A method for evaluating the aging level of heated surfaces based on feature transfer

Abstract

This application relates to a method for evaluating the aging level of heated surfaces based on feature transfer. The method includes the following steps: S10, setting up a sample assembly; S20, placing multiple sample assemblies in the device to be tested; S30, obtaining a sample dataset; S40, obtaining first sample data and first in-service data at that moment; S50, obtaining a correction value; S60, correcting the sample dataset using the correction value to obtain a corrected dataset. The feature transfer-based method for evaluating the aging level of heated surfaces described in this application has the advantage of convenient evaluation.

Description

Technical Field

[0001] This application relates to the field of nondestructive testing technology, and in particular to a method for evaluating the aging level of heated surfaces based on feature migration. Background Technology

[0002] With the advancement of industrialization, more high-temperature pressure-bearing equipment is widely used in industries such as power, metallurgy, power machinery, and petrochemicals. The safe and stable operation of this equipment is closely related to production safety, life safety, and economic development. Therefore, monitoring the service status and conducting risk assessments of materials used in high-temperature pressure-bearing equipment is a crucial means of ensuring production safety and economic benefits. In the thermal power generation industry, under the backdrop of energy conservation and emission reduction requirements, more large-scale, high-parameter power plant units are being put into operation, which places higher demands on the performance of heat-bearing pressure-bearing materials.

[0003] Currently, the service status monitoring of high-temperature pressure-bearing materials mainly relies on preventative planned maintenance (major overhauls, minor repairs, periodic maintenance, and emergency repairs). This involves inspecting the surface condition, geometric dimensions, metallographic structure, and mechanical properties of the monitored components during service to assess the material's failure state. However, current failure detection methods for high-temperature pressure-bearing materials all have varying degrees of shortcomings. Destructive testing methods require pipe cutting and sampling, which affects continuous equipment operation, has long testing cycles, and requires improved sample representativeness. Traditional non-destructive testing suffers from radiation risks and limitations in detecting only macroscopic defects or having limited depth of detection.

[0004] Therefore, there is a need to develop a new generation of non-destructive testing methods capable of rapidly / in-situ predictive analysis of the failure trends of metallic materials. Laser-induced breakdown spectroscopy (LIBS) is a technique that can meet these needs. By micro-ablating the object under test, it obtains breakdown spectral information characterizing the matrix properties of the material, enabling characteristic testing with minimal damage to the object. Furthermore, the measuring devices constituting this technology can be integrated into portable or remote measuring equipment, thus meeting the needs of convenient manual testing during online monitoring or maintenance of in-service materials, ensuring the safe operation of high-temperature and pressure-bearing critical equipment. However, this technique requires the system to be in a shutdown or maintenance state.

[0005] High-temperature pressure-bearing steel in service often operates in extremely harsh environments. The main causes of its failure and aging are changes in its metallographic structure, carbide precipitation and growth, and micro-area distribution of related elements under high-temperature service. The spectra of steel at different aging levels can reflect its current aging characteristics. However, power plants have long replacement cycles for high-temperature pressure-bearing equipment, and the aging levels of the replaced heated surface metal samples are relatively uniform. Furthermore, obtaining actual steel samples requires cutting and grinding, making the process complex. Therefore, in practical research, obtaining a sufficient number of high-temperature pressure-bearing steel samples with a wide range of aging levels under actual service conditions is difficult. This makes it challenging to obtain sufficient data to support data comparison, spectral characteristic analysis, and model building during the research process. Summary of the Invention

[0006] Therefore, the purpose of this application is to provide a method for evaluating the aging level of heated surfaces based on feature migration, which has the advantage of being able to detect at any time, thus facilitating the acquisition of test results.

[0007] One aspect of this application provides a method for evaluating the aging level of a heated surface based on feature transfer, comprising the following steps:

[0008] S10. Set up the sample assembly;

[0009] S20. Place the plurality of the sample assemblies into the device to be tested;

[0010] S30. At fixed time intervals, the sample assembly is taken out and aging is tested to obtain a sample data set.

[0011] S40. In the shutdown state, the aging of the sample assembly and the in-service equipment are detected respectively, and the first sample data and the first in-service data at that moment are obtained respectively.

[0012] S50. Compare the first sample data with the first in-service data, and correct the first sample data using the first in-service data to obtain the correction value;

[0013] S60. Correct the sample dataset using correction values ​​to obtain a corrected dataset.

[0014] The feature migration-based aging level evaluation method for heated surfaces described in this application involves setting up a sample assembly, placing it alongside a pressure vessel, and operating them simultaneously. After a set time interval, the sample assembly is removed, and its surface condition is inspected using LIBS (Liquidity-Induced Spatial Analysis). The obtained data is recorded and stored. This process of inspection and recording is repeated multiple times at the same time intervals until the system is shut down. Even when the system is shut down, the aging condition of both the sample assembly and the in-service equipment is inspected using LIBS.

[0015] Then, the two sets of data are compared, the deviation between the two sets of data is analyzed, and corrections are made based on the deviation to make the data consistent with the values ​​of the sample assembly and the in-service equipment.

[0016] After correcting the dataset, a corrected dataset is obtained, which is then used to evaluate in-service equipment.

[0017] Therefore, the feature transfer-based method for evaluating the aging level of heated surfaces in this application not only facilitates the acquisition of more data, but also, through the correction of the data, more accurately reflects the actual situation, thus enabling a more accurate evaluation. Furthermore, due to the larger and richer data volume, it also allows for more accurate predictions of subsequent aging conditions of the heated surfaces. In other words, the technical solution of this application is not only more convenient for detection, but also allows for prediction of similar situations.

[0018] Furthermore, it also includes the following steps:

[0019] S70. Repeat steps S30-S50 to obtain multiple correction values ​​at multiple time intervals, and obtain the correction curve based on these multiple correction values.

[0020] Furthermore, it also includes the following steps:

[0021] S80. Based on the correction curve, establish an aging level model.

[0022] Furthermore, the sample assembly includes a main pipe, a pressure-bearing end cap, a mounting cap, and a sample tube;

[0023] The pressure-bearing end cap has multiple mounting holes, and one end of the sample tube is mounted on the mounting hole of the pressure-bearing end cap;

[0024] The other end of the sample tube passes through and is fitted onto the mounting cover;

[0025] The pressure-bearing end cap has multiple branch through holes and a main through hole, which are connected to the multiple branch through holes respectively; the multiple branch through holes are connected to the corresponding mounting holes one by one.

[0026] The main pipe is connected to the pressure-bearing end cap, and the main pipe communicates with the main through hole;

[0027] An inner cavity is formed inside the sample tube;

[0028] The main through hole is connected to one end of multiple branch through holes, and the other end of each branch through hole is connected to the inner cavity of the sample tube.

[0029] Furthermore, the sample assembly also includes a first fixing post and a second fixing post, the first fixing post being installed in the middle of the pressure-bearing end cap, and the second fixing post being installed in the middle of the mounting cap; one end of the second fixing post is sleeved on the first fixing post;

[0030] The end of the sample tube is formed with a protrusion that abuts against the surface of the mounting cover to prevent the sample tube from becoming loose from the mounting cover.

[0031] A sealing ring is provided at the connection between the sample tube and the mounting hole.

[0032] Furthermore, the sample assembly also includes a mounting base, which is mounted on the mounting cover;

[0033] Alternatively, the mounting base is mounted on the mounting cover and the pressure-bearing end cover.

[0034] Furthermore, in step S30, aging detection is performed, including detecting the aging of the heated surface using LIBS.

[0035] Furthermore, it also includes step S90, evaluating the aging level of the test sample according to the aging level model.

[0036] Furthermore, in step S80, the aging level model is generated based on the correction curve through feature transfer.

[0037] To better understand and implement this application, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0038] Figure 1 A flowchart illustrating an exemplary feature transfer-based method for evaluating the aging level of a heated surface in this application;

[0039] Figure 2 This is a flowchart illustrating another exemplary feature transfer-based method for evaluating the aging level of a heated surface in this application.

[0040] Figure 3 This is a three-dimensional structural schematic diagram of an exemplary sample assembly of this application;

[0041] Figure 4This is a three-dimensional structural schematic diagram of an exemplary sample assembly of this application from another perspective;

[0042] Figure 5 This is a cross-sectional view of an exemplary sample assembly of this application;

[0043] Figure 6 This is a three-dimensional structural schematic diagram of another exemplary sample assembly of this application;

[0044] Figure 7 This is a cross-sectional view of another exemplary sample assembly of this application;

[0045] Figure 8 This is a side view of an exemplary pressure-bearing end cap of this application;

[0046] Figure 9 This is a cross-sectional view of half of the BB side of the pressure-bearing end cap, which is an example of this application.

[0047] Figure 10 This is a cross-sectional view of the other half of the BB side of the pressure-bearing end cap, which is an example of this application. Detailed Implementation

[0048] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0049] To obtain steel samples with different aging degrees and enrich their spectral data, the study of the aging characteristics of high-temperature pressure-bearing steel can initially focus on artificially set sample tubes 30. These artificially set sample tubes 30 can simulate real samples from in-service equipment to a certain extent, exhibiting similarities in metallographic structure, aging characteristics, and LIBS spectral information. By studying the spectra of artificially set sample tubes 30, the analyzed and summarized aging spectral characteristics of the steel samples can be extended to actual steel samples and verified using spectral correction. This approach aims to address the difficulties in obtaining actual steel samples and the limited range of aging levels in the research.

[0050] Please see Figures 1-10 This application exemplifies a method for evaluating the aging level of a heated surface based on feature transfer, comprising the following steps:

[0051] S10. Set up the sample assembly.

[0052] Please see Figures 3-10 In some preferred embodiments, the sample assembly includes a main tube 10, a pressure-bearing end cap 20, a mounting cap 40, and a sample tube 30.

[0053] The pressure-bearing end cap 20 has a plurality of mounting holes C, and one end of the sample tube 30 is mounted on the mounting hole C of the pressure-bearing end cap 20;

[0054] The other end of the sample tube 30 passes through and is sleeved on the mounting cover 40;

[0055] The pressure-bearing end cap 20 has multiple branch through holes A and a main through hole, which are connected to the multiple branch through holes A respectively; the multiple branch through holes A are connected to the corresponding mounting holes C one by one.

[0056] The main pipe 10 is connected to the pressure-bearing end cap 20, and the main pipe 10 is in communication with the main through hole;

[0057] An inner cavity is formed inside the sample tube 30;

[0058] The main through hole is connected to one end of multiple branch through holes A, and the other end of each branch through hole A is connected to the inner cavity of the sample tube 30.

[0059] In some preferred embodiments, the sample assembly further includes a first fixing post 51 and a second fixing post 52, wherein the first fixing post 51 is installed in the middle of the pressure-bearing end cover 20, and the second fixing post 52 is installed in the middle of the mounting cover 40; one end of the second fixing post 52 is sleeved on the first fixing post 51;

[0060] The end of the sample tube 30 is provided with a protrusion 31, which abuts against the surface of the mounting cover 40 to prevent the sample tube 30 from becoming detached from the mounting cover 40.

[0061] A sealing ring is provided at the connection between the sample tube 30 and the mounting hole C.

[0062] Compared with existing technologies, the sample assembly of this application is more suitable for testing high-temperature pressure equipment.

[0063] In some preferred embodiments, the sample assembly further includes a mounting base 60, which is mounted on the mounting cover 40;

[0064] Alternatively, the mounting base 60 is mounted on the mounting cover 40 and the pressure-bearing end cover 20.

[0065] Furthermore, the mounting base 60 has two structures, one for horizontal placement, such as... Figure 3 and4 As shown, the mounting base 60 is installed on the mounting cover 40 and the pressure-bearing end cover 20. One type is for vertical placement, such as... Figure 6 As shown, the mounting base 60 is installed on the bottom surface of the mounting cover 40 at this time.

[0066] S20. Place the plurality of the sample assemblies into the device to be tested;

[0067] S30. At fixed time intervals, the sample assembly is taken out and aging is tested to obtain a sample data set.

[0068] After pretreatment by grinding and polishing, the samples to be tested were subjected to LIBS detection to obtain spectral data of continuous laser pulses at 30 different measurement points in each sample tube.

[0069] Differences in the mechanical properties of metal surfaces, such as hardness, will cause variations in laser ablation samples. Furthermore, slight differences in the curvature of metal pipe surfaces and variations in surface layer structure can all affect the interaction between the laser and the sample, thus interfering with the correlation between LIBS data and the properties of heat-resistant steel. Therefore, preprocessing of the acquired LIBS data is necessary to eliminate error interference.

[0070] The effects of surface roughness, different sizes of solid particles, and surface scattering on the test sample can be eliminated by using standard normal transformation (SNV).

[0071] In addition, baseline drift can be corrected using multivariate scattering correction (MSC) and the scattering effects caused by inhomogeneity of the metal surface can be eliminated.

[0072] In some preferred embodiments, in step S30, aging condition detection is performed, including detecting the aging condition of the heated surface using LIBS.

[0073] S40. In the shutdown state, the aging of the sample assembly and the in-service equipment are detected respectively, and the first sample data and the first in-service data at that moment are obtained respectively.

[0074] Because the artificially set sample tube 30 and the real sample from the in-service equipment have certain characteristic differences, the artificially set sample tube 30 cannot completely simulate the metallic characteristics of the real sample from the in-service equipment. Although the spectra of the two have certain similarities and correlations, some characteristic differences still exist. Therefore, it is necessary to perform characteristic analysis on the spectral data of the two samples.

[0075] Specifically, the correlation between the spectra of the two types of samples was calculated using the Pearson correlation coefficient.

[0076] The intensity differences of characteristic spectral lines were analyzed and corrected. Random samples were taken from the spectral data of artificially set sample tube 30 and real samples of in-service equipment at different aging levels. The ratio of the spectral intensity of the artificially set sample tube 30 to that of the real sample of in-service equipment at each wavelength point was closer to 1 under the same aging level. This indicates that the spectral intensity of the uncorrected artificially set sample tube 30 and the real sample of in-service equipment at that wavelength point is closer. In this case, the artificially set sample tube 30 can better characterize the spectral feature information of the real sample of in-service equipment.

[0077] The average spectral intensity ratio under the sampling data can be calculated using the above formula from the randomly sampled spectral data, and is called the spectral intensity correction curve of the artificially set sample tube 30 under different aging levels.

[0078] S50. Compare the first sample data with the first in-service data, and correct the first sample data using the first in-service data to obtain the correction value;

[0079] S60. Correct the sample dataset using correction values ​​to obtain a corrected dataset.

[0080] The feature migration-based aging level evaluation method for heated surfaces described in this application involves setting up a sample assembly, placing it alongside a pressure vessel, and operating them simultaneously. After a set time interval, the sample assembly is removed, and its surface condition is inspected using LIBS (Liquidity-Induced Spatial Analysis). The obtained data is recorded and stored. This process of inspection and recording is repeated multiple times at the same time intervals until the system is shut down. Even when the system is shut down, the aging condition of both the sample assembly and the in-service equipment is inspected using LIBS.

[0081] Then, the two sets of data are compared, the deviation between the two sets of data is analyzed, and corrections are made based on the deviation to make the data consistent with the values ​​of the sample assembly and the in-service equipment.

[0082] After correcting the dataset, a corrected dataset is obtained, which is then used to evaluate in-service equipment.

[0083] Therefore, the feature transfer-based method for evaluating the aging level of heated surfaces in this application not only facilitates the acquisition of more data, but also, through the correction of the data, more accurately reflects the actual situation, thus enabling a more accurate evaluation. Furthermore, due to the larger and richer data volume, it also allows for more accurate predictions of subsequent aging conditions of the heated surfaces. In other words, the technical solution of this application is not only more convenient for detection, but also allows for prediction of similar situations.

[0084] In some preferred embodiments, the steps further include:

[0085] S70. Repeat steps S30-S50 to obtain multiple correction values ​​at multiple time intervals, and obtain the correction curve based on these multiple correction values.

[0086] In some preferred embodiments, the steps further include:

[0087] S80. Based on the correction curve, establish an aging level model.

[0088] In some preferred embodiments, in step S80, the aging level model is generated based on the correction curve by means of feature transfer.

[0089] For the manually set sample tube 30 spectrum β (source domain data) and the real sample spectrum α (target domain data) of the in-service equipment, the correlation is greatly improved after intensity correction. Then, spatial mapping φ is performed through a polynomial kernel function to further achieve the goal of feature transfer, so that the source domain data and target domain data after mapping and transfer have the same or very similar distribution, that is, P(φ(β))≈P(φ(α)). The specific algorithm flow is as follows:

[0090] (1) Introduce the function D describing the distance between domains. For two one-dimensional vector sets X and Y of the same length, we have the following equation:

[0091]

[0092] Where n and m represent the number of vectors. D(X,Y) is the average distance between the vectors in each of the two vector sets.

[0093] (2) High-dimensional space mapping is performed using a polynomial kernel function, as shown in equation (9).

[0094] κ(x,y)=(αx T y+c) d

[0095] Where x and y are spectral data from two different domains, α is the slope, c is a constant term, and d is the polynomial order.

[0096] (3) Calculate the domain distance between the spectral data of the artificially set sample tube 30 with the same aging level after mapping and the spectral data of the real sample of the in-service equipment.

[0097] D n =D(φ(α) n ),φ(β n ), where n represents the aging level.

[0098] By iteratively calculating by changing the parameters of the kernel function, the D values ​​for different aging levels can be optimized.n All parameters should be minimized as much as possible, and the optimal parameters for the multi-form kernel function mapping should be determined.

[0099] (4) The optimal kernel mapping function is obtained from the solved optimal parameters, and the new spectral data after high-dimensional space mapping is obtained.

[0100] α * =φ(α),β * =φ(β)

[0101] In some preferred embodiments, the method further includes step S90: evaluating the aging level of the test sample according to the aging level model.

[0102] The model building process involves establishing the relationship between the spectrum and the aging level. The basic process involves building a model based on spectral data from multiple samples with known aging levels, and then using the model to evaluate the aging level of the target sample. The data source for modeling uses the spectrum processed as described above.

[0103] Models with good predictive performance often require a large amount of data for training. However, obtaining LIBS data from real samples of heat-resistant steel in-service equipment is difficult, resulting in limited spectral data for real samples of in-service equipment. This limited training data leads to poor model training performance. Based on the aforementioned spectral correction method, feature transfer is performed on the data from the artificially set sample tube 30. The corrected LIBS data from the artificially set sample tube 30 is then used as the training set for model training, while the LIBS data from real samples of in-service equipment is used as the validation set to fine-tune the model parameters.

[0104] This method uses a support vector machine (SVM) to build a feature analysis model, selects only the feature variables of some sample points as support vectors, and maps the training sample data points to a higher-dimensional space through a kernel function.

[0105] Furthermore, because LIBS spectral data contains rich spectral features, it is prone to the curse of dimensionality, meaning the number of feature variables far exceeds the number of samples, resulting in sparsity of the sample data in high-dimensional space. Due to the sparsity of the samples, classification models can easily find a complex hyperplane in high-dimensional space, leading to overfitting problems.

[0106] Therefore, in order to improve the prediction and generalization ability of the classification model and better understand the contribution of each variable to the classification model, it is necessary to perform feature selection on the spectral feature variables. Based on recursive feature elimination (RFE), the contribution of variables to the model is represented by the weight vector coefficients of the learner, thereby eliminating invalid and redundant variables.

[0107] In this application, in order to ensure the accuracy of the test data, each sample tube 30 is numbered and the placement position and angle of each sample tube 30 are recorded.

[0108] Compared with existing technologies and equipment, the present invention has the following advantages and beneficial effects:

[0109] (1) To address interference factors such as surface differences and environmental fluctuations, targeted spectral preprocessing methods are used to reduce the impact of surface scattering, baseline drift, and surface pits on the quality and stability of spectral data, thereby ensuring the stability and effectiveness of spectral characterization of matrix properties.

[0110] (2) By performing spectral correlation analysis and feature analysis, the LIBS data of the artificially set sample tube 30 of heat-resistant steel is spectrally corrected and data dimensionality reduced, redundant feature information is eliminated, and the correlation between the artificially set sample tube 30 and the real sample spectral data of the in-service equipment is improved.

[0111] (3) Using a feature mapping matrix, the LIBS data (source domain data) of the artificially set sample tube 30 and the LIBS data (target domain data) of the real sample of the in-service equipment are mapped to a new high-dimensional space. After mapping, the two have the same distribution and retain the original feature information.

[0112] (4) Combining spectral correction and feature mapping, the data of artificially set sample tube 30 after feature transfer is used as the training set for modeling, and a heat-resistant steel aging level prediction model based on feature transfer learning is constructed to accurately predict the aging characteristics of real samples of in-service equipment.

[0113] The embodiments described above are merely examples of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. A method for evaluating the aging level of a heated surface based on feature transfer, characterized in that, Including the following steps: S10. Set up the sample assembly; S20. Place the plurality of the sample assemblies into the device to be tested; S30. At fixed time intervals, the sample assembly is removed and aging is tested to obtain a sample data set; the aging of the heated surface is tested using LIBS. S40. In the shutdown state, the aging of the sample assembly and the in-service equipment are detected respectively, and the first sample data and the first in-service data at that moment are obtained respectively. S50. Compare the first sample data with the first in-service data, and correct the first sample data using the first in-service data to obtain the correction value; S60. Correct the sample dataset using correction values ​​to obtain a corrected dataset; S70. Repeat steps S30-S50 to obtain multiple correction values ​​at multiple time intervals, and obtain the correction curve based on these multiple correction values. S80. Generate an aging level model based on the correction curve through feature transfer. S90. Evaluate the aging level of the test sample according to the aging level model. By employing a feature mapping matrix, the LIBS data of the sample assembly and the LIBS data of the real sample of the in-service equipment are mapped into a new high-dimensional space. Combining spectral correction and feature mapping, the feature-transferred sample assembly data is used as a training set for modeling, and an aging level prediction model based on feature transfer learning is constructed to accurately predict the aging characteristics of the real sample of the in-service equipment.

2. The method for evaluating the aging level of a heated surface based on feature transfer according to claim 1, characterized in that: The sample assembly includes a main pipe, a pressure-bearing end cap, a mounting cap, and a sample tube; The pressure-bearing end cap has multiple mounting holes, and one end of the sample tube is mounted on the mounting hole of the pressure-bearing end cap; The other end of the sample tube passes through and is fitted onto the mounting cover; The pressure-bearing end cap has multiple branch through holes and a main through hole, which are connected to the multiple branch through holes respectively; the multiple branch through holes are connected to the corresponding mounting holes one by one. The main pipe is connected to the pressure-bearing end cap, and the main pipe communicates with the main through hole; An inner cavity is formed inside the sample tube; The main through hole is connected to one end of multiple branch through holes, and the other end of each branch through hole is connected to the inner cavity of the sample tube.

3. The method for evaluating the aging level of a heated surface based on feature transfer according to claim 2, characterized in that: The sample assembly further includes a first fixing post and a second fixing post. The first fixing post is installed in the middle of the pressure-bearing end cap, and the second fixing post is installed in the middle of the mounting cap. One end of the second fixing post is sleeved on the first fixing post. The end of the sample tube is formed with a protrusion that abuts against the surface of the mounting cover to prevent the sample tube from becoming loose from the mounting cover. A sealing ring is provided at the connection between the sample tube and the mounting hole.

4. The method for evaluating the aging level of a heated surface based on feature transfer according to claim 2, characterized in that: The sample assembly also includes a mounting base, which is mounted on the mounting cover; Alternatively, the mounting base is mounted on the mounting cover and the pressure-bearing end cover.

Images