pipeline along the pipe wall thickness are determined, including: From the leakage magnetic field signal curve at the defect center, the signal point corresponding to the maximum peak and the signal point corresponding to the minimum valley are determined, and are respectively used as the peak point and valley point. The difference between the signal amplitude at the peak and the signal amplitude at the trough is calculated to obtain the peak-trough difference. Calculate the signal spacing between the peak and trough points; The steepness of the leakage magnetic signal curve at the defect center is calculated using the peak-to-valley difference and the signal spacing. Based on the steepness, the depth of the defect and the location information of the defect in the wall thickness direction are determined.

[0010] In one possible design, the steepness of the leakage magnetic signal curve at the defect center is calculated using the peak-to-valley difference and the signal spacing, including: Obtain the scaling factor and calculate the product between the scaling factor and the signal spacing; The ratio between the peak-to-valley difference and the product is taken as the steepness; Among them, based on the steepness, the location information of the defect in the wall thickness direction is determined, including: Determine whether the steepness is greater than a first threshold; If not, then determine whether the steepness is less than the second threshold; otherwise, determine that the location of the defect is on the outer wall of the petrochemical pipeline, wherein the second threshold is less than the first threshold. If so, the location information of the defect is determined to be on the inner wall of the petrochemical pipeline. When the steepness is between the second threshold and the first threshold, the location information of the defect is determined to be between the outer wall and the inner wall of the petrochemical pipeline.

[0011] In one possible design, the defect depth is determined based on the steepness, including: The depth of the defect is calculated using the following formula; ; In the formula, Indicates steepness. Indicates the depth of the defect. These represent the position compensation factor and the amplitude scaling factor, respectively.

[0012] In one possible design, the defect classification model is trained in the following manner: Obtain the magnetic flux leakage defect signal set corresponding to different defect types; Based on each set of magnetic flux leakage defects, a priori noise signals for each defect type are generated. The priori noise signal for any defect type contains the priori distribution information of the magnetic flux leakage defect signal set corresponding to that defect type. Using each magnetic flux leakage defect signal set and the corresponding prior noise signal, a generated sample dataset for each defect type is generated. A data filtering algorithm based on quality error is used to filter out the optimal generated sample set for each defect type from the generated sample datasets corresponding to each defect type. An enhanced training dataset is constructed using the magnetic flux leakage defect signal sets of various defect types and the optimal generated sample set. The enhanced training dataset is converted into an enhanced magnetic flux leakage image set, and the defect classification model is trained using the enhanced magnetic flux leakage image set to obtain the trained defect classification model.

[0013] In one possible design, prior noise signals for each defect type are generated based on various leakage magnetic defect signal sets, including: For any set of magnetic flux leakage defect signals, feature extraction is performed on the set of magnetic flux leakage defect signals to obtain a set of feature samples, wherein the feature data in the set of feature samples corresponds one-to-one with the magnetic flux leakage defect signals in the set of magnetic flux leakage defect signals. Clustering is performed on the feature data in the feature sample set to obtain multiple feature clusters; Calculate the mean and covariance matrix of each feature cluster; Based on the mean and covariance matrix of each feature cluster, the prior distribution of each feature cluster is calculated. By utilizing the prior distribution of each feature cluster, a multivariate hybrid prior distribution model of any magnetic flux leakage defect signal set is constructed. Using a multivariate hybrid prior distribution model, a prior noise signal corresponding to the defect type of any magnetic flux leakage defect signal set is generated.

[0014] In one possible design, a data filtering algorithm based on quality error is used to select the optimal generated sample set for each defect type from the generated sample datasets corresponding to each defect type, including: For any generated sample dataset, a magnetic flux leakage test signal set corresponding to the defect type of the generated sample dataset is generated based on the magnetic flux leakage defect signal set corresponding to the defect type of the generated sample dataset. Perform the j-th data filtering from any of the generated sample datasets to obtain the j-th sample subset; Using the j-th sample subset, train the quality enhancement model to obtain the trained quality enhancement model; The j-th sample subset is input into the trained quality enhancement model to obtain the enhanced dataset; Based on the j-th sample subset and the augmented dataset, calculate the error threshold corresponding to the j-th sample subset; The target test signal set is input into the trained quality enhancement model to obtain the test enhancement signal set, wherein the target test signal set is the leakage magnetic field test signal set; Calculate the quality error between each target test signal in the target test signal set and the corresponding test enhancement signal in the test enhancement signal set, and count the number of target test signals whose quality error is greater than the error threshold, as the number of misjudgments; The number of misjudgments and the sample subset obtained from the j-th data filtering are linked and recorded; Increment j by 1 and perform the j-th data filtering again from any of the generated sample datasets until j equals J, resulting in J misclassifications. Here, the generated sample data in each sample subset is different, the initial value of j is 1, and J is the set maximum number of filtering times. From the J misclassification counts, select the sample subset associated with the smallest misclassification count, and use the selected sample subset as the optimal generated sample set for the defect type corresponding to any generated sample dataset.

[0015] Secondly, a digital intelligent management system for safe operation and maintenance in the petrochemical industry is provided, including: The signal conversion unit is used to acquire the magnetic flux leakage detection signal of the petrochemical pipeline and convert the magnetic flux leakage detection signal into a magnetic flux leakage image; The defect identification unit is used to identify defects in the magnetic flux leakage image using a pre-trained defect classification model, so as to segment defect image blocks in the magnetic flux leakage image and determine the defect type of the petrochemical pipeline. The signal segmentation unit is used to segment the magnetic flux leakage defect signal at the defect location of the petrochemical pipeline from the magnetic flux leakage detection signal based on the defect image block; The defect parameter calculation unit is used to generate a defect center magnetic flux leakage signal curve at the defect location based on the magnetic flux leakage defect signal. The defect parameter calculation unit is also used to determine the defect depth of the petrochemical pipeline and the location information of the defect in the wall thickness direction of the petrochemical pipeline based on the leakage magnetic signal curve of the defect center. The operation and maintenance unit is used to push the defect type, defect depth and defect location information of the petrochemical pipeline to the monitoring platform, and generate corresponding operation and maintenance work orders to send to the operation and maintenance terminal. The operation and maintenance work orders contain operation and maintenance strategies. The management unit is used to receive operation and maintenance feedback information uploaded by the operation and maintenance terminal, and to record operation and maintenance work orders and operation and maintenance feedback information as operation and maintenance logs for storage, so as to complete the intelligent management of operation and maintenance of petrochemical pipelines after storage.

[0016] Thirdly, a petrochemical safety operation and maintenance digital intelligent management device is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the petrochemical safety operation and maintenance digital intelligent management method as described in the first aspect or any possible design in the first aspect.

[0017] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the intelligent management method for safe operation and maintenance of petrochemical plants as described in the first aspect or any possible design of the first aspect.

[0018] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the intelligent management method for safe operation and maintenance of petrochemical plants as described in the first aspect or any possible design of the first aspect.

[0019] Beneficial effects: (1) This invention converts magnetic flux leakage detection signals into magnetic flux leakage images and uses a pre-trained defect classification model for automatic identification and segmentation. This allows for accurate location of defect regions and determination of defect types from images, solving the technical bottleneck of misjudgment and missed detection of small defects and defects in complex structures when manually analyzing one-dimensional time-domain signals or two-dimensional images. This significantly improves the accuracy and efficiency of defect identification. Furthermore, after identifying defects, this invention automatically generates a magnetic flux leakage signal curve at the defect center based on the magnetic flux leakage defect signal corresponding to the defect image block, and accurately quantifies the depth of the defect and its location information in the wall thickness direction. Then, these key data points, along with the defect types, are pushed to the monitoring platform, automatically generating work orders containing operation and maintenance strategies. In this way, the present invention achieves full-chain data linkage of "identification-analysis-decision-execution". Finally, the present invention receives operation and maintenance feedback and records it together with the operation and maintenance work orders as operation and maintenance logs for storage. Based on this, not only is a closed loop of hidden danger management ensured, but data support is also provided for subsequent knowledge accumulation and operation and maintenance strategy optimization, which greatly improves the transparency and efficiency of the entire operation and maintenance management chain and ultimately effectively ensures the operational safety of petrochemical pipelines. Therefore, the present invention is very suitable for large-scale application and promotion. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the steps of the intelligent management method for safe operation and maintenance in the petrochemical industry provided in this embodiment of the invention; Figure 2 This is a structural diagram of the intelligent management system for safe operation and maintenance of petrochemical plants provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0022] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0023] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0024] Example: See Figure 1 As shown, the intelligent management method for safe operation and maintenance of petrochemical plants provided in this embodiment can be executed by, but is not limited to, computer equipment with certain computing resources, such as servers, edge computers, personal computers (PCs, which are multi-purpose computers of a size, price and performance suitable for personal use; desktop computers, laptops to mini-laptops and tablets and ultrabooks are all personal computers). It is understood that the aforementioned execution subject does not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, the steps S1 to S7 below.

[0025] S1. Acquire the magnetic flux leakage (MF) detection signal of the petrochemical pipeline and convert the MF detection signal into a MF image. In a specific implementation, the MF detection signal is exemplified as a triaxial signal, wherein the three axes are the axial direction, the circumferential direction, and the radial direction. The axial direction is the length direction of the pipeline, which is the forward direction of the MF detector inside the pipeline. The axial direction is the direction that circles the pipeline and is perpendicular to the axial direction. The radial direction is the direction that is perpendicular to the pipeline wall, which is the thickness direction of the pipeline. Thus, after the MF detection signal is acquired based on the MF detector, it can be converted into a MF image for subsequent image recognition, thereby identifying the type of defect on the pipeline and the defect image block in the MF image.

[0026] In practical applications, magnetic flux leakage detection signals often contain a large amount of noise signals, especially low-frequency quasi-DC noise signals, whose dynamic range is even comparable to the amplitude of the useful magnetic flux leakage signal. Therefore, this seriously affects the effect of subsequent intelligent analysis of magnetic flux leakage data. Thus, in this embodiment, before converting it into a magnetic flux leakage image, it is first enhanced to improve the signal quality.

[0027] The signal enhancement process may include, but is not limited to, the following: Step 1: Extract the axial and radial signals from the magnetic flux leakage detection signal to serve as the first signal and the second signal, respectively. The first signal contains a signal sequence corresponding to multiple channels, and each channel represents a sensor in the magnetic flux leakage detector. The second signal is similar and will not be elaborated here.

[0028] The second step is to perform DC component removal processing on the signal sequences corresponding to each channel in the first and second signals to obtain the first preprocessed signal sequence corresponding to each channel in the first signal and the second preprocessed signal sequence corresponding to each channel in the second signal. In specific implementation, for any signal sequence in the first signal, the mean of all sampling points in the signal sequence is calculated, and then the mean is subtracted from each sampling point in the signal sequence to obtain the first preprocessed signal sequence corresponding to the signal sequence. Of course, the processing method for other signal sequences is the same, and will not be elaborated here.

[0029] Step 3: Calculate the first initial quality enhancement coefficient corresponding to each first preprocessed signal sequence, and the second initial quality enhancement coefficient corresponding to each second preprocessed signal sequence.

[0030] In specific implementation, for any first preprocessed signal sequence, the second preprocessed signal sequence corresponding to the target channel is first selected from each of the second preprocessed signal sequences corresponding to the second signal as the target sequence (where the target channel is the channel corresponding to any first preprocessed signal sequence); then, the partial derivative of the target sequence in the x-direction (x-direction is the axial direction of the pipe) is calculated; finally, the absolute value of the partial derivative is used as the first initial quality enhancement coefficient corresponding to any first preprocessed signal sequence; of course, the processing method for the other first preprocessed signal sequences is the same, and will not be elaborated here.

[0031] Meanwhile, for the second preprocessed signal sequence, the first preprocessed signal sequence with the same channel is found in each of the first preprocessed signal sequences and used as the target sequence. The subsequent processing method is the same as the example above, so it will not be described again in this embodiment.

[0032] Step 3: Correct each first initial quality enhancement coefficient and each second initial quality enhancement coefficient to obtain the first quality enhancement coefficient corresponding to each first preprocessed signal sequence and the second quality enhancement coefficient corresponding to each second preprocessed signal sequence. For any initial quality enhancement coefficient, first determine whether the initial quality enhancement coefficient is greater than the maximum enhancement coefficient. If so, select the largest initial quality enhancement coefficient from among the first initial quality enhancement coefficients as the correction parameter. Then, calculate the ratio between the initial quality enhancement coefficient and the correction parameter. Finally, multiply the ratio by the maximum enhancement coefficient to obtain the corresponding first quality enhancement coefficient. If any initial quality enhancement coefficient is less than or equal to the maximum enhancement coefficient, no correction is required, and it is directly used as the first quality enhancement coefficient.

[0033] Of course, the correction process for the other first initial mass enhancement coefficients and the other second initial mass enhancement coefficients is the same, and will not be repeated here.

[0034] Step 4: Multiply each first preprocessed signal sequence by the first quality enhancement coefficient corresponding to each first preprocessed signal sequence to obtain several enhanced first signal sequences, and multiply each second preprocessed signal sequence by the second quality enhancement coefficient corresponding to each second preprocessed signal sequence to obtain several enhanced second signal sequences.

[0035] Step 5: Using several enhanced first signal sequences, an enhanced axial magnetic flux leakage signal is formed, and using several enhanced second signal sequences, an enhanced radial magnetic flux leakage signal is formed. The magnetic flux leakage signal in the axial direction of the magnetic flux leakage detection signal, as well as the enhanced axial magnetic flux leakage signal and the enhanced radial magnetic flux leakage signal, are used to form an enhanced magnetic flux leakage detection signal.

[0036] Thus, after the aforementioned steps are used to enhance the magnetic flux leakage detection signal, signal conversion can be performed. The enhanced magnetic flux leakage detection signal is a triaxial signal, and its three axes are as follows: ,in, The following are the axial directions of the enhanced magnetic flux leakage detection signal, respectively. The b-th sampling point within the _ ... The b-th sampling point within the n channels, and the n-th sampling point in the radial direction The b-th sampling point within each channel; thus, the red channel R(a,b) in the RGB image is equal to The green channel G(a,b) is equal to The blue channel B(a,b) is equal to At this point, the mapping between the signal and pixel values ​​is complete. Finally, Using 'b' as the row and column index and filling in the corresponding RGB values, the initial magnetic flux leakage image can be obtained.

[0037] After obtaining the initial magnetic flux leakage image, the pixel values ​​in the initial magnetic flux leakage image are normalized to between 0 and 255 to obtain the magnetic flux leakage image.

[0038] After obtaining the magnetic flux leakage image, image recognition can be performed, as shown in step S2 below.

[0039] S2. Using a pre-trained defect classification model, defects are identified in the magnetic flux leakage image to segment defect image blocks in the magnetic flux leakage image and to determine the defect type of the petrochemical pipeline. In specific implementation, for example, but not limited to, a trained PP-YOLOE model can be used to identify defects in the magnetic flux leakage image, thereby outputting the corresponding defect type (such as corrosion, cracks, etc.). At the same time, the model will also output the bounding box coordinates of the defects in the magnetic flux leakage image. Thus, based on the bounding box coordinates, defect image blocks can be segmented from the magnetic flux leakage image.

[0040] Furthermore, since petrochemical pipelines are in a healthy state for most of the actual operation time, it is difficult to obtain a large number of defect signals. Based on this, this embodiment provides a model training data based on sample augmentation, that is, based on small samples, sample expansion is performed to provide sufficient training data for the training of the PP-YOLOE model.

[0041] Optionally, the training process of the aforementioned defect classification model may be, but is not limited to, the steps S21 to S26 below.

[0042] S21. Obtain the magnetic flux leakage defect signal set corresponding to different defect types; In this embodiment, the magnetic flux leakage defect signal set with defect types such as corrosion and cracks can be obtained from the historical inspection data of petrochemical pipelines and used as the original sample dataset; Then, the original sample dataset can be used to expand the sample, as shown in the following steps S22 to S25.

[0043] S21. Based on each set of magnetic flux leakage defects, generate prior noise signals for each defect type. The prior noise signal for any defect type contains prior distribution information of the corresponding set of magnetic flux leakage defects. In practice, traditional sample augmentation techniques typically rely on random noise, which does not contain prior distribution information of the original sample data. Therefore, it cannot provide prior knowledge for subsequent sample learning, and thus cannot help the generated samples approximate the original distribution during training. Therefore, based on the aforementioned shortcomings, this embodiment utilizes each set of magnetic flux leakage defects to generate prior noise signals containing prior distribution information of the original sample data, and uses this to generate samples, thereby making the subsequently generated samples closer to the original distribution.

[0044] Taking any set of leakage magnetic defect signals as an example, the generation process of its prior noise signal can be described, but is not limited to the steps S21a to S21f below.

[0045] S21a. For any set of magnetic flux leakage defect signals, feature extraction is performed on the set of magnetic flux leakage defect signals to obtain a feature sample set, wherein the feature data in the feature sample set corresponds one-to-one with the magnetic flux leakage defect signals in the set of magnetic flux leakage defect signals; in specific implementation, for any magnetic flux leakage defect signal in any set of magnetic flux leakage defect signals, features such as amplitude, mean, minimum value, and maximum value can be extracted to form corresponding feature data. In this way, after the feature extraction of all magnetic flux leakage defect signals is completed, a feature sample set can be formed.

[0046] After obtaining the feature sample set, feature clustering can be performed, as shown in step S21b below.

[0047] S21b. Cluster the feature data in the feature sample set to obtain multiple feature clusters. In specific implementation, examples, but not limited to, using the K-means clustering algorithm, can be used to cluster the feature data in the feature sample set to obtain multiple feature clusters. The number of clusters can be determined by the DBI index, which is a commonly used indicator for evaluating the effectiveness of clustering algorithms. Its calculation principle will not be elaborated here. Of course, the K-means clustering algorithm is also a commonly used clustering technique, and its principle will not be elaborated here either.

[0048] After obtaining multiple feature clusters, the mean and covariance matrix of each feature cluster can be calculated so that the prior distribution of each feature cluster can be constructed based on this. The process is shown in step S21c below.

[0049] S21c. Calculate the mean and covariance matrix of each feature cluster; after calculating the mean and covariance matrix of the feature clusters, the prior distribution can be calculated, as shown in step S21d below.

[0050] S21d. Based on the mean and covariance matrix of each feature cluster, calculate the prior distribution of each feature cluster; in specific implementation, for the m-th feature cluster, for example, but not limited to, the prior distribution of the m-th feature cluster can be calculated according to the following formula.

[0051] ; In the formula, This represents the prior distribution of the m-th feature cluster. Let represent the mean and covariance matrix of the m-th feature cluster, respectively. Denotes the dimension of any feature data in the m-th feature cluster. This indicates the transpose operation. Let m represent any feature data in the m-th feature cluster, where m = 1, 2, ..., M, and M is the total number of feature clusters.

[0052] Thus, based on the aforementioned formula, after calculating the prior distribution of each feature cluster, the multivariate hybrid prior distribution model of any leakage magnetic defect signal set can be constructed using all prior distributions, as shown in step S21e below.

[0053] S21e. Using the prior distribution of each feature cluster, construct a multivariate hybrid prior distribution model for any set of leakage magnetic defect signals; in specific applications, for example, but not limited to, constructing the aforementioned multivariate hybrid prior distribution model according to the following formula.

[0054] ; In the formula, This represents the multivariate mixed prior distribution model. Let be the distribution weight of the m-th feature cluster; in this embodiment, is the ratio between the number of feature data in the m-th feature cluster and the total number of feature data in the feature sample set.

[0055] Thus, based on the aforementioned formula, a multivariate hybrid prior distribution model of the defect type corresponding to any magnetic flux leakage defect signal set can be constructed. Then, based on this distribution model, a prior noise signal of the defect type corresponding to any magnetic flux leakage defect signal set is generated, as shown in step S21f below.

[0056] S21f. Using a multivariate hybrid prior distribution model, generate a prior noise signal for the defect type corresponding to any set of leakage magnetic defect signals; in specific implementation, for example, but not limited to, randomly selecting a vector from the multivariate hybrid prior distribution model as the prior noise signal.

[0057] Thus, through the aforementioned steps S21a to S21f, prior noise signals corresponding to each set of magnetic flux leakage defects can be generated. In this way, the prior noise signals not only have randomness, but also embed the rough characteristics of the set of magnetic flux leakage defects, which can provide prior knowledge for subsequent sample generation.

[0058] After obtaining the prior noise signal corresponding to each leakage magnetic defect signal set, sample expansion can be performed, that is, generating a new sample for each defect type. The process is shown in step S22 below.

[0059] S22. Using each set of magnetic flux leakage defect signals and the prior noise signals corresponding to each set of magnetic flux leakage defect signals, generate a sample dataset corresponding to each defect type. In specific implementation, for example, but not limited to, a generative adversarial network can be used to generate the sample dataset corresponding to each defect type. That is, the prior noise signals corresponding to each set of magnetic flux leakage defect signals and the defect labels of each set of magnetic flux leakage defect signals are input into the generative adversarial network to generate the sample dataset corresponding to each defect type.

[0060] Furthermore, the specific process by which the generator in the generative adversarial network outputs a generated sample dataset corresponding to a defect type is as follows: The first fully connected layer inside the generator is used to map a priori noise signal corresponding to any set of magnetic flux leakage defects and the defect label of the set of magnetic flux leakage defects to obtain initial features.

[0061] The generator's internal multi-head attention layer transforms the initial features using three independent linear transformation matrices to generate a query matrix, a key matrix, and a value matrix; where, In the formula, These represent the query matrix, key matrix, and value matrix, respectively. As initial features, This query retrieves the linear transformation matrix, the key linear transformation matrix, and the value linear transformation matrix.

[0062] The multi-head attention layer is used to divide the query matrix, key matrix, and value matrix into h attention heads, so as to obtain the query submatrix, key matrix, and value submatrix corresponding to each attention head.

[0063] A multi-head attention layer is used to calculate the relevance of each query submatrix, key submatrix, and value submatrix, obtaining the attention output corresponding to each attention head; where, for the c-th attention head, its attention output is: ; In the formula, This represents the attention output corresponding to the c-th attention head. For activation function, These represent the query submatrix, key submatrix, and value submatrix corresponding to the c-th attention head, respectively. For transpose operation, Let c be the feature dimension of the attention head, i.e. It equals d / h, where d is the feature dimension of the initial feature and h is the total number of attention heads.

[0064] The multi-head attention layer is also used to concatenate the attention outputs corresponding to all attention heads to obtain concatenated features, which are then used in conjunction with... Different linear transformation matrices convert the spliced ​​feature dimensions into fused features, where the dimensions of the fused features are the same as the dimensions of the initial features.

[0065] The second fully connected layer in the generator is used to perform linear or nonlinear transformations on the fused features to obtain mapped features.

[0066] The deconvolutional layer in the generator is used to upsample the mapped features layer by layer to obtain the initial generated sample data.

[0067] The output layer in the generator is used to map the initial generated sample data to the original signal space using a convolutional layer to generate a pseudo signal sample with the same size as the leakage magnetic defect signal in any of the leakage magnetic defect signal sets. In this way, the generated pseudo signal sample can be used as a generated sample data.

[0068] Based on this, by performing multiple random samplings in a multivariate mixed prior distribution model, multiple generated sample data can be generated, thus forming a generated sample dataset.

[0069] Of course, when training a generative adversarial network, the pseudo-sample data output by the generator will also be input into the discriminator along with the original samples for adversarial training. This will enable the generator to learn to generate realistic data and the discriminator to learn to distinguish between the data generated by the generator and the real data. By continuously training and iterating, the trained generative adversarial network can be obtained.

[0070] Therefore, this embodiment introduces a multi-head attention mechanism into the generator and extracts information from different subspaces by executing multiple independent attention heads in parallel. This improves the generative model's ability to capture global key magnetic leakage features, thereby enhancing the model's understanding and generation capabilities of complex signals. Based on this, the goal of improving the quality of sample generation can be achieved.

[0071] Thus, based on the aforementioned method, after generating sample datasets corresponding to each defect type, the generated samples can be screened, that is, the generated samples with the best quality can be selected as supplementary samples; the process of screening generated samples is as shown in step S23 below.

[0072] S23. Employ a data filtering algorithm based on quality error to filter out the optimal generated sample set corresponding to each defect type from the generated sample datasets corresponding to each defect type. In specific implementation, take any generated sample dataset as an example to guide the quality filtering process. Examples may include, but are not limited to, the following steps S23a to S23j.

[0073] S23a. For any generated sample dataset, based on the magnetic flux leakage defect signal set corresponding to the defect type of the generated sample dataset, generate a magnetic flux leakage test signal set corresponding to the defect type of the generated sample dataset. In this embodiment, several magnetic flux leakage defect signals can be randomly selected from the magnetic flux leakage defect signal set corresponding to the defect type of the generated sample dataset to form a magnetic flux leakage test signal set. Then, multiple data filtering can be performed so that the data filtered each time can be used to train a quality enhancement model. The quality enhancement model obtained from each training can be used to enhance the magnetic flux leakage test signal set. Finally, the optimal generated sample data can be selected from multiple filtered generated sample data by the quality error between the enhanced data and the original magnetic flux leakage test data.

[0074] The process of selecting the optimal generated sample set is shown in steps S23b to S23j below.

[0075] S23b. Perform the j-th data filtering from any of the generated sample datasets to obtain the j-th sample subset.

[0076] After completing the j-th data filtering and obtaining the j-th sample subset, it can be used as training data to train the quality enhancement model, as shown in step S23c below.

[0077] S23c. Use the j-th sample subset to train the quality enhancement model and obtain the trained quality enhancement model. In specific implementation, for example, but not limited to, a variational autoencoder can be used as the quality enhancement model, and the j-th sample subset can be used to train the model to obtain the trained quality enhancement model.

[0078] Thus, after obtaining the trained quality enhancement model, the encoder and decoder in the trained quality enhancement model can be used to perform data augmentation, as shown in step S23d below.

[0079] S23d. Input the j-th sample subset into the trained quality enhancement model to obtain the enhanced dataset; after inputting the j-th sample subset into the trained quality enhancement model, the enhanced data corresponding to each sample data in the j-th sample subset can be obtained; then, the corresponding error threshold can be determined by the quality error between the two, as shown in step S23e below.

[0080] S23e. Based on the j-th sample subset and the augmented dataset, calculate the error threshold corresponding to the j-th sample subset. In specific applications, for example, but not limited to, first calculate the mean square error between each sample data in the j-th sample subset and the corresponding augmented data in the augmented dataset; then, sort the augmented data in the augmented dataset according to the order of the mean square error from smallest to largest to obtain a sorted sequence; next, obtain the misjudgment ratio value, and determine the truncation threshold based on the misjudgment ratio value; finally, from the sorted sequence, select the augmented data whose sort number is the truncation threshold, and use the mean square error corresponding to the selected augmented data as the error threshold corresponding to the j-th sample subset.

[0081] In practical applications, the formula for calculating the truncation threshold is: G = P × (1 - R), where G is the truncation threshold, P is the total number of data in the sorted sequence, and R is the misjudgment ratio. Thus, the mean square error corresponding to the Gth augmented data in the sorted sequence can be used as the error threshold.

[0082] After calculating the error threshold, the magnetic flux leakage test signal can be used to test the trained quality enhancement model. Then, based on the error threshold, the number of data points in the magnetic flux leakage test signal that are judged as abnormal can be obtained. The process is shown in steps S23f and S23g below.

[0083] S23f. Input the target test signal set into the trained quality enhancement model to obtain the test enhancement signal set, wherein the target test signal set is the leakage magnetic field test signal set.

[0084] After obtaining the test enhancement signal set, its quality error with the target test signal set can be calculated, as shown in step S23g below.

[0085] S23g. Calculate the quality error between each target test signal in the target test signal set and the corresponding test enhancement signal in the test enhancement signal set, and count the number of target test signals with quality errors greater than the error threshold, which is used as the number of misclassifications. In this embodiment, the quality error between any target test signal and its corresponding test enhancement signal is the mean square error of the two. If the quality of the generated samples is high, the reconstruction capability of the trained quality enhancement model is strong, and therefore, the number of real samples misclassified as abnormal is small (i.e., the number of quality errors greater than the error threshold is small). Conversely, if there are many low-quality samples in the generated samples, it will affect the reconstruction capability of the model, which will lead to more real samples being misclassified. Based on this, when the quality error is greater than the error threshold, it can be determined that the corresponding target test signal is misclassified as abnormal. At this time, the number of target test signals with quality errors greater than the error threshold can be used as the number of misclassifications.

[0086] After obtaining the number of misclassifications, it can be associated with and stored with the sample subset obtained from the j-th data filtering, so that the sample subset can be searched later based on the number of misclassifications. The process is shown in step S23h below.

[0087] S23h. Record the number of misjudgments and the sample subset obtained from the j-th data filtering.

[0088] After the associated storage is completed, the data can be filtered again, and the above steps can be repeated until the maximum number of filtering times is reached. Then, multiple sample subsets and the corresponding number of misclassifications can be obtained. Finally, the generated samples can be filtered according to the size of the number of misclassifications. The process is shown in steps S23i and S23j below.

[0089] S23i. Increment j by 1, and perform the j-th data filtering again from any of the generated sample datasets until j equals J, to obtain J misclassifications. Here, the generated sample data in each sample subset is different, the initial value of j is 1, and J is the set maximum number of filtering times.

[0090] S23j. From the J number of misjudgments, select the sample subset associated with the smallest number of misjudgments, and use the selected sample subset as the optimal generated sample set for the defect type corresponding to any generated sample dataset; in this embodiment, the sample subset associated with the smallest number of misjudgments indicates that the quality of the generated samples is high, and in this case, it can be used as the optimal generated sample set.

[0091] Thus, through the aforementioned steps S23a to S23j, this embodiment quantifies the quality of the generated sample dataset by calculating the number of anomalies misclassified in the original samples of the test set. The larger the number of misclassified anomalies, the higher the proportion of low-quality generated samples in the generated sample dataset (i.e., using low-quality samples for model training has a significant negative impact on the reconstruction capability of the quality enhancement network, resulting in a larger number of misclassified original samples in the subsequent test set). Conversely, it indicates that the quality of the generated sample dataset is higher. Therefore, higher-quality samples can be selected from the generated samples to enhance the defect recognition capability of the subsequent defect classification model.

[0092] After obtaining the optimal generated sample set for each defect type, it can be merged with the corresponding magnetic leakage defect signal set to obtain the enhanced training dataset, as shown in step S24 below.

[0093] S24. Use the leakage magnetic defect signal sets of each defect type and the optimal generated sample set to form an enhanced training dataset; after obtaining the enhanced training dataset, the defect classification model can be trained, as shown in step S25 below.

[0094] S25. The enhanced training dataset is converted into an enhanced magnetic flux leakage image set, and the enhanced magnetic flux leakage image set is used to train the defect classification model to obtain the trained defect classification model. In specific implementation, the defect classification model is trained by taking each enhanced magnetic flux leakage image as input and the defect type and defect image block of the corresponding pipeline of each enhanced magnetic flux leakage image as output.

[0095] Thus, based on the aforementioned steps S21 to S25, after the model training is completed, the magnetic leakage image can be input into the trained defect classification model to obtain the corresponding defect image block and defect type.

[0096] After obtaining the defect image block, the leakage magnetic defect signal at the defect location can be extracted so that the defect depth can be calculated and the defect can be located. The process is shown in steps S3 to S5 below.

[0097] S3. Based on the defect image block, the magnetic flux leakage defect signal at the defect location of the petrochemical pipeline is segmented from the magnetic flux leakage detection signal. In specific implementation, based on the defect image block, the triaxial components of the defect image block in the magnetic flux leakage detection signal can be deduced. Thus, the triaxial components of the defect image block in the magnetic flux leakage detection signal can be used as the magnetic flux leakage defect signal. Then, the defect depth calculation and defect location can be performed, as shown in steps S4 and S5 below.

[0098] S4. Based on the magnetic flux leakage defect signal, generate a magnetic flux leakage signal curve at the defect center location. In specific implementations, axial or radial magnetic flux leakage signals are more sensitive to defect depth. Therefore, in this embodiment, the magnetic flux leakage signals in the axial or radial direction are first extracted from the magnetic flux leakage defect signal as target signals. Then, the signal point corresponding to the defect center is determined from the target signal as the defect center point (wherein, the point with the largest signal amplitude in the target signal is taken as the defect center point). Next, all signal points containing the defect center point can be extracted from the target signal along the axial direction (i.e., along the x-axis direction, which is the moving direction of the magnetic flux leakage detector). Finally, all extracted signal points can be used to generate the magnetic flux leakage signal curve at the defect center.

[0099] After generating the leakage magnetic flux signal curve at the defect center at the defect location, the defect depth and defect location can be calculated based on this curve, as shown in step S5 below.

[0100] S5. Based on the leakage magnetic signal curve at the defect center, determine the defect depth and the location information of the defect in the wall thickness direction of the petrochemical pipeline; in specific implementation, for example, but not limited to, the following steps S51 to S55 can be used to calculate the defect depth and determine the aforementioned location information.

[0101] S51. From the leakage magnetic signal curve at the defect center, determine the signal point corresponding to the maximum peak and the signal point corresponding to the lowest trough, and use them as the peak point and trough point respectively; in this embodiment, find the highest point and the lowest point of the defect waveform; then, based on the amplitude difference between the highest point and the lowest point and the signal spacing, calculate the steepness of the defect waveform, as shown in steps S52 to S54 below.

[0102] S52. Calculate the difference between the signal amplitude at the peak and the signal amplitude at the trough to obtain the peak-trough difference.

[0103] S53. Calculate the signal spacing between the peak and trough points.

[0104] After calculating the amplitude difference between the highest and lowest points and the lateral spacing based on the aforementioned steps S52 and S53, the steepness of the leakage magnetic signal curve at the defect center can be calculated based on this, as shown in step S54 below.

[0105] S54. Calculate the steepness of the leakage magnetic signal curve at the defect center using the peak-to-valley difference and the signal spacing. In specific implementation, for example, but not limited to, first obtain the scaling factor (which is a preset value); then calculate the product between the scaling factor and the signal spacing; finally, the ratio between the peak-to-valley difference and the product can be used as the steepness.

[0106] In this embodiment, the steepness directly reflects the degree of magnetic field diffusion, thus revealing the location and depth of the defect. A greater steepness results in a steeper signal, indicating concentrated magnetic field lines with less diffusion. This means the defect is very close to the sensor (Hall element), meaning the magnetic field is detected before it has a chance to diffuse outwards, resulting in a sharp peak in the signal. Conversely, a smaller steepness results in a flatter signal, indicating divergent magnetic field lines with significant diffusion. This means the defect is far from the sensor (located on the inner wall or deep within the pipe), meaning the magnetic field travels from the defect through the pipe wall to the outer surface. During this process, the magnetic field lines diffuse outwards, causing the signal to broaden, while energy attenuation results in a lower signal intensity.

[0107] Thus, after calculating the steepness, depth calculation and defect location can be performed based on it, as shown in step S55 below.

[0108] S55. Based on the steepness, determine the depth of the defect and the location information of the defect in the wall thickness direction; in specific implementation, first determine whether the steepness is greater than a first threshold; if so, determine that the location information of the defect is on the outer wall of the petrochemical pipeline; otherwise, continue to determine whether the steepness is less than a second threshold (the second threshold is less than the first threshold); if so, determine that the location information of the defect is on the inner wall of the petrochemical pipeline; similarly, when the steepness is between the second threshold and the first threshold, determine that the location information of the defect is between the outer wall and the inner wall of the petrochemical pipeline.

[0109] Furthermore, for example, the first threshold is 0.2 and the second threshold is 0.07.

[0110] In this way, the location information of the defect in the wall thickness direction can be determined.

[0111] Furthermore, in this embodiment, the positions of each sampling point corresponding to the defect leakage magnetic signal can be combined to perform precise positioning, that is, to obtain the distance between the defect location and the detection starting point on the pipeline, thereby forming more accurate positioning information.

[0112] After obtaining the location information, the defect depth can be calculated. For example, but not limited to, the defect depth can be calculated using a formula.

[0113] ; In the formula, Indicates steepness. Indicates the depth of the defect. These represent the position step size factor and the magnitude scaling factor, respectively.

[0114] In this embodiment, defects of different depths can be simulated in advance through finite element simulation, and corresponding defect leakage magnetic signals can be generated. Then, the relationship between depth and steepness of defect leakage magnetic signals is fitted to obtain the aforementioned correlation formula between defect depth and steepness. Furthermore, the aforementioned position compensation factor is affected by lift-off value and pipe wall thickness. Therefore, different position compensation factors can be set for different pipe wall thicknesses and lift-off values. Of course, the amplitude scaling factor is also preset and is not specifically limited here.

[0115] Thus, based on the aforementioned steps S51 to S55, after calculating the defect depth and determining the location information, it can be sent to the monitoring platform along with the defect type for management personnel to view. At the same time, it can also be used to generate an operation and maintenance work order containing operation and maintenance policies, so as to send it to operation and maintenance personnel for timely maintenance. The process is shown in step S6 below.

[0116] S6. The defect type, defect depth, and defect location information of the petrochemical pipeline are pushed to the monitoring platform, and a corresponding maintenance work order is generated and sent to the maintenance terminal. The maintenance work order includes a maintenance strategy. In this embodiment, for example, but not limited to, a maintenance strategy library can be pre-set. This strategy library stores maintenance strategies corresponding to different defect information (defect information includes pipeline type, defect type, defect depth, and defect location information). Thus, after obtaining the aforementioned defect type, defect depth, and defect location information of the petrochemical pipeline, the corresponding maintenance strategy can be matched from the maintenance strategy library based on the pipeline type of the petrochemical pipeline. In this way, the defect type, defect depth, defect location information, and maintenance strategy can be used to generate a maintenance work order and send it to the currently available maintenance personnel who are closest to the defective pipeline (i.e., the detected pipeline segment). Based on this, maintenance personnel can arrive at the site in a timely manner to inspect the defect and carry out maintenance to prevent safety accidents.

[0117] After completing defect maintenance, the corresponding pipeline images after maintenance can be uploaded. At the same time, information such as the maintenance method used, the actual maintenance personnel involved, and the maintenance time can be uploaded to form maintenance feedback information. The process is shown in step S7 below.

[0118] S7. Receive the operation and maintenance feedback information uploaded by the operation and maintenance terminal, and record the operation and maintenance work order and operation and maintenance feedback information as operation and maintenance logs for storage, so as to complete the intelligent management of operation and maintenance of petrochemical pipelines after storage; in this embodiment, the operation and maintenance feedback information and operation and maintenance work order are recorded together, which can facilitate subsequent traceability and form knowledge accumulation data, thereby providing data support for subsequent operation and maintenance strategy optimization.

[0119] Therefore, through the intelligent management method for safe operation and maintenance of petrochemical pipelines described in detail in steps S1 to S7 above, this invention addresses the industry pain points of traditional magnetic flux leakage detection, which relies on manual experience, resulting in low identification efficiency and poor accuracy, as well as the information silos formed by the disconnect between detection and maintenance. By intelligently converting magnetic flux leakage signals into images and automatically identifying them using models, this method achieves precise quantitative analysis of defect type, depth, and location information, significantly improving the accuracy and efficiency of defect identification. Simultaneously, this method constructs a closed-loop data linkage from defect identification and quantitative analysis to maintenance work order generation and feedback recording. This not only achieves transparent management and efficient collaboration of the operation and maintenance process but also provides data support for subsequent strategy optimization through the accumulation of maintenance logs, thereby effectively eliminating safety hazards and greatly improving the intelligence level and safety of petrochemical pipeline operation and maintenance management.

[0120] like Figure 2As shown, the second aspect of this embodiment provides a hardware system for implementing the intelligent management method for safe operation and maintenance of petrochemical plants as described in the first aspect of the embodiment, comprising: The signal conversion unit is used to acquire the magnetic flux leakage detection signal of the petrochemical pipeline and convert the magnetic flux leakage detection signal into a magnetic flux leakage image.

[0121] The defect identification unit is used to identify defects in the magnetic flux leakage image using a pre-trained defect classification model, so as to segment defect image blocks in the magnetic flux leakage image and determine the defect type of the petrochemical pipeline.

[0122] The signal segmentation unit is used to segment the magnetic flux leakage defect signal at the defect location of the petrochemical pipeline from the magnetic flux leakage detection signal based on the defect image block.

[0123] The defect parameter calculation unit is used to generate a defect center magnetic flux leakage signal curve at the defect location based on the magnetic flux leakage defect signal.

[0124] The defect parameter calculation unit is also used to determine the defect depth of the petrochemical pipeline and the location information of the defect in the wall thickness direction of the petrochemical pipeline based on the leakage magnetic signal curve of the defect center.

[0125] The operation and maintenance unit is used to push the defect type, defect depth and defect location information of the petrochemical pipeline to the monitoring platform, and generate corresponding operation and maintenance work orders to send to the operation and maintenance terminal. The operation and maintenance work orders contain operation and maintenance strategies.

[0126] The management unit is used to receive operation and maintenance feedback information uploaded by the operation and maintenance terminal, and to record operation and maintenance work orders and operation and maintenance feedback information as operation and maintenance logs for storage, so as to complete the intelligent management of operation and maintenance of petrochemical pipelines after storage.

[0127] The working process, working details and technical effects of the system provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0128] like Figure 3 As shown, the third aspect of this embodiment provides a petrochemical safety operation and maintenance digital intelligent management device. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the petrochemical safety operation and maintenance digital intelligent management method as described in the first aspect of the embodiment.

[0129] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0130] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0131] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0132] The fourth aspect of this embodiment provides a storage medium that stores instructions containing the intelligent management method for petrochemical safety operation and maintenance described in the first aspect of the embodiment. That is, the storage medium stores instructions, and when the instructions are run on a computer, the intelligent management method for petrochemical safety operation and maintenance described in the first aspect of the embodiment is executed.

[0133] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0134] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0135] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the intelligent management method for safe operation and maintenance of petrochemical plants as described in the first aspect of this embodiment. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0136] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A digital intelligent management method for safe operation and maintenance in the petrochemical industry, characterized in that, include: The magnetic flux leakage detection signal of the petrochemical pipeline is acquired and converted into a magnetic flux leakage image; Using a pre-trained defect classification model, defects are identified in the magnetic flux leakage image to segment defect image blocks in the magnetic flux leakage image and to determine the defect type of the petrochemical pipeline. Based on the defect image block, the magnetic flux leakage defect signal at the defect location of the petrochemical pipeline is segmented from the magnetic flux leakage detection signal; Based on the magnetic flux leakage defect signal, a magnetic flux leakage signal curve at the defect center at the defect location is generated; Based on the magnetic flux leakage signal curve at the defect center, the defect depth and the location of the defect in the wall thickness direction of the petrochemical pipeline are determined. The defect type, defect depth, and defect location information of the petrochemical pipeline are pushed to the monitoring platform, and a corresponding operation and maintenance work order is generated and sent to the operation and maintenance terminal. The operation and maintenance work order contains the operation and maintenance strategy. It receives maintenance feedback information uploaded from the maintenance terminal and records maintenance work orders and maintenance feedback information as maintenance logs for storage, so as to complete the intelligent management of maintenance of petrochemical pipelines after storage.

2. The method according to claim 1, characterized in that, The leakage magnetic flux detection signal is a triaxial signal, and the three axes are the axial direction, the circumferential direction and the radial direction, respectively; The process of generating a magnetic flux leakage signal curve at the defect center location based on the magnetic flux leakage defect signal includes: From the leakage magnetic defect signal, the leakage magnetic signal in the axial or radial direction is extracted as the target signal; From the target signal, determine the signal point corresponding to the defect center, and use it as the defect center point; Along the axial direction, all signal points containing the defect center point are extracted from the target signal, and the defect center leakage magnetic signal curve is generated using all the extracted signal points.

3. The method according to claim 1, characterized in that, Based on the magnetic flux leakage signal curve at the defect center, the defect depth and the location of the defect in the pipeline along the wall thickness direction are determined, including: From the leakage magnetic field signal curve at the defect center, the signal point corresponding to the maximum peak and the signal point corresponding to the minimum valley are determined, and are respectively used as the peak point and valley point. The difference between the signal amplitude at the peak and the signal amplitude at the trough is calculated to obtain the peak-trough difference. Calculate the signal spacing between the peak and trough points; The steepness of the leakage magnetic signal curve at the defect center is calculated using the peak-to-valley difference and the signal spacing. Based on the steepness, the depth of the defect and the location information of the defect in the wall thickness direction are determined.

4. The method according to claim 3, characterized in that, Using the peak-to-valley difference and the signal spacing, the steepness of the leakage magnetic field signal curve at the defect center is calculated, including: Obtain the scaling factor and calculate the product between the scaling factor and the signal spacing; The ratio between the peak-to-valley difference and the product is taken as the steepness; Among them, based on the steepness, the location information of the defect in the wall thickness direction is determined, including: Determine whether the steepness is greater than a first threshold; If not, then determine whether the steepness is less than the second threshold; otherwise, determine that the location of the defect is on the outer wall of the petrochemical pipeline, wherein the second threshold is less than the first threshold. If so, the location information of the defect is determined to be on the inner wall of the petrochemical pipeline. When the steepness is between the second threshold and the first threshold, the location information of the defect is determined to be between the outer wall and the inner wall of the petrochemical pipeline.

5. The method according to claim 3, characterized in that, Based on the steepness, the defect depth is determined, including: The depth of the defect is calculated using the following formula; ； In the formula, Indicates steepness. Indicates the depth of the defect. These represent the position compensation factor and the amplitude scaling factor, respectively.

6. The method according to claim 1, characterized in that, The defect classification model is trained using the following method: Obtain the magnetic flux leakage defect signal set corresponding to different defect types; Based on each set of magnetic flux leakage defects, a priori noise signals for each defect type are generated. The priori noise signal for any defect type contains the priori distribution information of the magnetic flux leakage defect signal set corresponding to that defect type. Using each magnetic flux leakage defect signal set and the corresponding prior noise signal, a generated sample dataset for each defect type is generated. A data filtering algorithm based on quality error is used to filter out the optimal generated sample set for each defect type from the generated sample datasets corresponding to each defect type. An enhanced training dataset is constructed using the magnetic flux leakage defect signal sets of various defect types and the optimal generated sample set. The enhanced training dataset is converted into an enhanced magnetic flux leakage image set, and the defect classification model is trained using the enhanced magnetic flux leakage image set to obtain the trained defect classification model.

7. The method according to claim 6, characterized in that, Based on the various magnetic flux leakage defect signal sets, prior noise signals for each defect type are generated, including: For any set of magnetic flux leakage defect signals, feature extraction is performed on the set of magnetic flux leakage defect signals to obtain a set of feature samples, wherein the feature data in the set of feature samples corresponds one-to-one with the magnetic flux leakage defect signals in the set of magnetic flux leakage defect signals. Clustering is performed on the feature data in the feature sample set to obtain multiple feature clusters; Calculate the mean and covariance matrix of each feature cluster; Based on the mean and covariance matrix of each feature cluster, the prior distribution of each feature cluster is calculated. By utilizing the prior distribution of each feature cluster, a multivariate hybrid prior distribution model of any magnetic flux leakage defect signal set is constructed. Using a multivariate hybrid prior distribution model, a prior noise signal corresponding to the defect type of any magnetic flux leakage defect signal set is generated.

8. The method according to claim 6, characterized in that, A data filtering algorithm based on quality error is used to select the optimal generated sample set for each defect type from the generated sample datasets corresponding to each defect type, including: For any generated sample dataset, a magnetic flux leakage test signal set corresponding to the defect type of the generated sample dataset is generated based on the magnetic flux leakage defect signal set corresponding to the defect type of the generated sample dataset. Perform the j-th data filtering from any of the generated sample datasets to obtain the j-th sample subset; Using the j-th sample subset, train the quality enhancement model to obtain the trained quality enhancement model; The j-th sample subset is input into the trained quality enhancement model to obtain the enhanced dataset; Based on the j-th sample subset and the augmented dataset, calculate the error threshold corresponding to the j-th sample subset; The target test signal set is input into the trained quality enhancement model to obtain the test enhancement signal set, wherein the target test signal set is the leakage magnetic field test signal set; Calculate the quality error between each target test signal in the target test signal set and the corresponding test enhancement signal in the test enhancement signal set, and count the number of target test signals whose quality error is greater than the error threshold, as the number of misjudgments; The number of misjudgments and the sample subset obtained from the j-th data filtering are linked and recorded; Increment j by 1, and perform the j-th data filtering again from any of the generated sample datasets until j equals J, to obtain J misclassifications. Here, the generated sample data in each sample subset is different, the initial value of j is 1, and J is the set maximum number of filtering times. From the J misclassification counts, select the sample subset associated with the smallest misclassification count, and use the selected sample subset as the optimal generated sample set for the defect type corresponding to any generated sample dataset.

9. A petrochemical safety operation and maintenance digital intelligent management system, characterized in that, include: The signal conversion unit is used to acquire the magnetic flux leakage detection signal of the petrochemical pipeline and convert the magnetic flux leakage detection signal into a magnetic flux leakage image; The defect identification unit is used to identify defects in the magnetic flux leakage image using a pre-trained defect classification model, so as to segment defect image blocks in the magnetic flux leakage image and determine the defect type of the petrochemical pipeline. The signal segmentation unit is used to segment the magnetic flux leakage defect signal at the defect location of the petrochemical pipeline from the magnetic flux leakage detection signal based on the defect image block; The defect parameter calculation unit is used to generate a defect center magnetic flux leakage signal curve at the defect location based on the magnetic flux leakage defect signal. The defect parameter calculation unit is also used to determine the defect depth of the petrochemical pipeline and the location information of the defect in the wall thickness direction of the petrochemical pipeline based on the leakage magnetic signal curve of the defect center. The operation and maintenance unit is used to push the defect type, defect depth and defect location information of the petrochemical pipeline to the monitoring platform, and generate corresponding operation and maintenance work orders to send to the operation and maintenance terminal. The operation and maintenance work orders contain operation and maintenance strategies. The management unit is used to receive operation and maintenance feedback information uploaded by the operation and maintenance terminal, and to record operation and maintenance work orders and operation and maintenance feedback information as operation and maintenance logs for storage, so as to complete the intelligent management of operation and maintenance of petrochemical pipelines after storage.

10. A computer program product containing instructions, characterized in that, When the instruction is executed on the computer, it causes the computer to perform the intelligent management method for safe operation and maintenance of petrochemical plants as described in any one of claims 1 to 8.

Images