P91 pipeline aging grade prediction method based on micro-magnetic characteristic parameters

Abstract

The application belongs to the field of material nondestructive testing, and particularly relates to a P91 pipeline aging grade prediction method based on micro-magnetic characteristic parameters, which comprises the following steps: a two-dimensional calibration system covering microstructure characteristics and macroscopic mechanical properties is constructed, and multi-level standard samples are prepared; multi-dimensional micro-magnetic data of each level of samples under step-by-step tensile load is collected, and a full-dimensional data set containing aging grade, stress level and micro-magnetic parameters is constructed; a ReliefF feature screening algorithm is used to identify core characteristic parameters which are highly sensitive to aging and can effectively distinguish stress interference; a back propagation neural network prediction model is constructed with the core parameters as input and the calibration grade as output, and is applied to field detection. Through core parameter screening and multi-parameter model fusion, the application realizes effective decoupling of stress and aging influence, guarantees the integrity of the pipeline structure, and significantly improves the accuracy of quantitative prediction of the aging grade.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology for materials, and in particular to a method for predicting the aging level of P91 pipes based on micromagnetic characteristic parameters. Background Technology

[0002] With the continuous evolution of the power industry towards supercritical and ultra-supercritical units, martensitic heat-resistant steel has been widely used in high-temperature and high-pressure pipelines of thermal power units due to its excellent high-temperature creep strength and oxidation resistance. As a key pressure-bearing component material, the microstructural stability of P91 steel under long-term high-temperature and high-pressure extreme conditions directly affects the operational safety of the power system. During service, the material inevitably undergoes microstructural evolution such as martensitic lath fragmentation, carbide coarsening, and the precipitation of new phases. This thermally activated microstructural aging is a significant contributing factor to the degradation of the material's mechanical properties and even to pipe rupture accidents.

[0003] Accurate assessment of the aging level of P91 pipelines is a technical prerequisite for achieving safe pipeline operation and maintenance and predicting remaining life. Traditional aging assessment schemes typically follow industry standards, aiming to classify the degree of material aging from light to severe by cutting and sampling in-service pipelines, conducting laboratory microstructural observations and macroscopic mechanical property tests. However, with the increasing intelligence of power equipment operation and maintenance, how to achieve rapid and quantitative on-site assessment of pipeline aging status under complex operating conditions without damaging the pipeline structural integrity has become an important research direction in the field of nondestructive testing of materials.

[0004] Current technologies primarily rely on the aforementioned destructive testing methods for aging assessment. These methods not only cause irreversible damage to pipeline structures but also suffer from drawbacks such as lengthy testing cycles, high on-site construction risks, and the potential for missed detections in large-scale testing. Although conventional non-destructive testing techniques such as ultrasonic testing and hardness testing are used in engineering, they have extremely low sensitivity to the microstructural evolution caused by early material aging, and the test results are easily affected by surface conditions and residual stress, resulting in predictive accuracy that cannot meet engineering requirements.

[0005] Meanwhile, emerging micromagnetic detection methods have become a research hotspot in this field due to their advantages such as high sensitivity to microstructure evolution, fast detection speed, and ability to perform large-area scanning. For example, Chinese patent application CN120064433A discloses a method for evaluating the performance degradation of heat-resistant steel based on magnetic property parameters. This method involves subjecting P91 steel specimens to accelerated aging heat treatment for different durations, establishing a criterion for judging the degree of thermal aging by combining metallographic observation and mechanical property testing, using the ReliefF algorithm to screen magnetic property parameters sensitive to aging, and establishing a mapping relationship between magnetic parameters and aging levels based on the Gaussian process regression algorithm, thus achieving preliminary non-destructive testing of the degree of thermal aging of P91 steel.

[0006] However, this technical solution has fundamental limitations: all micromagnetic data acquisition is completed under stress-free laboratory conditions, completely neglecting the interference of unavoidable working and residual stresses in operating pipelines. It needs to be clarified that the tensile and hardness tests involved in this method are only used for destructive calibration of aging levels, not for actively introducing and controlling stress variables during micromagnetic testing, and further fail to study the influence of different stress levels on micromagnetic signals and decoupling methods. In actual field conditions, pipeline internal pressure, welding residual stress, and assembly stress can cause a 20%-50% deviation in micromagnetic signals. Directly applying this method to the inspection of in-service pipelines results in an aging level prediction error exceeding 25%, failing to meet engineering accuracy requirements.

[0007] In addition, existing micromagnetic detection methods generally suffer from the following common technical problems: First, they lack an effective decoupling mechanism for working stress interference, fail to identify core characteristic parameters that have differentiated responses to stress and aging, and cannot distinguish whether changes in micromagnetic signals are caused by material aging or stress fluctuations; second, they have failed to construct a dual-dimensional standardized calibration system covering microstructure characteristics and macroscopic mechanical properties, and the sample preparation process lacks unified specifications, resulting in poor reliability of model training data; third, they only use a single signal or a simple linear model, which cannot fully explore the correlation information of multi-dimensional micromagnetic data, resulting in insufficient generalization ability and reliability of the model in complex environments.

[0008] In summary, existing technologies cannot solve the core problem of low accuracy in predicting the aging level of P91 pipelines caused by on-site stress interference. There is an urgent need to develop a method for predicting the aging level of P91 pipelines that can effectively decouple the influence of stress and achieve rapid quantitative detection on-site. Summary of the Invention

[0009] To address the shortcomings of existing micromagnetic detection methods, such as the inability to eliminate on-site stress interference and low prediction accuracy, this invention provides a method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters, which can decouple on-site stress interference.

[0010] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0011] A method for predicting the aging level of P91 pipes based on micromagnetic characteristic parameters includes the following specific steps: Step S1: Constructing a standardized calibration system for the aging level of P91 pipes: Based on relevant metallographic inspection and evaluation technical guidelines and industry standards, and combined with the microscopic evolution mechanism and mechanical property deterioration law of P91 type high-temperature aging, the aging level of P91 pipes is divided into multiple levels, and a two-dimensional standardized calibration system covering microstructure characteristics and macroscopic mechanical property parameters is established; Step S2: Preparing P91 type standard samples of different aging levels: Selecting new P91 type pipes and old pipes from the same batch as the engineering site as the base material, and preparing standard samples of multiple aging levels matching the calibration system in Step S1 through accelerated aging heat treatment experiments; the old pipes are used to verify the calibration accuracy of the accelerated aging samples, and the main body of the standard samples is prepared through accelerated aging heat treatment. Step S3: Acquire multi-dimensional micromagnetic data considering stress interference: For the standard specimens prepared in Step S2, conduct micromagnetic data acquisition experiments under step-by-step tensile loading to obtain full-dimensional micromagnetic characteristic signals under different aging levels and stress levels, constructing a full-dimensional dataset containing aging level, stress level, and micromagnetic characteristic parameters. Step S4: Screen highly sensitive aging micromagnetic characteristic parameters: For the full-dimensional micromagnetic dataset obtained in Step S3, use the ReliefF feature selection algorithm to screen features, eliminating interference parameters that are unresponsive to stress and aging level, and selecting those highly sensitive to the P91 aging level. Furthermore, it can effectively distinguish the core characteristic parameters of stress interference; Step S5 constructs a P91 pipeline aging level prediction model based on backpropagation neural network: using the core micro-magnetic characteristic parameters selected in step S4 as the input layer and the P91 pipeline aging level calibrated in step S1 as the output layer, an aging level prediction model based on backpropagation neural network is constructed; Step S6 implements pipeline on-site inspection and aging level prediction: the prediction model constructed in step S5 is integrated into the micro-magnetic inspection equipment, on-site inspection of P91 pipeline is carried out, core micro-magnetic characteristic parameters are collected and input into the prediction model, and the aging level prediction value of the area to be tested is directly output.

[0012] Preferably, in step S1, the microstructure characteristics are characterized by metallographic microscopy, scanning electron microscopy, electron backscatter diffraction, and energy dispersive spectroscopy. The characteristics of the characteristics include the morphology and orientation of martensite laths, the proportion of ferrite phase, the equivalent circle diameter of grains, the precipitation and coarsening state of carbides, and the aggregation and distribution characteristics of specific alloying elements.

[0013] Preferably, the microstructure characteristics corresponding to the first aging level are: martensite in complete lath shape with clear lath orientation, no ferrite formation, fine and dispersed carbide precipitation within and at the lath boundaries, no obvious carbide aggregation at the grain boundaries, grain equivalent circle diameter area weighted average not greater than 6 μm, and uniform distribution of alloying elements.

[0014] Preferably, the microstructure characteristics corresponding to the second aging level are that the martensite is still lath-shaped, the lath orientation is slightly dispersed, a small amount of independent small ferrite is mixed in, granular carbide precipitation occurs at lath boundaries and grain boundaries, and the area weighted average of the equivalent circle diameter of the grain is within the first preset size range of 6μm to 10μm.

[0015] Preferably, the microstructure characteristics corresponding to the third aging level are that the martensite laths begin to break down, the orientation is obviously dispersed, large ferrite blocks appear, the number of carbides decreases and the size coarsens, spherical carbides appear, and the area-weighted average of the equivalent circle diameter of the grains is within the second preset size range of 10μm to 15μm.

[0016] Preferably, the microstructure characteristics corresponding to the fourth aging level are as follows: only a small amount of martensite traces remain, a large amount of ferrite phase is generated, the microstructure is mainly ferrite, the number of spherical carbides increases and chain-like carbides appear, and the area-weighted average of the equivalent circle diameter of the grains is located in the third preset size range of 15μm to 25μm.

[0017] Preferably, the microstructure characteristics corresponding to the fifth aging level are that the martensite has completely disappeared, the structure is a mixture of ferrite and carbides, the carbides are obviously coarsened and appear in chain or strip shape, and the area-weighted average of the equivalent circle diameter of the grains is greater than 25 μm.

[0018] Preferably, in step S1, the macroscopic mechanical performance parameters are calibrated through room temperature tensile test, preset high temperature tensile test and room temperature Charpy impact test. The calibration indicators include room temperature tensile strength, preset high temperature tensile strength, yield strength and impact energy, and clarify the mechanical performance range and reliability judgment rules corresponding to multiple aging levels.

[0019] Preferably, the room temperature tensile strength corresponding to the first aging level is not less than a first preset strength threshold; the probability that the room temperature tensile strength corresponding to the second aging level is within the first preset strength range is not less than a first preset probability threshold; the probability that the room temperature tensile strength corresponding to the third aging level is within the second preset strength range is not less than a second preset probability threshold; the probability that the room temperature tensile strength corresponding to the fourth aging level is within the third preset strength range is not less than a third preset probability threshold; and the probability that the room temperature tensile strength corresponding to the fifth aging level is not greater than the second preset strength threshold is not less than a fourth preset probability threshold.

[0020] Preferably, in step S2, the heat treatment process for aging samples of different grades is as follows: the first aging sample is the untreated base material; the second aging sample is heated at 810℃ for 180 min and then cooled; the third aging sample is heated at 820℃ for 90 min and then cooled; the fourth aging sample is heated at 820℃ for 160 min and then cooled; the fifth aging sample is heated at 1055℃ for 130 min and then cooled to 740℃, then heated for another 480 min and then cooled.

[0021] Preferably, in step S3, the step tensile test uses the yield strength of the specimen as the boundary, sets multiple sets of gradient stress loads in the elastic range, and sets multiple sets of gradient stress loads in the range from yield strength to tensile strength, with each load point held for 60s.

[0022] Preferably, the micromagnetic signals acquired in step S3 include tangential magnetic field strength, Barkhausen noise, incremental permeability, and multi-frequency eddy currents, covering all detectable micromagnetic characteristic parameters.

[0023] Preferably, in step S4, the ReliefF feature selection algorithm iteratively calculates the weight value of each micromagnetic feature parameter and selects the core feature parameters with the highest weight values, including the maximum value of Barkhausen noise, the mean value of Barkhausen noise, the equivalent remanence of Barkhausen noise, and the peak width at a specific proportion of the peak value of Barkhausen noise.

[0024] Preferably, among the core characteristic parameters, the peak width at the 75% peak of the Barkhausen noise only responds to changes in the aging level when the stress is lower than the yield strength, and is not affected by stress; while the maximum value of Barkhausen noise, the average value of Barkhausen noise, and the equivalent remanence of Barkhausen noise are all positively linearly correlated with stress and aging level, and the influence of on-site pipeline stress on the micro-magnetic signal is eliminated by combining the parameters.

[0025] The peak width at 75% of the Barkhausen noise peak specifically refers to the signal width corresponding to 75% of the height of the Barkhausen noise signal peak. This parameter was selected based on the observation in the examples that it exhibits extremely high sensitivity to microstructural evolution caused by aging (such as decreased martensite dislocation density and carbide coarsening) within the elastic range where stress is below the yield strength, but its response to magnetoelastic effects caused by elastic stress is extremely weak. This achieves preliminary isolation of stress interference at the physical level, effectively improving the accuracy of aging level prediction.

[0026] Specifically, a multivariate nonlinear regression algorithm can be used to achieve nonlinear correlation modeling. Multivariate nonlinear regression is a statistical method that predicts the value of a dependent variable by establishing a nonlinear functional relationship between the dependent variable and multiple independent variables. Taking this invention as an example, the peak width at the 75% peak of Barkhausen noise, the maximum value of Barkhausen noise, the mean value of Barkhausen noise, and the equivalent remanence of Barkhausen noise are used as independent variables, and the aging level is used as the dependent variable to construct a multivariate nonlinear regression model. By fitting a large amount of experimental data, the parameters in the model are determined, thereby achieving decoupling between stress effects and aging effects. In addition, support vector machine regression is also a feasible algorithm. Support vector machine regression is a regression method based on support vector machines, which finds an optimal hyperplane that minimizes the distance from all sample points to the hyperplane. In this invention, the core feature parameters are used as input, and the aging level is used as output to train the support vector machine regression model, which is then used to achieve decoupling between stress and aging effects.

[0027] Preferably, in step S5, a multi-layer backpropagation neural network structure is constructed, with the number of neurons in the input layer being consistent with the number of core feature parameters selected, 10 neurons in the hidden layer, and 1 neuron in the output layer corresponding to the aging level prediction value.

[0028] Preferably, the backpropagation neural network is trained iteratively using a training set, with mean squared error as the loss function, and the network weights and biases are optimized through the backpropagation algorithm.

[0029] Preferably, in step S6, the area to be tested is surface-treated before on-site testing to remove surface oxides and impurities, ensuring that the surface roughness meets the testing requirements.

[0030] Preferably, in step S6, while outputting the predicted aging level value, the mechanical property degradation assessment status of the material in the region is also output simultaneously based on the pre-established correlation between the aging level and mechanical properties.

[0031] Preferably, the formation of the standardized sample library in step S2 includes verifying the prepared samples of each level through the calibration system of step S1 to ensure that the microstructure and mechanical properties of the samples correspond to the target aging level.

[0032] Preferably, in step S3, multiple sets of valid data are repeatedly collected for each load point, and outliers are removed to ensure the reliability of the full-dimensional dataset.

[0033] Preferably, in step S5, the input core micromagnetic feature parameters are normalized to eliminate the influence of differences in the dimensions of different parameters, and the normalization interval is set to the range of [0,1].

[0034] Preferably, in step S6, when the micromagnetic detection probe is attached to the area to be measured, the force is kept constant to reduce the interference of the probe lift-off effect on the micromagnetic characteristic signal.

[0035] Preferably, in step S4, the number of iterations of the ReliefF feature selection algorithm is set to 100, the number of nearest neighbors is set to 10, and the weight value is positive, with the larger the value, the stronger the ability of the feature parameter to distinguish the aging level.

[0036] Preferably, the learning rate of the backpropagation neural network is set to 0.01, the maximum number of iterations is set to 1000, and the activation function is a non-linear activation function.

[0037] Preferably, in the mechanical performance range and reliability determination rules, the average high-temperature tensile strength of the first aging level is not less than 420 MPa, and the average high-temperature tensile strength of the fifth aging level is not greater than 330 MPa.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. This invention establishes a detection model based on micromagnetic characteristic parameters, achieving completely non-destructive testing of pipeline aging levels without requiring cutting and sampling of pressure-bearing pipelines, thus fundamentally protecting the structural integrity of the pipeline. This method eliminates the construction safety risks associated with on-site sampling, making it possible to conduct large-scale, multi-point rapid inspections during pipeline downtime. The detection time for a single point is reduced to less than 10 minutes, significantly improving detection efficiency compared to traditional metallographic inspection methods and significantly reducing operation and maintenance costs and time costs.

[0040] 2. This invention innovatively establishes a two-dimensional standardized calibration system encompassing microstructural characteristics and macroscopic mechanical performance parameters, clarifying the microscopic evolution indicators and mechanical performance ranges corresponding to multiple aging levels. This system solves the problem of insufficient calibration accuracy of micromagnetic model samples in existing technologies, providing a standardized and highly reliable data foundation for model training, and ensuring the objectivity and accuracy of aging level prediction from the source.

[0041] 3. Addressing the core weakness of existing micromagnetic detection technologies in eliminating stress interference, this invention utilizes the ReliefF algorithm to identify key micromagnetic feature parameters with differentiated responses, achieving effective decoupling of stress and aging effects. By employing the ReliefF feature selection algorithm, this invention identifies key micromagnetic feature parameters that are highly sensitive to aging and exhibit differentiated responses to stress. Leveraging the physical characteristic that the peak width at a specific proportion of Barkhausen noise remains unaffected by stress within the elastic range, combined with other feature parameters, this invention successfully achieves effective decoupling of stress and aging effects. This breakthrough solves the core problem of large prediction deviations in existing micromagnetic detection technologies under complex stress environments, resulting in extremely high accuracy in aging level identification and representing a leap from qualitative analysis to precise quantitative prediction. By identifying key feature parameters with differentiated stress responses, this invention achieves effective decoupling of stress and aging effects. In field environments with both working and residual stress, the aging level prediction error is less than 1.5%, far superior to the over 20% prediction error caused by stress interference in existing technologies.

[0042] 4. This invention employs a multi-parameter fusion backpropagation neural network model, which can not only accurately output the aging level of the pipeline but also simultaneously assess the degradation status of key mechanical properties such as the tensile strength of the material. This multi-dimensional output mode provides comprehensive data support for pipeline remaining life assessment and safety risk early warning, greatly improving the intelligence level of equipment operation and maintenance.

[0043] 5. The method and process of this invention are standardized and can be directly integrated into portable testing equipment, requiring minimal professional background from on-site operators. Test results are output in real time, eliminating the need for complex laboratory analysis and making it perfectly suited to harsh on-site operating environments. This technical solution is not only applicable to specific types of chromium-molybdenum heat-resistant steel but can also be extended to other similar martensitic heat-resistant steel materials, possessing broad industrial application prospects and significant economic and social benefits. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the overall technical solution architecture according to an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the core feature selection and stress decoupling principle framework based on the ReliefF algorithm in the prediction method according to an embodiment of the present invention;

[0046] Figure 3 This is a logical framework diagram of a standardized calibration system covering both microscopic organization and macroscopic mechanics in the prediction method according to an embodiment of the present invention;

[0047] Figure 4 This is a schematic diagram of the data flow of the aging level prediction model based on the backpropagation neural network in the prediction method according to an embodiment of the present invention;

[0048] Figure 5 This is a flowchart illustrating the logical process of implementing on-site pipeline inspection and quantitative output of aging levels in the prediction method according to an embodiment of the present invention. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this is not intended to limit the scope of the invention; it is merely illustrative. It should be noted that, unless otherwise specified, the embodiments and technical features described in this application can be combined with each other. Unless otherwise indicated, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. Unless otherwise specified, the experimental methods used in the following embodiments are conventional methods. Unless otherwise specified, the materials, reagents, etc., used in the following embodiments are commercially available.

[0050] Example 1

[0051] The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters specifically performs the following steps, as follows: Figure 1 As shown:

[0052] In the method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters, step S1 involves constructing a standardized calibration system for the aging level of P91 pipelines. Specifically, the construction logic of the calibration system is based on the coupled mapping relationship between the microstructural degradation and macroscopic mechanical property decay caused by long-term service of P91 pipelines under high temperature and high pressure conditions. The calibration system covers both the microstructural characteristic dimension and the macroscopic mechanical property parameter dimension, such as... Figure 3 As shown.

[0053] In the microstructure characteristics dimension, the calibration process was completed through a combination of metallographic microscopy (OM), scanning electron microscopy (SEM), electron backscatter diffraction (EBSD), and energy dispersive spectroscopy (EDS). Specific calibration indicators were set as the morphology and orientation of martensite laths, the proportion of ferrite phase, the equivalent circle diameter of grains, the carbide precipitation and coarsening state, and the aggregation and distribution characteristics of specific alloying elements (such as chromium (Cr) and molybdenum (Mo)).

[0054] The calibration system classifies aging levels into 5 levels:

[0055] The microstructure characteristics corresponding to the first aging level are as follows: the matrix presents as complete martensitic laths with high dislocation density inside the laths, clear lath orientation and strong preferred orientation, and no ferrite phase is formed in the microstructure; fine and diffusely distributed M23C6 type carbides are observed inside the laths and at the lath boundaries, no obvious carbide chain aggregation is observed at the grain boundaries, the area-weighted average of the equivalent circle diameter of the grains is not greater than 6 μm, and chromium and molybdenum elements are uniformly distributed in the matrix without segregation characteristics.

[0056] The microstructure characteristics corresponding to the second aging level are defined as follows: the martensite still maintains the lath morphology, but the lath orientation shows slight dispersion, and a small amount of independently distributed blocky ferrite begins to be mixed in the matrix; at the lath boundaries and the original austenite grain boundaries, granular carbides begin to precipitate and show a preliminary tendency to grow, the area-weighted average of the equivalent circle diameter of the grains is within the first preset size range of 6μm to 10μm, and the distribution of alloying elements remains basically uniform.

[0057] The microstructure characteristics corresponding to the third aging level are as follows: the martensite laths are significantly fragmented, and the original preferred orientation is significantly dispersed; large blocky ferrite regions appear in the matrix; due to the Oswald ripening effect, the number of carbides decreases and their size is significantly coarsened, and obvious spherical carbides appear; the area-weighted average of the equivalent circle diameter of the grains is located in the second preset size range of 10 μm to 15 μm; and molybdenum shows preliminary segregation and aggregation in the carbide enrichment area.

[0058] The microstructure characteristics corresponding to the fourth aging level are defined as follows: only a small amount of martensite traces remain in the matrix, a large amount of ferrite phase is generated and dominates, and the structure is transformed into a ferrite-dominated matrix; spherical carbides further grow and increase in number, and obvious chain-like distribution characteristics appear at the grain boundaries. The area-weighted average of the equivalent circle diameter of the grains is located in the third preset size range of 15μm to 25μm. Chromium and molybdenum elements accumulate in large quantities in the precipitated phase, resulting in a decrease in the abundance of iron elements at the corresponding positions in the matrix.

[0059] The microstructure characteristics corresponding to the fifth aging level are defined as follows: the martensite features completely disappear, and the structure evolves into a mixture of coarse ferrite grains and large-sized carbides; the carbides undergo severe coarsening and spheroidization, exhibiting a long strip or continuous chain-like distribution, and the area-weighted average of the equivalent circle diameter of the grains is greater than the second preset size threshold of 25 μm; chromium and molybdenum elements exhibit high-intensity aggregation at the same coordinate position.

[0060] Within the macroscopic mechanical performance parameters, the calibration process is completed through room temperature tensile tests, a preset high-temperature tensile test at 550°C, and a room temperature Charpy impact test. Calibration indicators include room temperature tensile strength, preset high-temperature tensile strength, yield strength, and impact energy.

[0061] The first aging level corresponds to a first preset strength threshold of room temperature tensile strength of not less than 650 MPa, and a third preset strength threshold of high temperature tensile strength of not less than 420 MPa.

[0062] The probability that the room temperature tensile strength corresponding to the second aging level is within the first preset strength range of 610MPa to 650MPa is not less than the first preset probability threshold of 80%.

[0063] The probability that the room temperature tensile strength corresponding to the third aging level is within the second preset strength range of 585MPa to 610MPa is not less than the second preset probability threshold of 80%.

[0064] The probability that the room temperature tensile strength corresponding to the fourth aging level is within the third preset strength range of 550MPa to 585MPa is not less than a third preset probability threshold of 95%.

[0065] The probability that the room temperature tensile strength corresponding to the fifth aging level is not greater than 550 MPa is not less than 20% of the second preset strength threshold and the average high temperature tensile strength is not greater than 330 MPa of the fourth preset strength threshold.

[0066] In the method for predicting the aging level of P91 pipes based on micromagnetic characteristic parameters, step S2 involves preparing P91 standard samples of different aging levels. Specifically, new P91 pipes from the same batch as those at the test site and old pipes with a service life exceeding 100,000 hours are selected as the experimental base materials.

[0067] Step S2 simulates a long-term service process through an accelerated aging heat treatment process:

[0068] The first aging test sample was made directly from new tube material that had not undergone heat treatment.

[0069] The second aging sample was subjected to a first preset heat treatment temperature of 810℃, and was held at that temperature for a first preset duration of 180 minutes before being cooled in the furnace.

[0070] The third aging sample was subjected to a second preset heat treatment temperature of 820℃, and after holding at that temperature for 90 minutes for a second preset duration, it was cooled in the furnace.

[0071] The fourth aging sample was subjected to a third preset heat treatment temperature of 820℃, and was held at that temperature for 160 minutes before being cooled in the furnace.

[0072] The fifth aging sample adopted a two-stage annealing process: first, it was held at the fourth preset heat treatment temperature of 1055℃ for 130 minutes for the fourth preset time, then cooled in the furnace to the fifth preset temperature of 740℃, and held again for 480 minutes for the fifth preset time before cooling to room temperature.

[0073] After preparation, step S2 verifies the standardized calibration system described in step S1 on samples of each grade. Through metallographic observation and mechanical analysis of 25 parallel samples for each grade, it ensures that the martensite integrity, carbide size and distribution, tensile strength, and other indicators of the samples accurately correspond to the target aging grade, thus forming a robust standardized sample library. At least 25 parallel samples are prepared for each aging grade to ensure the statistical reliability of the sample data. This invention employs furnace cooling to ensure the uniformity of the sample microstructure and avoid residual stress interference caused by rapid cooling.

[0074] In the P91 pipe aging level prediction method based on micromagnetic characteristic parameters, step S3 involves collecting multi-dimensional micromagnetic data considering stress interference. Specifically, the standardized specimen prepared in step S2 is processed into a plate tensile specimen with a gauge length width of 10 mm and a thickness of 2 mm. A step-by-step tensile test is performed using a WDW-100 microcomputer-controlled universal testing machine.

[0075] The stepped tensile process uses the measured yield strength of the specimen as the logical dividing point, setting five sets of equally spaced gradient loads within the elastic stress range from 0 MPa to the yield strength; and five sets of gradient loads within the plastic stress range from the yield strength to the tensile strength. At each load point, the testing machine maintains a preset constant load duration of 60 seconds to eliminate the disturbance of instantaneous plastic deformation to the magnetic domain structure.

[0076] During the loading period, a multifunctional micromagnetic nondestructive testing system was used to collect micromagnetic characteristic signals on a pre-marked area on the sample surface. The collected signal categories included:

[0077] A. Tangential magnetic field strength signal: used to characterize the magnetic flux distribution on the surface of a material and its corresponding force response.

[0078] B. Barkhausen noise signal: The induced electromotive force signal generated by the magnetic domain wall during discontinuous jumping process is captured by the induction coil, reflecting the pinning effect of dislocations, grain boundaries and precipitation on the magnetic domain wall.

[0079] C. Incremental permeability (MICP) signal: The variation law of microscopic permeability of materials is measured under alternating magnetic field modulation.

[0080] D. Multi-frequency eddy current signal: The combined response of electrical conductivity and magnetic permeability of deep tissues of materials is detected by excitation magnetic fields of multiple frequencies.

[0081] The full-dimensional micromagnetic characteristic parameters collected in step S3 total 44 indicators. To ensure the reliability of the full-dimensional dataset, five sets of valid data were collected repeatedly at each load point. Step S3 also includes a data preprocessing sub-step, which uses the Laida criterion (3σ criterion) to remove abnormal data values ​​caused by environmental electromagnetic interference. The final dataset is stored in matrix form, with each row containing: aging level label (1-5), real-time stress value (MPa), and 44 micromagnetic characteristic parameter values.

[0082] In the P91 pipeline aging level prediction method based on micromagnetic feature parameters, step S4 involves screening highly sensitive micromagnetic feature parameters for aging. Specifically, step S4 performs dimensionality reduction and correlation analysis on the 44-dimensional original feature space by executing the ReliefF feature screening algorithm.

[0083] The ReliefF feature selection algorithm is executed as follows: the preset number of iterations m is set to 100, and the preset number of nearest neighbors k is set to 10. In each iteration, a sample R is randomly selected from the training set, and then k nearest neighbor samples (Near Hits) are searched in the sample set of the same category as R, and k nearest neighbor samples (Near Misses) are searched in the sample set of each different category.

[0084] The weight value of each feature parameter is updated by calculating its contribution to the distance between R and its nearest neighbors. The formula for calculating the weight value is defined as follows:

[0085]

[0086] in, Indicates the first The weight values ​​of each feature parameter, Indicates the feature offset distance. Indicates similar or close neighbors. express Dissimilar neighbors express Prior probability of a class of samples.

[0087] Based on the calculation results of the weight values, the top four core feature parameters are selected: maximum Barkhausen noise (BN_Max), mean Barkhausen noise (BN_Mean), equivalent remanence of Barkhausen noise (BN_Mr), and peak width at the 75% peak of Barkhausen noise (BN75). Figure 2 As shown.

[0088] In terms of physical mechanism, the core characteristic parameters exhibit differentiated response characteristics: the peak width at the 75% peak of the Barkhausen noise (BN75), within the elastic range where the stress is below the yield strength, is highly sensitive to the microstructural evolution caused by aging (such as a decrease in martensite dislocation density and carbide coarsening), but its response to the magnetoelastic effect caused by elastic stress is extremely weak, thus achieving preliminary isolation of stress interference at the physical level. BN_Max, BN_Mean, and BN_Mr show a linear increasing trend with increasing stress, and also increase with increasing aging level. By modeling the nonlinear correlation between BN75 and the other three parameters, the method can effectively decouple the interference of residual stress and working stress in the pipeline on aging prediction.

[0089] In the method for predicting the aging level of P91 pipes based on micromagnetic feature parameters, step S5 involves constructing a prediction model for the aging level of P91 pipes based on a backpropagation (BP) neural network. Specifically, the BP neural network employs a three-layer topology.

[0090] The input layer contains four neurons, corresponding to BN_Max, BN_Mean, BN_Mr, and BN75 selected in step S4, respectively. The output layer has one neuron, whose output value is a floating-point number between 1.0 and 5.0, representing the quantitatively predicted aging level. The hidden layer has a preset number of 10 neurons.

[0091] During the model building process, the following sub-operations are performed:

[0092] a. Data normalization processing: In order to eliminate the influence of different micromagnetic parameter dimensions and numerical magnitudes on the gradient descent convergence speed, a linear mapping function is used to normalize the input BN feature parameters to the preset value range of [0,1].

[0093] b. Parameter initialization: Network weights and biases are initialized using a random initialization strategy.

[0094] c. Iterative Training: The backpropagation neural network is trained iteratively using the training set. The full-dimensional dataset is divided into a training set and a validation set in a 7:3 ratio. The preset learning rate is 0.01, and the preset maximum number of iterations is 1000. The activation function used is the Sigmoid non-linear activation function.

[0095] d. Loss Function Definition: The mean squared error (MSE) between the network output predicted value and the calibration level label in step S1 is used as the loss function, and its formula is defined as follows:

[0096]

[0097] in, To label the level, This is the predicted output of the neural network. This represents the number of samples.

[0098] The error gradient is calculated using the backpropagation algorithm, and the connection weights and biases of neurons in each layer are optimized layer by layer using gradient descent. Training is stopped and the model parameters are fixed when the number of iterations reaches a preset upper limit or the MSE descent slope falls below a preset threshold. Figure 4 As shown.

[0099] In the P91 pipeline aging level prediction method based on micromagnetic characteristic parameters, step S6 involves performing on-site pipeline inspection and aging level prediction. Specifically, step S6 integrates the fixed BP neural network prediction model into the embedded computing module of the portable micromagnetic nondestructive testing instrument.

[0100] The on-site implementation process includes:

[0101] First, surface treatment is performed on the area to be tested of the P91 pipeline (such as the main steam pipeline of a thermal power unit). A handheld angle grinder is used with 120- to 400-grit sandpaper to gradually remove the oxide scale, anti-rust paint, and pitting on the surface to be tested until the metal substrate is exposed, ensuring that the surface roughness Ra meets the testing requirement of less than 6.3 μm.

[0102] Subsequently, the micromagnetic detection probe is vertically attached to the center of the area to be measured. During the data acquisition process, a constant force spring support is used to maintain the force between the probe and the pipe surface at a constant range of 10N to 15N, in order to minimize the interference of the probe lift-off effect on the micromagnetic induction signal.

[0103] The detector drives the excitation coil in real time to generate a variable frequency triangular wave magnetic field, and simultaneously acquires the BN_Max, BN_Mean, BN_Mr, and BN75 core feature parameters of the area under test. These parameters are then input into the built-in BP neural network prediction model.

[0104] The model directly calculates and outputs the predicted aging level of the area under test. Step S6, while outputting the predicted aging level, also simultaneously outputs the degradation assessment status of the material's mechanical properties, such as room temperature tensile strength, high temperature tensile strength, and residual impact energy, based on a pre-established empirical correlation curve between aging level and mechanical properties. This provides a multi-parameter evidence chain for pipeline safety assessment. Figure 5 As shown.

[0105] Application examples:

[0106] During the annual overhaul of a 600MW supercritical thermal power generating unit, the method described in Example 1 was used to inspect the elbows and heat-affected zones of the main steam P91 pipeline. This pipeline has accumulated 125,000 hours of operation.

[0107] Following the requirements of step S6, the pipe surface was ground. Testing was then conducted using a micro-magnetic detection device integrating the prediction model from step S5. The device measured the BN75 characteristic value of the outer back arc region of a certain elbow to be 124.52 μV and the BN_Max characteristic value to be 856.31 μV. After substituting the data into the model, the system output a predicted aging level of 3.2.

[0108] Based on the synchronously output mechanical property correlation results, the system indicated a predicted room-temperature tensile strength of 592 MPa at this location. Subsequent on-site micro-sampling and laboratory metallographic analysis revealed that the microstructure at this point exhibited obvious three-stage aging characteristics of martensitic lath fragmentation and carbide growth. The measured room-temperature tensile strength was 601 MPa, with a prediction error of only 1.5%, demonstrating the effectiveness of the method in eliminating stress interference and improving quantitative prediction accuracy.

[0109] Example 2

[0110] This embodiment is an extended application based on the core principles of the present invention, applicable to P92 steel pipes, and does not affect the protection scope of the present invention for P91 pipes. To make the technical solution of the present invention more widely applicable under different material systems, this embodiment 2, based on embodiment 1, optimizes and adjusts the aging level prediction method for P92 pipes, whose composition differs slightly from that of P91 pipes.

[0111] In the aforementioned method for predicting the aging level of P91 pipes based on micromagnetic characteristic parameters, step S1 adds a calibration index for the Laves phase (Fe2W type intermetallic compound) for P92 steel. Since P92 steel contains approximately 1.5% to 2.0% tungsten (W), the precipitation and coarsening of the Laves phase during aging has a more significant pinning effect on magnetic domain walls than that of P91 steel.

[0112] In the first aging level, the Laves phase has not yet precipitated, and the tungsten element is completely dissolved in the matrix.

[0113] In the third to fifth aging levels, the equivalent diameter of the Laves phase was added to the calibration index. The Laves phase size corresponding to the fifth aging level is usually greater than 0.5 μm and is distributed in a continuous network at the grain boundaries.

[0114] In step S2, the temperature parameters of the accelerated aging heat treatment were fine-tuned to accommodate the higher thermal stability of P92 steel. The preset heat treatment temperature of the second aging sample was increased to 820℃, and the holding time of the fifth aging sample was increased to 600 min to ensure that the degree of microstructure evolution accurately covers the standardized calibration system.

[0115] In step S4, for the multi-dimensional dataset of P92 steel, in addition to BN_Max and BN75, the ReliefF algorithm introduced the incremental permeability peak position as a supplementary core feature. Experiments showed that as the precipitation of the Laves phase increases the matrix coercivity, the peak position of the incremental permeability waveform shifts towards higher field strengths. By introducing the Peak_Position parameter, the model can more sensitively capture the signal fluctuations in the early aging stage caused by Laves phase precipitation.

[0116] In step S5, the neural network structure of the prediction model is expanded to a 4-12-1 structure, that is, the number of hidden layer neurons is increased to 12, in order to cope with the complex tissue features of P92 steel.

[0117] In the data processing sub-step, the normalization algorithm adds specific weight assignments to the Peak_Position parameter. The learning rate of the BP neural network is reduced to 0.005 to obtain a smoother loss function convergence curve and improve the model's ability to identify subtle aging differences.

[0118] When conducting on-site testing in step S6, considering that P92 pipes typically have thicker walls and higher surface oxidation hardness, a plasma cleaning or more powerful mechanical polishing step is added to the surface treatment sub-step to ensure the electromagnetic coupling efficiency between the probe and the substrate.

[0119] This embodiment 2, through targeted adjustments, achieves accurate prediction of the aging level of P92 steel pipelines. In the actual test verification, multi-point inspections were carried out on a P92 main steam pipeline. The aging alarm points output by the model at level 4.5 were verified by subsequent destructive tests, and their impact energy had decreased to below 20J, fully demonstrating the engineering practical value of the prediction model in safety early warning.

[0120] Example 3

[0121] In the P91 pipeline aging level prediction method based on micromagnetic feature parameters, in order to further improve the model stability under complex and variable environments, this embodiment 3 introduces a BP neural network structure optimized by genetic algorithm (GA) under the architecture of embodiment 1, and makes in-depth improvements to step S5.

[0122] Specifically, in step S5, the model building process adds a weight and bias pre-optimization step based on a genetic algorithm:

[0123] 1. Chromosome encoding: Concatenate all connection weights and neuron biases of the BP neural network into an individual vector, which serves as the chromosome for the genetic algorithm.

[0124] 2. Fitness Function Design: The sum of prediction errors of the BP neural network on the training set is used as the reciprocal of the fitness function, i.e.:

[0125] The higher the fitness, the closer the weights and bias parameters of the set are to the global optimum.

[0126] 3. Genetic Operations: Perform selection, crossover, and mutation operations. Tournament selection is used to retain superior individuals, and real crossover is used to generate new individuals, with the mutation probability set to 0.05.

[0127] 4. Pre-optimized output: The optimal initial parameter set of the BP neural network is found by iterating 50 times through the GA algorithm, which avoids the problem that the traditional BP algorithm is prone to getting trapped in local minima due to random initialization.

[0128] In step S3, an environmental temperature compensation sub-step is added to the micromagnetic data acquisition. The on-site testing environment may experience high-temperature fluctuations ranging from 40℃ to 80℃, which can affect the Curie point and cause the magnetic saturation intensity to drift. The method integrates a high-precision thermistor within the probe, simultaneously recording the real-time temperature while acquiring the micromagnetic signal.

[0129] In step S5, the number of input layer neurons is expanded from 4 to 5, and "detecting ambient temperature" is added as an input feature. The BP neural network learns the thermodynamic relationship between temperature and micromagnetic parameters and automatically performs temperature compensation in its internal logic, ensuring that the prediction model has extremely high reproducibility in detection results under different seasons and different downtime stages.

[0130] In step S4, when screening core characteristic parameters, the redundancy between parameters is evaluated by calculating mutual information. In addition to the BN series parameters, the phase shift of multi-frequency eddy currents is selected as a characteristic auxiliary identification factor for the fifth aging stage. When the pipeline enters the fifth aging stage, due to extreme grain coarsening and the formation of a large amount of ferrite, the electrical conductivity of the material undergoes a significant abrupt change; Phase Shift can provide electrical auxiliary evidence in addition to magnetic characteristics.

[0131] In step S6, the on-site detection system adds a module for wireless data upload and cloud comparison. The core micromagnetic characteristic parameters of the detection points are transmitted to the server via 4G / 5G network and dynamically matched with a standardized sample library.

[0132] This embodiment 3 significantly improves the anti-interference capability of the prediction model under harsh field conditions through GA-BP algorithm optimization and environmental factor compensation. Practical application data shows that, using the method of this embodiment, even with slight differences in surface grinding quality across different shifts and batches, the consistency deviation of the aging level prediction results remains within 0.1 levels.

[0133] Example 4

[0134] Based on the above embodiments, this embodiment 4 focuses on the lift-off effect interference in the field inspection of P91 pipeline, and specifically strengthens steps S4 and S6.

[0135] In step S4, to eliminate signal attenuation caused by minute gap variations between the detection probe and the pipe surface, a ratio-type core characteristic parameter is introduced. Specifically, in addition to BN75, the selected characteristic parameter includes a ratio term between BN_Max and the Barkhausen noise RMS value (BN_RMS).

[0136] Physical studies have shown that when the probe undergoes a minute displacement (lift-off), BN_Max and BN_RMS typically decrease proportionally, while their ratio BN_Ratio is extremely insensitive to changes in the lift-off distance, but remains highly sensitive to noise pulse waveform distortion caused by aging. By introducing this ratioistic feature, the prediction model constructed in step S5 receives physical quantities with "self-collimation" characteristics at the input layer.

[0137] In step S6, the micromagnetic detection probe employs a special suspended array structure. The array contains two symmetrically distributed Hall elements.

[0138] In the acquisition sub-step, the system calculates the actual lift-off height of the probe in real time by using the difference in magnetic field strength output by the two Hall elements.

[0139] If the real-time lift-off height exceeds the preset threshold of 0.5mm, the device will automatically alarm and lock data acquisition; if the lift-off height is between 0.1mm and 0.5mm, the pre-stored lift-off compensation function will be called to correct the gain of the acquired core micro-magnetic characteristic parameters. The correction formula is as follows:

[0140]

[0141] in, These are the corrected parameter values. These are the original collected values. The compensation coefficient varies with the aging level. To provide real-time lifting height.

[0142] In the calibration system of step S1, the ultrasonic attenuation coefficient was added as a verification indicator for the fourth and fifth stages of aging. When preparing standard samples (step S2), the attenuation constant of each stage of the sample was determined by ultrasonic pulse reflection method and calibrated in conjunction with micromagnetic characteristics.

[0143] In the model training step S5, a regularization term was added to prevent overfitting. By adding an L2 norm penalty term to the loss function, the absolute values ​​of neuron weights are limited to prevent them from becoming too large, thus ensuring that the model can still provide reasonable predictions when faced with abnormally aged tissues in non-calibrated systems. Linear / nonlinear correlation curves between aging levels and room temperature tensile strength, 550℃ high-temperature tensile strength, and impact energy were pre-fitted using calibration data from standard samples at various levels.

[0144] This embodiment 4 solves the problem of unstable test data caused by manual operation on site by combining physical compensation and algorithm optimization. In an actual blind test experiment at a thermal power plant, for the same aging point, the test results under 10 different angles and clamping forces showed that the predicted aging level was always stable at 2.8, with a standard deviation of only 0.04, which greatly improved the standardization level of non-destructive testing process.

[0145] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters, characterized in that, Includes the following steps: S1. Construct a two-dimensional standardized calibration system for the aging levels of P91 pipelines: Establish a two-dimensional standardized calibration system covering microstructure characteristics and macroscopic mechanical performance parameters, and classify the aging degree of P91 pipelines into the first to fifth aging levels. S2, Prepare multi-level standard samples matching the calibration system: Prepare standard samples of multiple aging levels matching the calibration system in step S1 by accelerating the aging heat treatment of P91 base material, and prepare no less than 25 parallel samples for each aging level. S3, Construct a full-dimensional micromagnetic dataset containing stress variables: Micromagnetic data were collected from standard specimens at all levels under stepped tensile loads. Taking the measured yield strength of the specimen as the boundary, five sets of equally spaced gradient stress loads were set in the elastic range from 0 MPa to the yield strength, and five sets of gradient stress loads were set in the plastic range from the yield strength to the tensile strength. Each load point was held for 60 seconds to eliminate the disturbance of instantaneous plastic deformation to the magnetic domain structure. Micromagnetic characteristic signals under different aging levels and different stress levels were obtained, and a full-dimensional dataset containing aging level, stress level and 44 micromagnetic parameters was constructed. S4, Screening core feature parameters with differentiated stress response: The ReliefF feature screening algorithm with 100 iterations and 10 nearest neighbors is used to extract features from the full-dimensional dataset to identify core feature parameters that are highly sensitive to aging levels and can effectively distinguish stress interference. The core feature parameters include a first feature parameter that is not sensitive to stress changes in the elastic stress range below the yield strength and only responds to changes in aging levels, and a second feature parameter that is positively linearly correlated with both stress and aging levels. The first characteristic parameter is the peak width at the 75% peak of Barkhausen noise, whose signal change rate within the elastic stress range is less than 5%, while its signal change rate with respect to aging level changes is greater than 30%, which is used to achieve physical isolation of stress interference; the second characteristic parameter is the maximum value of Barkhausen noise, the mean value of Barkhausen noise, and the equivalent remanence of Barkhausen noise, which decouples the stress effect from the aging effect through the multivariate nonlinear fusion of the first characteristic parameter and the second characteristic parameter; S5, Construct an aging level prediction model based on a backpropagation neural network: Using the core feature parameters as input and the calibrated aging level as output, construct a three-layer backpropagation neural network prediction model with 4 neurons in the input layer, 10 neurons in the hidden layer, and 1 neuron in the output layer. S6, Implement on-site testing and prediction: Collect the core micro-magnetic characteristic parameters of the area to be tested on-site and input them into the prediction model, and output the aging level prediction value.

2. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, In step S1, the microstructure characteristic calibration indicators include martensite lath morphology and orientation, ferrite phase ratio, equivalent grain diameter, carbide precipitation and coarsening state, and specific alloying element aggregation and distribution characteristics; the macroscopic mechanical property parameter calibration indicators include room temperature tensile strength, 550℃ high temperature tensile strength, yield strength, and impact energy.

3. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 2, characterized in that, In step S1: the first aging level corresponds to martensite in a complete lath shape, with no ferrite formation, and the area-weighted average of the equivalent circle diameter of the grains is no greater than 6 μm; the fifth aging level corresponds to the complete disappearance of martensite, and the microstructure is a mixture of coarse ferrite and large-sized carbides, with the average equivalent circle diameter of the grains being greater than 25 μm.

4. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, Step S2 specifically includes: the first aging sample uses untreated base material; the second aging sample is heated at 810℃ for 180 min and then cooled; the third aging sample is heated at 820℃ for 90 min and then cooled; the fourth aging sample is heated at 820℃ for 160 min and then cooled; the fifth aging sample uses a two-stage annealing process, first heated at 1055℃ for 130 min and then cooled in the furnace to 740℃, then heated again for 480 min and then cooled.

5. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, The micro-magnetic characteristic signals acquired in step S3 include tangential magnetic field strength signals, Barkhausen noise signals, incremental permeability signals, and multi-frequency eddy current signals; step S3 also includes using the Laida criterion to remove abnormal data values.

6. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, In step S4, the peak width at the 75% peak of Barkhausen noise is used as the reference feature of the aging level to establish a reference mapping relationship between it and the aging level. Then, the maximum value of Barkhausen noise, the mean value of Barkhausen noise, and the equivalent remanence of Barkhausen noise are stress-corrected using the reference mapping relationship to achieve decoupling between stress influence and aging influence.

7. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, The peak width at 75% of the Barkhausen noise is highly sensitive to the microstructure evolution caused by the decrease in martensite dislocation density and carbide coarsening, but its response to the magnetoelastic effect caused by elastic stress is extremely weak.

8. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, In step S5, the connection weights and biases of the neural network are pre-optimized using a genetic algorithm before model training; during training, mean squared error is used as the loss function, and L2 regularization is used to prevent overfitting of the model.

9. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, In step S6, before on-site testing, the oxide scale of the area to be tested is removed so that the surface roughness Ra is below 6.3 μm; during the acquisition process, the probe force is kept constant between 10N and 15N by a constant force spring bracket to reduce lift-off interference.

10. The method for predicting the aging level of P91 pipelines based on micromagnetic characteristic parameters according to claim 1, characterized in that, In step S6, while outputting the predicted aging level, the degradation assessment status of mechanical properties such as room temperature tensile strength, 550℃ high temperature tensile strength, and residual impact energy is simultaneously output.

Images