Steel pipe nondestructive testing system and method based on machine learning

Abstract

The invention relates to the field of non-destructive testing, and particularly discloses a steel pipe non-destructive testing system and method based on machine learning, and the non-destructive testing specifically comprises the following steps: collecting and preprocessing an initial eddy current testing signal; performing time-frequency domain feature analysis on the signal to obtain basic time-frequency feature data; performing feature extraction and fusion processing on the eddy current signals through a convolutional neural network based on an attention mechanism and a bidirectional gating circulation unit; when the features are verified, an integrated classification module is adopted to classify the features. According to the method, weight distribution of signal features is optimized by introducing an attention mechanism, and time sequence feature extraction capabilities of CNN and Bi GRU are combined, so that the accuracy of eddy current detection in steel tube defect recognition is remarkably improved, the problem of recognition errors caused by noise interference and signal complexity in a traditional eddy current detection technology is solved, and the accuracy of steel tube defect recognition is improved. The accuracy of nondestructive testing is improved, and potential risks caused by steel pipe defects are reduced.

Description

Technical Field

[0001] The present invention belongs to the field of non-destructive testing, and specifically is a steel pipe non-destructive testing system based on machine learning. Background Art

[0002] Nondestructive testing is a technology that uses physical or chemical methods to detect the internal and surface structure, properties, state and defects of the object being tested without destroying its integrity and usability.

[0003] However, existing traditional non-destructive testing methods rely on manual operation or simple automated equipment, resulting in low testing efficiency and possible human errors. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a steel pipe nondestructive testing system based on machine learning to solve the problem that the nondestructive testing method in the prior art relies on manual operation or simple automated equipment, resulting in low detection efficiency.

[0005] In order to solve the above technical problems, the present invention provides a nondestructive testing system for steel pipes based on machine learning, wherein the nondestructive testing system includes an eddy current data acquisition module, an AI data analysis module, a monitoring and control module, and a user interaction module;

[0006] The eddy current data acquisition module uses eddy current detection equipment to measure N groups of eddy current signals of steel pipes in normal state and damaged state to form a data set, labels the signal sequence in the data set, and transmits the collected eddy current data to the AI ​​data analysis module;

[0007] The AI ​​data analysis module includes a signal preprocessing unit, a feature extraction unit, and a damage identification unit; the signal preprocessing unit filters and denoises the collected eddy current data to improve the signal quality, discretizes the filtered and denoised eddy current data, and converts them into digital signals suitable for wavelet transform; the feature extraction unit uses a hybrid model of a convolutional neural network based on an attention mechanism and a bidirectional gated recurrent unit to extract features from the eddy current frequency domain data, and saves the trained model parameters; the damage identification unit further processes and analyzes the extracted features, and identifies the damage of the steel pipe through a machine learning model;

[0008] The monitoring and control module monitors the steel pipe in real time according to the analysis results of the AI ​​data analysis module, and automatically triggers an alarm when damage is detected;

[0009] The user interaction module provides a user operation interface, allowing the user to view the monitoring data and damage status of the steel pipe and receive alarm information.

[0010] A better technical solution of the present invention: the eddy current data acquisition module includes an eddy current detection probe and a signal converter, the eddy current detection probe includes a plurality of high-sensitivity eddy current sensors fixed on the surface of the steel pipe, and collects data by detecting electromagnetic changes on the surface of the steel pipe; the signal converter is used to convert the eddy current signal into an electrical signal for collection and transmit it to the signal preprocessing unit.

[0011] The preferred technical solution of the present invention is as follows: the feature extraction unit selects some wavelet coefficients as features according to task requirements, and these features can reflect the characteristics of the signal at different frequencies and time scales, and adopts a feature extraction method based on wavelet transform to further decompose the signal into smaller sub-bands, and extract statistical features in the sub-bands, including mean and variance, and adopts a feature extraction method based on wavelet energy spectrum to calculate the energy of wavelet transform coefficients of different scales, and obtain the frequency domain features of the signal at different scales;

[0012] The obtained features are then input into a hybrid model of a convolutional neural network based on an attention mechanism and a bidirectional gated recurrent unit. The CNN layer further refines the features and captures local dependencies. At the same time, the attention mechanism ensures that the model focuses on the most critical features for damage identification. The BiGRU layer processes these features to capture the temporal correlation between features and fuses features from different sources through the Concatenate function. The model is then nonlinearly fitted through a neural network to improve the classification accuracy and generalization ability of the model. That is, after the model training is completed, the outputs of the CNN and BiGRU are fused through a fully connected layer, which uses a softmax function to convert the output vector into a probability distribution.

[0013] A better technical solution of the present invention is as follows: the damage identification unit uses the model constructed by the feature extraction unit to judge the degree of fit of the measured signal, and the probability of the model output is used to determine whether the signal indicates the presence of damage; the probability threshold is specifically set to 0.5, and the characteristic vector of the measured signal is nonlinearly mapped through the model constructed by the feature extraction unit. If the output probability corresponding to the damage category is greater than 0.5, the signal label is marked as 1, indicating that the component corresponding to the signal is damaged; if the output probability is less than 0.5, the signal label is marked as 0, indicating that the component corresponding to the signal is not damaged.

[0014] The preferred technical solution of the present invention is as follows: wherein wavelet transform and a hybrid model based on CNN and BiGRU are used to extract key features from preprocessed data. For a signal f(t), its continuous wavelet transform is defined as:

[0015]

[0016] Among them, a is the scale parameter, which determines the expansion and contraction of the wavelet function; b is the translation parameter, which determines the movement of the wavelet function on the time axis; ψ(t) is the mother wavelet, which is a function with finite energy and zero average value; ψ*(t) represents the complex conjugate of ψ(t); CWT f,ψ (a, b) represents the transform coefficients at scale a and translation b, which contain information about the signal at different times and scales;

[0017] The calculation process of the BiGRU unit is as follows:

[0018] z t =σ(W z ·[h t-1 ,x t ])

[0019] r t =σ(W r ·[h t-1 ,x t ])

[0020]

[0021] Among them, z t represents the update gate, r t Represents the reset gate;

[0022] Controls the amount of information from the previous node that is written to the current candidate set GRU will reset the previous gate output r t and the input state h at the previous moment t-1 , input x at the current time t , perform calculations to form new memory content for the current node GRU combines the input state h of the previous moment t-1 With the current new memory content To determine the state of the current node h t ;

[0023] σ is the Sigma function, tanh is the hyperbolic tangent function, W is the weight, x t represents the data input at time t, h t-1 Represents the data input at the last moment, h t is the output at the current moment, W, V, U represent weight matrices, b and c represent bias matrices, and y t is the final output of Bi GRU.

[0024] A preferred technical solution of the present invention: the nondestructive testing system also includes a remote monitoring module, which is used to realize remote steel pipe status monitoring and damage identification result viewing. The remote monitoring module is connected to the monitoring and control module through a network, allowing remote users to view the monitoring data and damage status of the steel pipe in real time and receive alarm information through a user interaction module and a special remote monitoring interface.

[0025] A better technical solution of the present invention: the nondestructive testing system also includes an intelligent maintenance suggestion module, which provides users with maintenance suggestions and repair plans for steel pipes based on the damage identification results of the AI ​​data analysis module. The intelligent maintenance suggestion module considers the current status, damage type, damage degree and development trend of the steel pipe, combines historical maintenance data and expert knowledge, and generates personalized maintenance strategies for users to reduce downtime and maintenance costs.

[0026] A preferred technical solution of the present invention: the AI ​​data analysis module also includes a data storage unit for storing the collected eddy current data, features extracted by wavelet transform, and damage identification results.

[0027] According to a machine learning-based nondestructive testing system for steel pipes according to claim 1 or 2, it is characterized in that the AI ​​data analysis module also includes a data storage unit for storing the collected eddy current data, features extracted by wavelet transform, and damage identification results.

[0028] The present invention also provides a nondestructive testing method for steel pipes based on machine learning, which specifically includes the following steps:

[0029] S1. Use eddy current testing equipment to measure N groups of eddy current signals of steel pipes in normal state and damaged state to form a data set, and label the signal sequence in the data set;

[0030] S2. Filter and denoise the eddy current data collected in step S1 to improve signal quality, discretize the filtered and denoised eddy current data, and convert them into digital signals suitable for wavelet transform; extract features of the eddy current frequency domain data based on a hybrid model of a convolutional neural network (CNN) and a bidirectional gated recurrent unit (Bi GRU) based on an attention mechanism, and save the trained model parameters; further process and analyze the extracted features, and identify the damage of the steel pipe through a machine learning model;

[0031] S3. Monitor the steel pipe in real time according to the damage identification result in step S2, and automatically trigger an alarm when damage is detected;

[0032] S4. Store and display the results identified in step S2 and the information monitored in step S3.

[0033] A further technical solution of the present invention: the step S2 specifically comprises the following steps:

[0034] S201. Signal preprocessing: filtering and denoising the collected eddy current data to improve the signal quality, discretizing the filtered and denoised eddy current data and converting them into digital signals suitable for wavelet transform; performing wavelet transform on the discretized signal, obtaining wavelet coefficients at different scales by calculating the inner product of the signal and the wavelet function, performing multi-scale analysis on the signal according to the set scale parameter, and extracting wavelet coefficients of different frequency components;

[0035] S202. Feature extraction: According to the task requirements, some wavelet coefficients are selected as features. These features can reflect the characteristics of the signal at different frequencies and time scales. The feature extraction method based on wavelet transform is used to further decompose the signal into smaller sub-bands and extract the statistical features in the sub-bands, including the mean and variance. The feature extraction method based on wavelet energy spectrum is used to calculate the energy of wavelet transform coefficients at different scales to obtain the frequency domain features of the signal at different scales. The obtained features are then input into a hybrid model of convolutional neural network based on attention mechanism and bidirectional gated recurrent unit. The CNN layer further refines the features and captures local dependencies. At the same time, the attention mechanism ensures that the model focuses on the most critical features for damage identification. The Bi GRU layer processes these features to capture the temporal correlation between features and fuses features from different sources through the Concatenate function. Then, the model is nonlinearly fitted through the neural network to improve the classification accuracy and generalization ability of the model. That is, after the model training is completed, the outputs of CNN and Bi GRU are fused through a fully connected layer. This layer uses the softmax function to convert the output vector into a probability distribution.

[0036] S203. Damage identification: The model constructed in step S202 is used to judge the degree of fit of the measured signal, and the probability of the model output is used to determine whether the signal indicates the presence of damage.

[0037] A further technical solution of the present invention is as follows: in said S203, the probability threshold is set to 0.5, and the characteristic vector of the measured signal is nonlinearly mapped through the model in step 2; if the output probability corresponding to the damage category is greater than 0.5, the signal label is marked as 1, indicating that the component corresponding to the signal is damaged; if the output probability is less than 0.5, the signal label is marked as 0, indicating that the component corresponding to the signal is not damaged.

[0038] Compared with the prior art, the present invention has the following beneficial effects: by collecting eddy current data of steel pipes in real time and accurately identifying damage through machine learning models, real-time monitoring and early warning of steel pipe quality are ensured; the application of AI data analysis modules makes the damage identification process more intelligent, reduces human intervention, and improves detection efficiency; the detection method of the present invention optimizes the weight distribution of signal features by introducing an attention mechanism, and combines the time series feature extraction capabilities of CNN and Bi GRU, which significantly improves the accuracy of eddy current detection in steel pipe defect identification, and solves the problem of identification error caused by noise interference and signal complexity in traditional eddy current detection technology. The present invention can not only perform comprehensive detection of steel pipes, but also its analysis method based on machine learning has good scalability and can adapt to changes in different materials and detection requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a system diagram of the present invention.

[0040] Figure 2 This is the damage identification criterion diagram of the present invention.

[0041] Figure 3 Schematic diagram of the CNN and Bi GRU hybrid model structure used in the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Embodiment 1 provides a nondestructive testing system for steel pipes based on machine learning, such as Figures 1 to 3 As shown, it includes eddy current data acquisition module, AI data analysis module, monitoring and control module and user interaction module

[0044] The eddy current data acquisition module includes an eddy current detection probe and a signal converter. The eddy current detection probe includes a plurality of high-sensitivity eddy current sensors fixed on the surface of the steel pipe, which collect data by detecting electromagnetic changes on the surface of the steel pipe; the signal converter is used to convert the eddy current signal into an electrical signal for collection. The eddy current data acquisition module is used to collect real-time data on the steel pipe through eddy current detection technology. The eddy current detection equipment is used to measure a steel pipe in good condition and a steel pipe in damaged condition for real-time data collection. First, 300 sets of eddy current data are collected to form a data set; the signal on the data set is marked with a damaged / undamaged label; and the collected eddy current data is transmitted to the AI ​​data analysis module.

[0045] The AI ​​data analysis module is used to perform signal preprocessing and feature extraction on the eddy current data set, build a machine learning model, output the label (damage / no damage) corresponding to the signal through the machine learning model, and identify the damage of the steel pipe; the AI ​​data analysis module includes a signal preprocessing unit, a feature extraction unit, and a damage identification unit.

[0046] The feature extraction module is used to normalize and wavelet transform the collected eddy current data; the normalization method is the Z-score normalization method, and the calculation formula is as follows:

[0047]

[0048]

[0049] Among them, x i represents the i-th data point in the dataset, and n is the total number of data points.

[0050] The normalized data is then subjected to wavelet transform, and key features are extracted from the data using wavelet transform. For a signal f(t), its continuous wavelet transform is defined as:

[0051]

[0052] Among them, a is the scale parameter, which determines the expansion and contraction of the wavelet function; b is the translation parameter, which determines the movement of the wavelet function on the time axis; ψ(t) is the mother wavelet, which is a function with finite energy and zero average value; ψ*(t) represents the complex conjugate of ψ(t); CWT f,ψ (a,b) represents the transform coefficients at scale a and translation b, which contains the information of the signal at different times and scales.

[0053] The feature extraction unit selects some wavelet coefficients as features, which can reflect the characteristics of the signal at different frequencies and time scales, and uses a feature extraction method based on wavelet transform to further decompose the signal into smaller sub-bands, and extracts statistical features in the sub-bands, including mean and variance, and uses a feature extraction method based on wavelet energy spectrum to calculate the energy of wavelet transform coefficients at different scales to obtain frequency domain features of the signal at different scales;

[0054] The obtained features are then input into a hybrid model of a convolutional neural network (CNN) and a bidirectional gated recurrent unit (BiGRU) based on an attention mechanism. The CNN layer further refines the features. The CNN layer performs convolution operations on the input features through multiple convolution kernels to detect local features at different locations in the signal. Each convolution kernel extracts a specific pattern or local correlation, and the results of multiple layers of convolution are superimposed. At the same time, the attention mechanism ensures that the model focuses on the most critical features for damage identification. The attention mechanism assigns different weights to features at different time and frequency locations according to the importance of the damage features. Specifically, the attention layer assigns weight values ​​to the features of each time step in the wavelet frequency domain data. The larger the weight value, the more important the feature is for damage identification; the BiGRU layer processes these features to capture the temporal correlation between features. The Bi GRU processes the signal simultaneously through two independent GRU layers, forward and backward, to capture the temporal dependency between the front and back of the signal. The forward GRU processes the input of the current time step and the information of the previous time step, while the backward GRU traces back information from the future time step. This bidirectional structure enables the model to understand the temporal characteristics of the signal more comprehensively. Each GRU unit contains an update gate and a reset gate. The update gate controls whether the current time step retains the previous memory, and the reset gate determines whether to forget the previous state information. In the damage detection process, this mechanism enables Bi GRU to flexibly adjust its dependence on past and future information. After BiGRU processing, the features from different sources are fused through the Concatenate function; after CNN, attention mechanism and BiGRU processing, the extracted features are sent to the fully connected layer for fusion, and finally the classification results are output through the softmax function. The system will output the corresponding damage probability based on the characteristics of the eddy current signal.

[0055] The calculation process of the Bi GRU unit is as follows:

[0056] z t =σ(W z ·[h t-1 ,x t ])

[0057] r t =σ(W r ·[h t-1 ,x t ])

[0058]

[0059] Among them, z t represents the update gate, r t Represents the reset gate, which controls the amount of information of the previous node to be written into the current candidate set GRU will reset the previous gate output r tand the input state h at the previous moment t-1 , input x at the current time t , perform calculations to form new memory content for the current node GRU combines the input state h of the previous moment t-1 With the current new memory content To determine the state of the current node h t σ is the Si gmoid function, tanh is the hyperbolic tangent function, W is the weight, x t represents the data input at time t, h t-1 Represents the data input at the last moment, h t is the output at the current moment. W, V, and U all represent weight matrices; b and c represent bias matrices. t is the final output of BiGRU.

[0060] The damage identification unit uses the model constructed in the feature extraction unit to judge the degree of fit of the measured signal, and the probability of the model output is used to determine whether the signal indicates damage. The probability threshold is set to 0.5. If the output probability corresponding to the damage category is greater than 0.5, it is considered that the component corresponding to the signal is damaged; if the output probability is less than 0.5, it is considered that there is no damage;

[0061] The monitoring and control module is used to monitor the steel pipe in real time based on the analysis results of the AI ​​data analysis module and automatically trigger an alarm when damage is detected;

[0062] The user interaction module is used to provide a user operation interface, allowing the user to view the monitoring data and damage status of the steel pipe and receive alarm information.

[0063] From the above, it can be seen that this system obtains the eddy current data of the steel pipe in real time through the eddy current data acquisition module, and performs signal preprocessing and feature extraction through the AI ​​data analysis module, effectively extracting the key features reflecting the damage of the steel pipe, and using the machine learning model to conduct in-depth analysis of these features to accurately identify the damage of the steel pipe. The monitoring and control module monitors the status of the steel pipe in real time according to the analysis results. Once damage is found, it automatically triggers an alarm to ensure production safety. The user interaction module provides users with an intuitive operation interface, allowing users to easily view the monitoring data and damage status of the steel pipe and receive alarm information. The entire system realizes the automation of the entire process from data acquisition to damage prediction, which greatly improves the intelligence level of steel pipe detection.

[0064] Embodiment 2: This embodiment is basically the same as the previous embodiment, except that the monitoring and control module is connected to the AI ​​data analysis module, which is used to monitor the steel pipe in real time according to the analysis results of the AI ​​data analysis module and automatically trigger an alarm when damage is detected.

[0065] Specifically, the user interaction module is connected to the monitoring and control module, providing a user operation interface, allowing the user to view the monitoring data and damage status of the steel pipe and receive alarm information.

[0066] The second embodiment also includes a remote monitoring module and an intelligent maintenance suggestion module. The remote monitoring module is used to realize remote monitoring of the steel pipe status and viewing of damage identification results. The remote monitoring module is connected to the monitoring and control module through the network, allowing remote users to view the monitoring data and damage status of the steel pipe in real time and receive alarm information through the user interaction module or a special remote monitoring interface. The intelligent maintenance suggestion module provides users with maintenance suggestions and repair plans for steel pipes based on the damage identification results of the AI ​​data analysis module. The intelligent maintenance suggestion module considers the current damage status of the steel pipe, combines historical maintenance data and expert knowledge, and generates personalized maintenance strategies for users to reduce downtime and maintenance costs.

[0067] The AI ​​data analysis module also includes a data storage unit for storing the collected eddy current data, features extracted by wavelet transform, and damage identification results.

[0068] Embodiment 3 provides a nondestructive testing method for steel pipes based on machine learning, which specifically includes the following steps:

[0069] S1. Use eddy current testing equipment to measure N groups of eddy current signals of steel pipes in normal state and damaged state to form a data set, and label the signal sequence in the data set;

[0070] S2. Filter and denoise the eddy current data collected in step S1 to improve signal quality, discretize the filtered and denoised eddy current data, and convert them into digital signals suitable for wavelet transform; extract features from the eddy current frequency domain data using a hybrid model of a convolutional neural network (CNN) based on an attention mechanism and a bidirectional gated recurrent unit (BiGRU), and save the trained model parameters; further process and analyze the extracted features, and identify the damage of the steel pipe through a machine learning model; specifically, the following steps are included:

[0071] S201. Signal preprocessing: filtering and denoising the collected eddy current data to improve the signal quality, discretizing the filtered and denoised eddy current data and converting them into digital signals suitable for wavelet transform; performing wavelet transform on the discretized signal, obtaining wavelet coefficients at different scales by calculating the inner product of the signal and the wavelet function, performing multi-scale analysis on the signal according to the set scale parameter, and extracting wavelet coefficients of different frequency components;

[0072] S202. Feature extraction: According to the task requirements, some wavelet coefficients are selected as features. These features can reflect the characteristics of the signal at different frequencies and time scales. The feature extraction method based on wavelet transform is used to further decompose the signal into smaller sub-bands and extract the statistical features in the sub-bands, including the mean and variance. The feature extraction method based on wavelet energy spectrum is used to calculate the energy of wavelet transform coefficients at different scales to obtain the frequency domain features of the signal at different scales. The obtained features are then input into a hybrid model of convolutional neural network based on attention mechanism and bidirectional gated recurrent unit. The CNN layer further refines the features and captures local dependencies. At the same time, the attention mechanism ensures that the model focuses on the most critical features for damage identification. The Bi GRU layer processes these features to capture the temporal correlation between features and fuses features from different sources through the Concatenate function. Then, the model is nonlinearly fitted through the neural network to improve the classification accuracy and generalization ability of the model. That is, after the model training is completed, the outputs of CNN and BiGRU are fused through a fully connected layer. This layer uses the softmax function to convert the output vector into a probability distribution.

[0073] S203. Damage identification: The degree of fit of the measured signal is judged using the model constructed in step S202, and the probability of the model output is used to determine whether the signal indicates damage. The probability threshold is set to 0.5 in S203, and the characteristic vector of the measured signal is nonlinearly mapped through the model in step 2. If the output probability corresponding to the damage category is greater than 0.5, the signal label is marked as 1, indicating that the component corresponding to the signal is damaged; if the output probability is less than 0.5, the signal label is marked as 0, indicating that the component corresponding to the signal is not damaged.

[0074] S3. Monitor the steel pipe in real time according to the damage identification result in step S2, and automatically trigger an alarm when damage is detected;

[0075] S4. Store and display the results identified in step S2 and the information monitored in step S3.

[0076] From the above, it can be seen that the steel pipe nondestructive testing system based on machine learning of the present invention realizes efficient monitoring and damage prediction of steel pipes through the collaborative work of multiple modules. The eddy current data acquisition module converts eddy current signals into electrical signals and improves the signal quality through preprocessing. The AI ​​data analysis module uses machine learning technology to perform in-depth analysis on the collected data and accurately identify steel pipe damage. The monitoring and control module tracks the status of the steel pipe in real time and automatically triggers an alarm once damage is found. The user interaction module provides an intuitive interface so that users can easily view monitoring data, damage conditions and receive alarms. In addition, the system is also equipped with a data storage unit for saving key data and results to ensure the security and traceability of the data. The entire system provides an intelligent and automated solution for steel pipe detection.

[0077] The standard parts used in the present invention can all be purchased from the market, and the special-shaped parts can be customized according to the description and the drawings. The specific connection methods of each part adopt conventional means such as mature bolts, rivets, welding, etc. in the prior art. The machinery, parts and equipment all adopt conventional models in the prior art, and the circuit connection adopts the conventional connection method in the prior art, which will not be described in detail here. The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

[0078] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A non-destructive testing system for steel pipes based on machine learning, characterized in that: The nondestructive testing system includes an eddy current data acquisition module, an AI data analysis module, a monitoring and control module, and a user interaction module; The eddy current data acquisition module uses eddy current detection equipment to measure N groups of eddy current signals of steel pipes in normal state and damaged state to form a data set, labels the signal sequence in the data set, and transmits the collected eddy current data to the AI ​​data analysis module; The AI ​​data analysis module includes a signal preprocessing unit, a feature extraction unit, and a damage identification unit; the signal preprocessing unit filters and denoises the collected eddy current data to improve the signal quality, discretizes the filtered and denoised eddy current data, and converts them into digital signals suitable for wavelet transform; the feature extraction unit uses a hybrid model of a convolutional neural network based on an attention mechanism and a bidirectional gated recurrent unit to extract features from the eddy current frequency domain data, and saves the trained model parameters; the damage identification unit further processes and analyzes the extracted features, and identifies the damage of the steel pipe through a machine learning model; The monitoring and control module monitors the steel pipe in real time according to the analysis results of the AI ​​data analysis module, and automatically triggers an alarm when damage is detected; The user interaction module provides a user operation interface, allowing the user to view the monitoring data and damage status of the steel pipe and receive alarm information.

2. According to claim 1, a nondestructive testing system for steel pipes based on machine learning is characterized in that: The eddy current data acquisition module includes an eddy current detection probe and a signal converter. The eddy current detection probe includes a plurality of high-sensitivity eddy current sensors fixed on the surface of the steel pipe, and collects data by detecting electromagnetic changes on the surface of the steel pipe; the signal converter is used to convert the eddy current signal into an electrical signal for collection and transmit it to the signal preprocessing unit.

3. According to claim 1, a nondestructive testing system for steel pipes based on machine learning is characterized in that: The feature extraction unit selects some wavelet coefficients as features according to task requirements. These features can reflect the characteristics of the signal at different frequencies and time scales. The feature extraction method based on wavelet transform is used to further decompose the signal into smaller sub-bands, and extract statistical features in the sub-bands, including mean and variance. The feature extraction method based on wavelet energy spectrum is used to calculate the energy of wavelet transform coefficients at different scales to obtain the frequency domain features of the signal at different scales. The obtained features are then input into a hybrid model of a convolutional neural network based on an attention mechanism and a bidirectional gated recurrent unit. The CNN layer further refines the features and captures local dependencies. At the same time, the attention mechanism ensures that the model focuses on the most critical features for damage identification. The BiGRU layer processes these features to capture the temporal correlation between features and fuses features from different sources through the Concatenate function. The model is then nonlinearly fitted through a neural network to improve the classification accuracy and generalization ability of the model. That is, after the model training is completed, the outputs of the CNN and BiGRU are fused through a fully connected layer, which uses a softmax function to convert the output vector into a probability distribution.

4. The nondestructive testing system for steel pipes based on machine learning according to claim 1, characterized in that: The damage identification unit uses the model constructed by the feature extraction unit to judge the degree of fit of the measured signal, and the probability of the model output is used to determine whether the signal indicates the presence of damage; the probability threshold is specifically set to 0.5, and the characteristic vector of the measured signal is nonlinearly mapped through the model constructed by the feature extraction unit. If the output probability corresponding to the damage category is greater than 0.5, the signal label is marked as 1, indicating that the component corresponding to the signal is damaged; if the output probability is less than 0.5, the signal label is marked as 0, indicating that the component corresponding to the signal is not damaged.

5. A steel pipe nondestructive testing system based on machine learning according to claim 1 or 2, characterized in that: in, Wavelet transform and hybrid model based on CNN and BiGRU are used to extract key features from the preprocessed data. For a signal f(t), its continuous wavelet transform is defined as: Among them, a is the scale parameter, which determines the expansion and contraction of the wavelet function; b is the translation parameter, which determines the movement of the wavelet function on the time axis; ψ(t) is the mother wavelet, which is a function with finite energy and zero average value; ψ*(t) represents the complex conjugate of ψ(t); CWT f,ψ (a, b) represents the transform coefficients at scale a and translation b, which contain information about the signal at different times and scales; The calculation process of the BiGRU unit is as follows: z t =σ(W z ·[h t-1 ,x t ]) r t =σ(W r ·[h t-1 ,x t ]) Among them, z t represents the update gate, r t Represents the reset gate; Controls the amount of information from the previous node that is written to the current candidate set GRU will reset the previous gate output r t and the input state h at the previous moment t-1 , input x at the current time t , perform calculations to form new memory content for the current node GRU combines the input state h of the previous moment t-1 With the current new memory content To determine the state of the current node h t ; σ is the Sigmoid function, tanh is the hyperbolic tangent function, W is the weight, x t represents the data input at time t, h t-1 Represents the data input at the last moment, h t is the output at the current moment, W, V, U represent weight matrices, b and c represent bias matrices, and y t is the final output of BiGRU.

6. A steel pipe nondestructive testing system based on machine learning according to claim 1 or 2, characterized in that: The nondestructive testing system also includes a remote monitoring module, which is used to realize remote steel pipe status monitoring and damage identification result viewing. The remote monitoring module is connected to the monitoring and control module through a network, allowing remote users to view the monitoring data and damage status of the steel pipe in real time and receive alarm information through a user interaction module and a special remote monitoring interface.

7. A steel pipe nondestructive testing system based on machine learning according to claim 1 or 2, characterized in that: The nondestructive testing system also includes an intelligent maintenance suggestion module, which provides users with maintenance suggestions and repair plans for steel pipes based on the damage identification results of the AI ​​data analysis module. The intelligent maintenance suggestion module considers the current status, damage type, damage degree and development trend of the steel pipe, combines historical maintenance data and expert knowledge, and generates personalized maintenance strategies for users to reduce downtime and repair costs.

8. A steel pipe nondestructive testing system based on machine learning according to claim 1 or 2, characterized in that: The AI ​​data analysis module also includes a data storage unit for storing the collected eddy current data, features extracted by wavelet transform, and damage identification results.

9. A nondestructive testing method for steel pipes based on machine learning, characterized in that: The specific steps include: S1. Use eddy current testing equipment to measure N groups of eddy current signals of steel pipes in normal state and damaged state to form a data set, and label the signal sequence in the data set; S2. Filter and denoise the eddy current data collected in step S1 to improve signal quality, discretize the filtered and denoised eddy current data, and convert them into digital signals suitable for wavelet transform; extract features from the eddy current frequency domain data using a hybrid model of a convolutional neural network (CNN) based on an attention mechanism and a bidirectional gated recurrent unit (BiGRU), and save the trained model parameters; further process and analyze the extracted features, and identify the damage of the steel pipe through a machine learning model; S3. Monitor the steel pipe in real time according to the damage identification result in step S2, and automatically trigger an alarm when damage is detected; S4. Store and display the results identified in step S2 and the information monitored in step S3.

10. The method for nondestructive testing of steel pipes based on machine learning according to claim 9, characterized in that: The S2 step specifically includes the following steps: S201. Signal preprocessing: filtering and denoising the collected eddy current data to improve the signal quality, discretizing the filtered and denoised eddy current data and converting them into digital signals suitable for wavelet transform; performing wavelet transform on the discretized signal, obtaining wavelet coefficients at different scales by calculating the inner product of the signal and the wavelet function, performing multi-scale analysis on the signal according to the set scale parameter, and extracting wavelet coefficients of different frequency components; S202. Feature extraction: According to the task requirements, some wavelet coefficients are selected as features. These features can reflect the characteristics of the signal at different frequencies and time scales. The feature extraction method based on wavelet transform is used to further decompose the signal into smaller sub-bands and extract the statistical features in the sub-bands, including the mean and variance. The feature extraction method based on wavelet energy spectrum is used to calculate the energy of wavelet transform coefficients at different scales to obtain the frequency domain features of the signal at different scales. The obtained features are then input into a hybrid model of convolutional neural network based on attention mechanism and bidirectional gated recurrent unit. The CNN layer further refines the features and captures local dependencies. At the same time, the attention mechanism ensures that the model focuses on the most critical features for damage identification. The BiGRU layer processes these features to capture the temporal correlation between features and fuses features from different sources through the Concatenate function. Then, the model is nonlinearly fitted through the neural network to improve the classification accuracy and generalization ability of the model. That is, after the model training is completed, the outputs of CNN and BiGRU are fused through a fully connected layer. This layer uses the softmax function to convert the output vector into a probability distribution. S203. Damage identification: The model constructed in step S202 is used to judge the degree of fit of the measured signal, and the probability of the model output is used to determine whether the signal indicates the presence of damage.

11. The method for nondestructive testing of steel pipes based on machine learning according to claim 10, characterized in that: In S203, the probability threshold is set to 0.5, and the characteristic vector of the measured signal is nonlinearly mapped through the model in step 2. If the output probability corresponding to the damage category is greater than 0.5, the signal label is marked as 1, indicating that the component corresponding to the signal is damaged; if the output probability is less than 0.5, the signal label is marked as 0, indicating that the component corresponding to the signal is not damaged.

Images