Pulse vortex pipeline wall thickness detection method based on RSBO optimized LSTM network

Abstract

The invention provides a pulse eddy current pipeline wall thickness detection method based on an RSBO optimized LSTM network, and the method comprises the steps: estimating the thickness of a pipeline through employing a pulse eddy current signal, determining an interval where a measurement signal is reserved through employing a data sample interception length evaluation criterion based on a least square fitting algorithm, inputting the processed signal into an LSTM model, and carrying out the detection of the thickness of the pipeline. And carrying out hyper-parameter setting on the LSTM model by adopting combined optimization based on random search and Bayesian optimization.

Description

Technical Field

[0001] The present invention belongs to the technical fields of PEC detection, pipeline thickness measurement, etc., and specifically relates to a pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network. Background Art

[0002] Corrosion of pressure vessels, especially the thinning of the wall caused by corrosion of steel pressure vessels, poses a serious threat to the safety and reliability of industrial operations. As a common pipeline structure in pressure vessels, effective detection methods are urgently needed. PEC detection technology has the characteristics of non-contact, deep penetration depth and wide spectrum, and has been widely used for thickness measurement of metal specimens, especially under high thickness and high lift-off conditions. For the detection of pipeline structures, the eddy current transient diffusion signal affected by the plate properties can be received by a coil or a magnetic field sensor, and the thickness change of the sample tube will be reflected by this signal.

[0003] In the prior art, the relationship between signal characteristics and thickness is studied and the inverse problem is solved by optimization method to estimate the thickness based on the received signal. Y.Shin et al. proposed to estimate the sample thickness and conductivity parameters by the peak point and zero-crossing time characteristics of the received signal (see Shin, Y.-K.; Choi, D.-M.; Kim, Y.-J.; Lee, S.-S., Signal characteristics of differential-pulsed eddy current sensors in the evaluation of plate thickness. NDT&EInternational 2009, 42, (3), 215-221.), but this method is affected by the lift-off height between the sensor probe and the sample. M. Fan et al. proposed that the phase spectrum of the converted received signal into the frequency domain is not affected by the liftoff effect (see Fan, M.; Cao, B.; Sunny, AI; Li, W.; Tian, ​​G.; Ye, B., Pulsed eddy current thickness measurement using phase features immune to liftoff effect. Ndt&E International 2017, 86, 123-131.). This also applies to the phase spectrum of multi-frequency eddy current testing if the liftoff change is not obvious (see Yin, W.; Peyton, AJ, Thickness measurement of metallic plates with an electromagnetic sensor using phase signature analysis. IEEE Transactions on Instrumentation and Measurement 2008, 57, (8), 1803-1807.).A frequency feature mixing method combining the characteristics of each frequency is proposed to obtain a linear relationship between the characteristics and the sample thickness (see Ge, J.; Yusa, N.; Fan, M., Frequency component mixing of pulsed or multi-frequency eddycurrent testing for nonferromagnetic plate thickness measurement using a multi-gene genetic programming algorithm. NDT&EInternational 2021, 120, 102423.).

[0004] In recent years, data-driven deep learning (DL) methods have been frequently used in eddy current testing research. DL methods learn the nonlinear mapping between the measured values ​​and the physical properties of the sample, providing rich data samples that can be obtained from intensive measurements and numerical simulations (see Zhu, P.; Cheng, Y.; Banerjee, P.; Tamburrino, A.; Deng, Y., A novel machine learning model for eddy current testing with uncertainty. ndt&e International 2019, 101, 104-112. and Kucheryavskiy, S.; Egorov, A.; Polyakov, V., Coupling analytical models and machine learning methods for fast and reliable resolution of effects in multifrequency eddy-current sensors. Sensors 2021, 21, (2), 618.). H. Seo et al. implemented a deep neural network for pipe thickness estimation, with a dataset of 720 samples (see Seo, HG; Jun, JH; Park, DG; Shin, JW, PipeThickness Estimation byDeep Learning of Pulsed Eddy Current Time-Series Data. Journal of the Korean Society for Nondestructive Testing 2021, 41, (3),164-171.).M. Tian et al. embedded a convolutional neural network into a customized PEC system to estimate the thickness of aluminum and mild steel plates (see Meng, T, Xiong, L, Zheng, X, et al. Real-time automatic thickness recognition using pulse eddycurrent with deep learning. 2023 IEEE International Instrumentation and Measurement Technology Conference (I2MTC). IEEE, 2023: 1-6.). The dataset consists of more than 10k samples with various PEC measurement configurations and is available online. The richness of PEC measurement samples helps to achieve accurate and efficient sample thickness estimation.

[0005] The optimization algorithm can be used to solve the least squares problem of the received signal between measurement and calculation, and the sample thickness is obtained by iteration. X. Chen et al. studied the inverse PEC problem and the optimization method for simultaneously estimating thickness and electromagnetic properties (see Chen, X.; Li, J.; Wang, Z., Inversion method in pulsed eddy current testingfor wallthickness of ferromagnetic pipes. IEEE Transactions onInstrumentation and Measurement 2020, 69, (12), 9766-9773.). Without prior knowledge, it is still difficult to provide a reasonable initial point for optimization, which affects the estimation results to a certain extent. Ulapane, Net al. proposed that the decay rate of the received signal is not affected by the lift-off and depends only on the diffusion of the eddy current in the sample (see Ulapane, N.; Alempijevic, A.; Miro, JV; Vidal-Calleja, T., Non-destructive evaluation of ferromagnetic material thickness using pulsed eddy current sensor detector coil voltage decay rate. NDT&E International 2018, 100, 108-114.). However, for small-sized samples, the position of the probe will also affect the signal of the eddy current diffusion, which will bring difficulties to the estimation of the characteristics of the tube. For example, when the probe is at the edge of the sample, an edge effect will occur. Due to the complex boundary conditions, it is very complicated to accurately simulate the various conditions of the PEC test using an analytical model. Summary of the invention

[0006] Therefore, in view of the defects and deficiencies of the above prior art, the present invention proposes a pulse eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network, which is a prediction method of deep learning, based on a long and short-term memory (Long and Short Term Memory, LSTM) network optimized by random search and Bayesian optimization (Random Searching Bayesian Optimization, RSBO) combination, and the pulse eddy current (Pulsed Eddy Current, PEC) signal is used to estimate the pipeline thickness. In view of the contradiction that signal data redundancy and feature retention cannot be taken into account, a data sample interception length evaluation criterion based on the least squares fitting algorithm is introduced to improve the operational efficiency and accuracy of the model. In addition, for the LSTM model, the RSBO optimization method is used for hyperparameter setting, which is convenient for users to use, and an estimation scheme is designed, which can automatically adjust the hyperparameters of the LSTM network, avoiding the manual parameter setting process, and having good estimation performance.

[0007] The technical solution specifically adopted by the present invention to solve the technical problem is: A pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network: the pipeline thickness is estimated by pulsed eddy current signal, the interval of retaining measurement signal is determined by data sample interception length evaluation criterion based on least squares fitting algorithm, the processed signal is input into LSTM model to detect the pipeline thickness, and the hyperparameters of LSTM model are adjusted by combined optimization based on random search and Bayesian optimization.

[0008] Furthermore, the data sample interception length evaluation criteria based on the least squares fitting algorithm are specifically as follows: after fitting the measurement curve using the least squares method, the slope curve is calculated, and the standard deviation of the fitting matrix is ​​used to evaluate the overall data fluctuation of the intercept segment. The slope change of the overall data curve is evaluated by calculating the mean of each intercepted segment, and the standard deviation of the fitting matrix is ​​obtained and plotted on the same graph as the slope average, so that the intersection of the two curves is used as the basis for determining the sample interval.

[0009] Furthermore, the LSTM model consists of an input layer, multiple stacked hidden layers, a fully connected layer and a regression layer. The output of each time step of each layer is used as the input of the time step of the next layer. The feature extraction and feature transformation of the pulsed eddy current measurement signal are completed through the input layer and the hidden layer, and the final thickness prediction value is output through the nonlinear regression of the fully connected layer.

[0010] Furthermore, the training of the LSTM model adopts a learning rate decay mechanism.

[0011] Furthermore, the hyperparameters include the number of hidden layers, the number of hidden neurons Num, the initial learning rate learnRate0 of the network, the number of iterations Inter, and the training batch BatchSize.

[0012] Furthermore, the hyperparameter tuning of the LSTM model using a combination of random search and Bayesian optimization specifically includes: Randomly sample the combination of hyperparameters through random search to find the optimal hyperparameters; Bayesian optimization is used to further optimize the hyperparameters after random search by constructing a probability model of the objective function. After all iterations are completed, the hyperparameter set with the smallest target loss is recorded. When the thickness error in the validation set reaches the minimum value, the model hyperparameters are saved.

[0013] Furthermore, the specific steps of using Bayesian optimization to further optimize the hyperparameters after random search by constructing a probability model of the objective function are as follows: Step S1: Select a set of initial parameters, train and evaluate the LSTM model, and use the optimal result as the initial observation value of Bayesian optimization; Step S2: using the initial observations to train a Gaussian process model, and using the trained Gaussian process model to predict the objective function value and uncertainty of unobserved points; Step S3: using the acquisition function to determine the next sampling point; The acquisition function is from the input space , Observation Space and hyperparameter space Functions that map to real number space ; Through the observation data set The obtained posterior distribution is constructed, and the next collection point is selected by maximizing the collection function : Where s represents the collection point; Step S4: Select a new hyperparameter combination for evaluation based on the results of the acquisition function and update the Gaussian process model; Step S5: Repeat steps S1 to S4 until the iteration stop condition is met and output the optimal hyperparameter result.

[0014] And, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of a pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network are implemented as described above.

[0015] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a pulsed eddy current pipeline wall thickness detection method based on an RSBO optimized LSTM network as described above.

[0016] In view of the defects and shortcomings of the prior art, the main design points and advantages of the present invention and its preferred solution are at least reflected in: 1. Aiming at the contradiction between PEC signal data redundancy and feature retention, a data sample truncation length evaluation criterion based on the least squares fitting algorithm is introduced to improve the computational efficiency and accuracy of the model.

[0017] 2. In view of the large amount of existing sample data and multiple optimization objectives, the traversal optimization algorithm takes a lot of time, so a parameter optimization method combining RS and BO is proposed.

[0018] 3. For the LSTM model, the RSBO optimization method is used for hyperparameter tuning, which avoids the manual parameter tuning process, improves the accuracy of pipeline thickness detection, and has better estimation performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a schematic diagram of the sensor measurement position according to an embodiment of the present invention; Figure 2 This is a measurement result diagram of the sensor of the embodiment of the present invention at the center position with a lift of 6 mm and a thickness of 20 mm; Figure 3 It is a graph showing the slope standard deviation and mean value changes at different sample intervals in an embodiment of the present invention; Figure 4 This is a structural diagram of an LSTM network used for thickness estimation in an embodiment of the present invention; Figure 5 This is the basic LSTM network structure diagram; Figure 6 The figure is a flowchart of hyperparameter tuning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the features and advantages of the present invention more clearly understood, the following embodiments are specifically described in detail as follows: It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0021] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0022] The embodiment of the present invention proposes a pulsed eddy current pipeline wall thickness detection based on RSBO optimized LSTM, which realizes a fast and accurate detection method for the thickness of the sample pipe. The specific implementation steps are as follows: Step 1: PEC feature data collection and preprocessing like Figure 1 As shown in the figure, experiments were conducted on four positions of the sample {center, edge, corner, random}, six lift-off distances l∈{0,3,…,15} mm, and nine thicknesses d∈{20,25,…,60} mm. 50 PEC tests were performed on each type of experiment, and a total of 10,800 data sets were collected, each containing 270 data. The diversity of detection thickness, position and lift-off height brought some differences in pulse response. However, some pulse response curves are relatively similar. In order to improve computational efficiency and reduce data redundancy, data enhancement is required to capture the main features in the signal curve and include as many signal features as possible while keeping the data length as short as possible.

[0023] The main characteristics of the PEC signal curve include three parts: jump, attenuation, and flatness. Among them, the jump part is short and contains the information of the sample tube being tested, so the intercepted data segment should include the entire jump in the first half; and the second half of the pulse eddy current signal curve is a process from flatness to exponential decay. This transition is caused by the diffusion of eddy currents, the gradual slowing of the diffusion speed, and the reduction of intensity, so this part of the data should also be included.

[0024] In this embodiment, the sensor is placed in the center position, and a total of 54 sets of experimental data are used as representative data when the lifting distance l∈{0,3,…,15} mm and the thickness d∈{20,25,…,60} mm are used. Within a sampling period, the values ​​between the curves are different from the degree of attenuation rate. The slope of the curve corresponds to the diffusion rate of the eddy current on the surface, and the degree of slope fluctuation is considered as a judgment of the degree of difference between the curves at different time points. Therefore, after fitting the measurement curve using the least squares method, the slope curve is selected for calculation. The fitted measurement curve is used to evaluate the overall change in the slope of the data, and the standard deviation of the fitting matrix is ​​used to evaluate the overall data fluctuation of the intercept segment (for specific implementation, please refer to the literature Feng Ding, "Least squares parameter estimation and multi-innovation leastsquares methodsfor linear fitting problems from noisy data", Journalof Computational andApplied Mathematics, vol.426, pp.115107, 2023.).

[0025] from Figure 2 It can be seen that the curve slowly falls back after 0.4-1s. Therefore, this embodiment selects the time span of 0.37-0.9s (corresponding to data points 100-250), performs least squares fitting on it, and the fitting step size is 10, with 15 segments, and then obtains the fitting matrix [54×15]. By calculating the mean of each intercepted segment, the slope change of the overall data curve can be evaluated, and the standard deviation of the fitting matrix can be obtained at the same time, and the two curves are plotted in the same figure, as shown in Figure 3 shown.

[0026] It can be seen that, as time goes by, the standard deviation of the slope gradually increases, the data fluctuates slowly, and tends to a stable maximum value; while the average value of the slope becomes smaller and tends to a minimum value. The intersection of the two curves is taken as sample interval 2. At this time, the changes and values ​​between the curves are more balanced, and the measurement signal of the corresponding length is retained, that is, the time period of 0.407-0.444 s is intercepted. In order to make the intercepted data contain more characteristics of the backoff response process, this embodiment takes 0.444 s (120 data points) as the maximum value of the sample interval.

[0027] Step 2: Build an LSTM model based on RSBO The deep learning LSTM model proposed in this invention for pipeline thickness detection consists of an input layer, multiple stacked hidden layers, a fully connected layer and a regression layer. Its network structure is as follows: Figure 4As shown in the figure, the output of each time step of each layer is used as the input of the time step of the next layer. Therefore, the feature extraction and feature transformation of the eddy current measurement signal are completed through the input layer and the hidden layer, and the final thickness prediction value is output through the nonlinear regression of the fully connected layer.

[0028] The multi-hidden layer LSTM model is composed of the most basic LSTM unit, and its structure is mainly composed of a forget gate, an input gate, and an output gate. Figure 5 As shown in the figure. For example, the forget gate determines what information should be forgotten by the neuron. t and some time ago t-1 The weight calculation formula is: (1) The neuron updates its state vector through the input gate layer and the tanh layer to save the relevant information of the input sequence. The input gate layer uses the Sigmoid function to determine the value to be updated, while the tanh layer obtains the candidate value g through nonlinear transformation. t , which will be added to the state of the neuron. Combining these two steps to produce an updated state vector, the state update of the information is achieved, and the formula is: By transforming the input layer and neuron state through Sigmoid and tanh functions, neurons can filter out unnecessary information and retain the information that needs to be output. The output h t It will be passed to the next neuron for information update, and the output of the last layer of the network will be output as the final result after weight and bias calculation. The formula is: Where W and b are the weight and bias of the corresponding layer respectively, It is the Sigmoid operation.

[0029] Selecting appropriate hyperparameters can significantly improve the performance of the LSTM model. In the present invention, the hyperparameters required for the LSTM neural network structure include the number of hidden layers in the network, the number of hidden neurons Num, the initial learning rate learnRate0 of the network, the number of iterations Inter, and the training batch BatchSize. In order to achieve convergence in the later stage of deep learning network training, a learning rate decay mechanism should be adopted: (8) The specific process of hyperparameter tuning in the embodiment of the present invention is as follows Figure 6As shown in the figure, RS randomly samples the combinations of hyperparameters to find the optimal hyperparameters. Compared with grid search, it does not traverse all possible hyperparameter combinations, but randomly selects a certain number of combinations in the hyperparameter space for evaluation. Compared with swarm intelligence optimization algorithms, it has lower computational intensity and can quickly find better hyperparameter combinations in a limited time.

[0030] Therefore, this embodiment first uses RS to optimize the LSTM hyperparameter combination, divides the data set into a training set and a validation set at a ratio of 4:1, and uses the Adam gradient descent algorithm for training. The network input is 120, the output is 1, and the mean squared error (MSE) is used as the target loss function. The parameter configuration is shown in Table 2. After all iterations are completed, the hyperparameter set with the smallest target loss is set for the parameter variables. When the thickness error in the validation set reaches the minimum value, the model hyperparameters are saved.

[0031] Then BayesOPT is used to further optimize the hyperparameters after random search to improve the thickness estimation accuracy. For BO, by constructing a probability model of the objective function, the global optimal solution can be obtained under a small number of experiments, which is suitable for neural network parameter optimization with large computational complexity.

[0032] The specific steps are as follows: 1. Select a set of initial parameters, train and evaluate the model, and use the optimal result as the initial observation value for Bayesian optimization; 2. Use these observations to train a Gaussian process model and use the model to predict the objective function value and uncertainty of unobserved points; 3. Use an acquisition function to determine the next sampling point, which balances the trade-off between exploration and exploitation.

[0033] The acquisition function is from the input space , Observation Space and hyperparameter space Functions that map to real number space . This function is composed of the observed data set The obtained posterior distribution is constructed and the next evaluation point is selected by maximizing it : (9) 4. Select new hyperparameter combinations for evaluation based on the results of the acquisition function and update the Gaussian process model.

[0034] 5. Repeat steps (1)-(4) until the iteration stop condition is met and output the optimal hyperparameter results.

[0035] In a test case, different hyperparameter sets were tested with different hidden layers and neurons. Based on RS optimization, the network has the best estimation effect when the hyperparameters are set to Inter = 2000, nInter = 8, learnRate0 = 0.01, α = 0.6, Layer = 3, BatchSize = 1000, and Num = 150 per hidden layer. Under this setting, the model performs better through experimental comparison with other methods, with MAPE improved to 5.53% and MAE value improved to 1.98mm thickness estimation.

[0036] Table 1 Estimation results using different neural networks Table 2. Thickness estimation results of training set Thickness(mm) 20 25 30 35 40 45 50 55 60 quantity 944 884 943 960 944 979 943 950 953 MAPE 0.0605 0.0970 0.0466 0.0529 0.0427 0.0416 0.0385 0.0370 0.0333 Table 3. Thickness estimation results of test set Thickness(mm) 20 25 30 35 40 45 50 55 60 quantity 260 267 264 244 263 227 265 256 254 MAPE 0.0638 0.1071 0.0510 0.0591 0.0415 0.0418 0.0409 0.0373 0.0359 Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0037] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0038] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0039] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0040] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0041] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

[0042] The present invention is not limited to the above-mentioned optimal implementation mode. Anyone can derive other various forms of a pulse eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network under the inspiration of the present invention. All equal changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network, characterized in that: The pulsed eddy current signal is used to estimate the pipeline thickness, and the data sample interception length evaluation criterion based on the least squares fitting algorithm is used to determine the interval of retaining the measurement signal. The processed signal is input into the LSTM model to detect the pipeline thickness, and the combined optimization based on random search and Bayesian optimization is used to tune the hyperparameters of the LSTM model.

2. A pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network according to claim 1, characterized in that: The data sample interception length evaluation criteria based on the least squares fitting algorithm are specifically as follows: after fitting the measurement curve using the least squares method, the slope curve is calculated, and the standard deviation of the fitting matrix is ​​used to evaluate the overall data fluctuation of the intercept segment. The slope change of the overall data curve is evaluated by calculating the mean of each intercepted segment, and the standard deviation of the fitting matrix is ​​obtained and plotted on the same graph with the slope average value, so that the intersection of the two curves is used as the basis for determining the sample interval.

3. A pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network according to claim 1, characterized in that: The LSTM model consists of an input layer, multiple stacked hidden layers, a fully connected layer and a regression layer. The output of each time step of each layer is used as the input of the time step of the next layer. The feature extraction and feature transformation of the pulsed eddy current measurement signal are completed through the input layer and the hidden layer, and the final thickness prediction value is output through the nonlinear regression of the fully connected layer.

4. A pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network according to claim 3, characterized in that: The training of the LSTM model adopts a learning rate decay mechanism.

5. A pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network according to claim 1, characterized in that: The hyperparameters include the number of hidden layers, the number of hidden neurons Num, the initial learning rate learnRate0 of the network, the number of iterations Inter, and the training batch BatchSize.

6. The pulse eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network according to claim 1 is characterized in that: The hyperparameter tuning of the LSTM model using a combination of random search and Bayesian optimization specifically includes: Randomly sample the combination of hyperparameters through random search to find the optimal hyperparameters; Bayesian optimization is used to further optimize the hyperparameters after random search by constructing a probability model of the objective function. After all iterations are completed, the hyperparameter set with the smallest target loss is recorded. When the thickness error in the validation set reaches the minimum value, the model hyperparameters are saved.

7. A pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network according to claim 6, characterized in that: The specific steps of using Bayesian optimization to further optimize the hyperparameters after random search by constructing a probability model of the objective function are as follows: Step S1: Select a set of initial parameters, train and evaluate the LSTM model, and use the optimal result as the initial observation value of Bayesian optimization; Step S2: using the initial observations to train a Gaussian process model, and using the trained Gaussian process model to predict the objective function value and uncertainty of unobserved points; Step S3: using the acquisition function to determine the next sampling point; The acquisition function is from the input space , Observation Space and hyperparameter space Functions that map to real number space ; Through the observation data set The obtained posterior distribution is constructed, and the next collection point is selected by maximizing the collection function : Where s represents the collection point; Step S4: Select a new hyperparameter combination for evaluation based on the results of the acquisition function and update the Gaussian process model; Step S5: Repeat steps S1 to S4 until the iteration stop condition is met and output the optimal hyperparameter result.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of a pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network as described in any one of claims 1 to 7 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a pulsed eddy current pipeline wall thickness detection method based on RSBO optimized LSTM network as described in any one of claims 1 to 7 are implemented.

Images