Method and device for detecting fatigue damage of metal member based on positron annihilation

Abstract

This invention discloses a method and apparatus for fatigue damage detection of metal components based on positron annihilation. The method involves collecting positron annihilation lifetime spectrum data and Doppler broadening data of the detection area of ​​the metal component, performing defect capture rate correlation analysis to establish a defect evolution curve, locating the inflection point of damage evolution based on the defect evolution curve to determine the damage-sensitive time window, and generating an adaptive detection window by combining momentum ratio parameters to form an annihilation parameter correction factor. The adaptive detection window is then adapted to form a detection configuration table, identifying the defect capture path and calculating the annihilation lag time. Based on the annihilation lag time, damage levels are defined to generate a damage stratification configuration. Cross-state lag consistency evaluation is performed on the damage stratification configuration to determine the optimal excitation mode. A detection execution sequence is formed by prioritizing fluctuations, and the fatigue damage detection results are jointly analyzed and output, achieving quantitative characterization of microscopic defect types and damage spatial distribution in metal components.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and in particular to a method and apparatus for fatigue damage detection of metal components based on positron annihilation. Background Technology

[0002] Under long-term alternating loads, metallic components gradually accumulate lattice defects at the microscopic scale. Early damage exists in the form of point defects such as vacancies and dislocations, which gradually evolve into vacancy aggregation and micropores as service time increases, eventually leading to the initiation and propagation of fatigue cracks. This damage process is difficult to detect at the macroscopic level, but once it enters the rapid crack propagation stage, it can easily cause sudden fracture failure, posing a serious threat to the safe operation of engineering structures.

[0003] Existing fatigue testing methods for metal components largely rely on macroscopic physical field methods such as ultrasound, magnetic particle testing, and eddy current testing. While these methods are sensitive to established macroscopic defects, they have limited ability to distinguish early microscopic defects such as lattice vacancies and vacancy clusters, making it difficult to provide effective early warning at the damage initiation stage. Furthermore, when faced with complex damage scenarios involving multiple types of defects, existing testing methods lack the ability to quantitatively differentiate between different defect types and cannot accurately characterize the differences in damage evolution rates across different parts of the metal component. This results in insufficient precision in the testing conclusions, making it difficult to support differentiated maintenance decisions based on different damage states. Summary of the Invention

[0004] This invention discloses a method and apparatus for fatigue damage detection of metal components based on positron annihilation. Addressing the challenges of early detection of microscopic damage in metal components, insufficient quantitative differentiation of multiple defect types, and lack of refined characterization of damage spatial distribution, the invention establishes defect evolution curves by collecting positron annihilation lifetime spectrum data and Doppler broadening data. An adaptive detection window is generated by combining momentum ratio parameter correction, thereby identifying defect capture paths and quantifying annihilation lag time to complete damage stratification. Finally, fatigue damage detection results are output through joint analysis of optimal excitation mode matching and fluctuation priority ranking.

[0005] The first aspect of this invention proposes a fatigue damage detection method for metal components based on positron annihilation, comprising the following steps: Positron annihilation lifetime spectrum data and Doppler broadening data of the detection area of ​​the metal component are collected. Based on the correlation analysis of the defect capture rate of the positron annihilation lifetime spectrum data and the Doppler broadening data, a defect evolution curve is established. Based on the defect evolution curve, the damage evolution inflection point is located to determine the damage-sensitive time window. The defect identification momentum ratio parameter is obtained from the positron annihilation lifetime spectrum data and the Doppler broadening data to form an annihilation parameter correction factor. The annihilation parameter correction factor is used to perform window narrowing correction for the defect type transformation stage of the damage-sensitive time window to generate an adaptive detection window. The adaptive detection window is adapted to form a detection configuration table. Based on the detection configuration table, the defect capture path dominated by low-intensity long-life components is identified. The annihilation rate gradient of the defect capture path is calculated to obtain the annihilation lag time. Based on the annihilation lag time, the damage level is defined to generate a damage layer configuration. The optimal excitation mode is determined by cross-state hysteresis consistency evaluation of the damage stratification configuration. The optimal component decomposition window data of the optimal excitation mode is detected. The fluctuation priority sorting is performed based on the optimal component decomposition window data to form a detection execution sequence. The fatigue damage detection result is output by joint analysis of the damage degree based on the detection configuration table and the detection execution sequence.

[0006] A second aspect of this invention provides a fatigue damage detection device for metal components based on positron annihilation, comprising: The signal acquisition module is used to acquire positron annihilation lifetime spectrum data and Doppler broadening data of the detection area of ​​the metal component, and to establish a defect evolution curve based on the correlation analysis of the defect capture rate of the positron annihilation lifetime spectrum data and the Doppler broadening data. The window correction module is used to locate the inflection point of damage evolution based on the defect evolution curve to determine the damage-sensitive time window, obtain the defect identification momentum ratio parameter from the positron annihilation lifetime spectrum data and the Doppler broadening data to form an annihilation parameter correction factor, and use the annihilation parameter correction factor to perform window narrowing correction for the defect type transformation stage of the damage-sensitive time window to generate an adaptive detection window. The path analysis module is used to adapt the detection scheme to the adaptive detection window to form a detection configuration table, identify the defect capture path dominated by low-intensity long-life components according to the detection configuration table, calculate the annihilation rate gradient of the defect capture path to obtain the annihilation lag time, and define the damage level and generate a damage layer configuration based on the annihilation lag time. The result output module is used to perform cross-state hysteresis consistency evaluation on the damage stratification configuration to determine the optimal excitation mode, detect the optimal component decomposition window data of the optimal excitation mode, perform fluctuation priority sorting based on the optimal component decomposition window data to form a detection execution sequence, and perform joint analysis of damage degree based on the detection configuration table and the detection execution sequence to output fatigue damage detection results.

[0007] The beneficial effects of this invention are reflected in the following points: 1. Positron annihilation lifetime spectrum data and Doppler broadening data are collected in the detection area of ​​metal components according to their classification. A defect evolution curve is established using the correlation coefficient matrix of the co-evolution of the two types of parameters with the detection cycle. Based on this, the inflection point of damage evolution is located to determine the damage-sensitive time window. Transition node markers are extracted from the momentum spectrum decomposition results to construct annihilation parameter correction factors. Window narrowing correction is performed on the defect type transition stage in the damage-sensitive time window, realizing refined adaptation of the detection timing for the coexistence stage of different types of defects. 2. A detection configuration table is formed based on the adaptive detection window. The defect capture path dominated by low-intensity long-life components is located through trap competition effect analysis. Non-uniform spacing weighted difference calculation is performed on the path to generate an annihilation rate gradient sequence. The damage front section that maintains a high level after a sudden gradient change is identified and the annihilation lag time is quantified. The detection area of ​​metal components is divided into three types of damage layers based on the lag time, establishing a quantitative correlation between defect spatial distribution and damage evolution rate. 3. Perform cross-state hysteresis consistency assessment on the damage stratification configuration to screen out the optimal excitation mode where the deviation monotonically converges as the damage deepens. Based on the optimal component decomposition window data, identify abnormal silent sections and perform saturation capture stability scoring to form a detection execution sequence. Combine the acquisition condition constraints of the detection configuration table with the priority weights of the detection execution sequence to drive the weighted fusion of lifetime spectrum component parameters and broadening feature parameters, and output the spatial distribution of damage level and defect capture rate covering the entire detection area of ​​the metal component. Attached Figure Description

[0008] Figure 1 This is a schematic flowchart of the fatigue damage detection method for metal components based on positron annihilation according to the present invention.

[0009] Figure 2 This is a schematic diagram showing the division of the detection area between the positron annihilation detection device and the metal component in this invention.

[0010] Figure 3 This is a schematic diagram of the evolution curve of the defect capture rate κ as a function of the detection cycle in this invention.

[0011] Figure 4 This is a structural block diagram of the metal component fatigue damage detection device based on positron annihilation according to the present invention.

[0012] Wherein: 1-γ detector A; 2-γ coincidence ray path; 3-metal component; 4-detection and acquisition point; 5-high stress concentration zone; 6-welding heat-affected zone; 7-fatigue crack initiation prediction zone; 8- 22Na point source; 9-γ detector B; 10-κ axis; 11-detection cycle number axis; 12-Stage I curve segment (point defect dominant); 13-rate inversion inflection point T1; 14-Stage II curve segment (vacancy cluster dominant); 15-rate inversion inflection point T2; 16-Stage III curve segment (micropore dominant); 17-data marker points for each detection cycle. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0015] The technical solutions of the embodiments of this application will be described below.

[0016] like Figure 1 As shown, this embodiment of the invention provides a fatigue damage detection method for metal components based on positron annihilation, including the following steps S110-S140: Step S110: Collect positron annihilation lifetime spectrum data and Doppler broadening data of the detection area of ​​the metal component, and establish a defect evolution curve based on the correlation analysis of defect capture rate using positron annihilation lifetime spectrum data and Doppler broadening data.

[0017] Specifically, positron annihilation lifetime spectrum data and Doppler broadening data are collected from the detection area of ​​the metal component. For example... Figure 2 As shown, the detection area of ​​metal component 3 is pre-divided into three categories based on the component's service conditions: a high stress concentration zone 5, a welding heat-affected zone 6, and a fatigue crack initiation prediction zone 7. Detection and acquisition points 4 are arranged in each category according to the stress gradient difference. The spacing between acquisition points and the cumulative detection time in the high stress concentration zone 5 are superior to those in the other two categories. 22Na point source 8 (simultaneously configured on both sides of the detection surface of metal component 3 in a "source-sample-source sandwich" manner) Figure 2 (Only one side of the source body is shown) The metal component 3 is attached to the detection surface. The dual-probe γ coincidence detection system composed of γ detector A (1) and γ detector B (9) collects data from various regions 4 through the γ coincidence ray path 2, and outputs positron annihilation lifetime spectrum data and Doppler broadening data. In the positron annihilation lifetime spectrum data of the high stress concentration region 5, the intensity ratio of the long lifetime component τ3 is usually 12% to 20% lower than that of the fatigue crack initiation prediction region 7. The absolute difference of the S parameter of the Doppler broadening data between the two regions can reach 0.003 to 0.008. The spatial distribution difference between the two types of data outputs obvious layering characteristics, which is the signal basis for distinguishing the defect types of different damage levels. The cumulative number of events of the positron annihilation lifetime spectrum data at each detection point is not less than 2×10 6 To meet the statistical accuracy requirements of multi-component fitting, the energy resolution of the 511 keV peak in the Doppler broadening data is no less than 1.2 keV (FWHM). The positron annihilation lifetime spectrum data and the Doppler broadening data are strictly synchronized and recorded using the same time reference at all acquisition points in the metal component detection area, with a time reference error not exceeding 50 ps; the cumulative number of events is less than 2 × 10⁻⁶. 6 Or the peak count rate at 511 keV is less than 2 × 10 4 The detection point outputs a low statistical value marker, which triggers the point to participate in the weight reduction process during the defect capture rate correlation analysis stage.

[0018] Defect evolution curves were established by correlation analysis of defect capture rate based on positron annihilation lifetime spectrum data and Doppler broadening data. Correlation analysis was performed sequentially in four steps: parameter extraction, correlation coefficient matrix construction, defect capture rate conversion, and curve slope segmentation and weight mapping. From the lifetime spectrum data, the lifetime values ​​τi (ns) and intensity ratios Ii of each component were extracted. The intensity I1 of the τ1 component, corresponding to the matrix free annihilation contribution, served as the benchmark reference for defect capture rate conversion. The intensities I2 and I3 of the τ2 and τ3 components exhibited phased changes with damage evolution. From the Doppler broadening data, S-parameters (integral ratio ±1.5 keV at the peak center) and W-parameters (integral ratio ±2.5–5.0 keV) were extracted. A Pearson correlation coefficient matrix was established periodically using these two parameters, outputting the co-evolution intensity values ​​of each element. In the early stages of fatigue, the correlation coefficient between I2 and S-parameters was higher than 0.92. When the τ3 component appeared and I3 exceeded 8%, the W-parameter showed non-monotonic fluctuations, and the correlation coefficient dropped to 0.75–0.85. This phased decrease calibrated the defect type transition period. The defect capture rate κ (dimensions ns) was calculated. -1 The three-state capture model is based on the matrix reference annihilation rate λ_bulk (dimension ns). -1 ) and the annihilation rate of each component λi=1 / τi (dimension ns) -1The difference between the two values ​​is weighted and output according to Ii. The change of the S-parameter along the detection period direction, δS(t), is used as the input for cycle-by-cycle cross-validation. When the direction deviates, the confidence level of the κ value is downgraded. Figure 3 As shown, the defect evolution curve uses the κ axis 10 and the detection cycle number axis 11 as the coordinate reference. The κ value of each cycle is connected by the data marker point 17 of each detection cycle to output the complete evolution trajectory. It shows a three-stage evolution pattern with the slope rising first and then slowing down: Stage I curve segment 12 (point defect dominated) has the slowest slope, and the rate inversion inflection point T1 (13) marks the transition boundary of the directional vacancy cluster; Stage II curve segment 14 (vacancy cluster dominated) has a moderate slope, and the rate inversion inflection point T2 (15) marks the transition boundary of the directional micropore; Stage III curve segment 16 (micropore dominated) has the defect capture rate approaching saturation and the curve slope gradually slows down. The segmented structure of the curve slope corresponds to the staged transition of the defect from point defect to vacancy cluster and then to micropore.

[0019] Step S120: Based on the defect evolution curve, locate the inflection point of damage evolution to determine the damage-sensitive time window. Obtain the defect identification momentum ratio parameter from the positron annihilation lifetime spectrum data and Doppler broadening data to form the annihilation parameter correction factor. Use the annihilation parameter correction factor to perform window narrowing correction for the defect type transformation stage of the damage-sensitive time window to generate an adaptive detection window.

[0020] In some embodiments, determining the damage-sensitive time window by locating the inflection point of damage evolution based on the defect evolution curve includes: obtaining the annihilation rate time series of the defect evolution curve according to the detection cycle to generate an annihilation rate evolution sequence; calculating the second-order difference to identify curvature inversion nodes based on the annihilation rate evolution sequence to generate a rate inversion inflection point set; reversely calculating the defect concentration change trend based on the rate inversion inflection point set to generate a damage acceleration interval; and using the damage acceleration interval to perform boundary mapping to determine the damage-sensitive time window.

[0021] An annihilation rate evolution sequence is generated by obtaining the annihilation rate time series of the defect evolution curve according to the detection cycle. The κ value of the defect evolution curve at each detection cycle node carries the defect capture rate information of the metal component at the corresponding time. The annihilation rate time series extraction obtains the equivalent annihilation rate λ_eff (dimension ns) by summing the κ value of the defect evolution curve at each detection cycle with the matrix reference annihilation rate λ_bulk according to λ_eff=λ_bulk+κ. -1During the conversion, the λ_bulk value calibrated from the non-destructive test specimen of the same material and the annihilation rate λ_trap of the defect location feature are used as the conversion benchmarks. λ_trap is selected by weighting among three characteristic values—vacancy monomers, vacancy clusters, and micropores—based on the intensity ratios of τ2 and τ3 in the current detection cycle. The converted λ_eff sequence is arranged in ascending order according to the detection cycle to form the annihilation rate evolution sequence. The λ_eff extraction points corresponding to the low-confidence marker segments in the defect evolution curve are given an uncertainty weight in the annihilation rate evolution sequence. The weight value ranges from 0.4 to 0.8, and the weight value is determined by a linear mapping of the confidence level of the corresponding detection cycle κ value. The temporal resolution of the annihilation rate evolution sequence is consistent with the detection cycle interval. In typical fatigue damage detection of metal components, the detection cycle interval is 100-500 hours. The total sequence length is usually 20-60 nodes in the whole life detection scenario. In the accelerated damage section with a large slope of the defect evolution curve, the node density is increased to 2-3 times that of the conventional interval by dense sampling. The densely sampled nodes are included in the annihilation rate evolution sequence with the same conversion benchmark as the conventional nodes and participate in the subsequent second-order difference calculation.

[0022] The second-order difference is used to identify the inflection point set of curvature inversion node generation rate based on the annihilation rate evolution sequence. The first-order difference of the annihilation rate evolution sequence, Δλ_eff(t) = λ_eff(t+1) - λ_eff(t), reflects the periodic rate of change of the annihilation rate (in nanoseconds). -1 / period), second-order difference Δ 2 λ_eff(t) = Δλ_eff(t+1) - Δλ_eff(t) describes the acceleration (dimensions in nanoseconds) at the rate of change. -1 / cycle 2 After performing a second-order difference on the annihilation rate evolution sequence point by point, Δ 2Nodes where λ_eff changes from negative to positive correspond to curvature inversion positions where the annihilation rate growth rate changes from slowing to accelerating, and are the core candidate source for the rate inversion inflection point set. Before the second-order difference calculation, nodes with a weight below 0.6 in the annihilation rate evolution sequence are replaced by linear interpolation of λ_eff from one normal weight node before and after them to avoid low-quality detection points introducing local noise peaks into the second-order difference sequence and generating false inflection points. Nodes in the second-order difference sequence whose absolute value exceeds 1.5 times the standard deviation of the entire sequence mean are marked as significant curvature inversion candidates. When candidate nodes appear repeatedly in two adjacent detection periods, they are merged and the average detection period is taken as a single inflection point position to prevent false inflection points caused by adjacent sampling noise from being repeatedly counted. The rate inversion inflection point set includes all curvature inversion nodes that have passed the significance test. Each record carries three attributes: the corresponding detection cycle number, the absolute value of the second difference, and the local slope of the annihilation rate evolution sequence at that node. The local slope is calculated by linearly fitting the slope of λ_eff of the two nodes before and after the inflection point. The three attributes together characterize the position and intensity of the inflection point in the annihilation rate evolution sequence. The size of the rate inversion inflection point set is usually 2-4 nodes in a typical fatigue damage process. If the number of nodes exceeds 5, it indicates that there is strong noise disturbance in the annihilation rate evolution sequence, and the quality traceability of the original collected data of the corresponding section in the defect evolution curve needs to be performed.

[0023] The damage acceleration interval is generated by reverse-engineering the defect concentration change trend based on the rate inversion inflection point set. The annihilation rate increase change time corresponding to each inflection point in the rate inversion inflection point set is physically lagging behind the actual starting time of the accelerated increase in defect concentration. This lag is due to the fact that the positron capture probability only increases significantly after the defect concentration accumulates to the critical trap density. The quantitative value of the lag is based on the measured calibration correction amount of each material at the corresponding service temperature. For typical steel components, it is about 1-3 detection cycles at room temperature, while for austenitic stainless steel components, it can be extended to 3-5 detection cycles due to the higher vacancy migration activation energy. The actual starting point of the accelerated increase in defect concentration is obtained by subtracting the calibration lag correction amount of the corresponding material from the detection cycle of each inflection point. If the second-order difference of the annihilation rate evolution sequence segment between two adjacent inflection points in the rate inversion inflection point set is continuously positive, the segment is determined to be a monotonically accelerating defect concentration growth segment. This segment, together with the starting node after reverse calculation correction, constitutes a damage acceleration interval record. If the second-order difference shows alternating positive and negative values, only the sub-segment with continuously positive second-order differences is extracted to form an independent record, preventing non-monotonic segments from expanding the boundary range of the damage-sensitive time window. After generation, the damage acceleration interval is labeled with three attributes: the interval start and end detection cycle number, the average annihilation rate growth rate within the interval, and the mean of the second-order difference. These three attributes are used as weight inputs for adaptive adjustment of the window width when mapping the boundary of the damage-sensitive time window. The larger the mean of the second-order difference, the more significant the window boundary contraction corresponding to the damage acceleration interval.

[0024] Damage-sensitive time windows are determined by boundary mapping using damage acceleration intervals. The start and end detection cycles of the damage acceleration interval correspond to the left and right boundaries of the damage-sensitive time window in the boundary mapping. The left boundary is taken as the starting node of the damage acceleration interval, extending forward by L_left detection cycles, and the right boundary is taken as the ending node, extending backward by L_right detection cycles. The extension width is determined by linear mapping of the second-order difference mean of the interval. The higher the second-order difference mean, the narrower the extension width to focus on the rapid evolution stage. The extension width of the left and right boundaries is 0.5-2 detection cycles on typical steel components. Boundary mapping introduces a material-temperature correction coefficient α to adjust L_left and L_right respectively. α is pre-calibrated based on the functional relationship between material vacancy formation energy and component service temperature. For steel components, α is usually taken as 1.0-1.2 under room temperature conditions, and α is increased to 1.5-2.0 under high temperature service conditions to compensate for the hysteresis response lag caused by thermal activation effect, so that the expansion of the damage sensitive time window is fully extended with the high temperature response delay. The α calibration value is stored in the material parameter library. When boundary mapping is executed, the corresponding α value is automatically retrieved by indexing the component material type and service temperature. When the α value exceeds the range of 1.0-2.0, a material parameter library missing warning is triggered, and the calibration data of the material at the corresponding service temperature needs to be manually entered. When multiple damage acceleration intervals exist, the damage-sensitive time window takes the union of the mapping results of each interval boundary. If the distance between adjacent sub-windows in the union is less than one detection cycle, they are merged into a continuous time window. If the distance exceeds one detection cycle, they are retained as independent sub-windows. Each sub-window carries two attributes: the mean of the second-order difference of the corresponding damage acceleration interval and the average growth rate of the annihilation rate. These two attributes are used to distinguish the weights at the sub-window level when the annihilation parameter correction factor performs narrowing correction. The narrowing amplitude of the sub-window corresponding to the higher mean of the second-order difference is enhanced first.

[0025] In some embodiments, obtaining the defect identification momentum ratio parameter from positron annihilation lifetime spectrum data and Doppler broadening data to form an annihilation parameter correction factor includes: performing momentum spectrum decomposition on the positron annihilation lifetime spectrum data and the Doppler broadening data to generate low momentum distribution and high momentum distribution; calculating the ratio deviation based on the low momentum distribution and the high momentum distribution to generate a momentum ratio sequence; identifying nodes where the slope of the ratio sequence changes from positive to negative according to the momentum ratio sequence to generate transition node markers; and performing transition node directional weighted correction based on the transition node markers and the momentum ratio sequence to form an annihilation parameter correction factor.

[0026] Momentum spectrum decomposition was performed on positron annihilation lifetime spectrum data and Doppler broadening data to generate low-momentum and high-momentum distributions. The Doppler broadening data, with its energy broadening of 511 keV annihilated photons, directly reflects the longitudinal momentum distribution of the annihilated electron-positron pair and serves as the main input data for momentum spectrum decomposition. The positron annihilation lifetime spectrum data, being time-domain data, does not directly carry momentum information. Its component lifetimes τi, corresponding to different defect-trapped states, exhibit differentiated contribution ratios in the Doppler broadened low-momentum region. Therefore, it is used as an auxiliary cross-validation input to classify and verify the momentum spectrum decomposition results. In momentum spectrum decomposition, the boundary between the low-momentum and high-momentum regions is defined centered on the 511 keV annihilation peak. The low-momentum region covers the count distribution within the range of 511 keV ± 1.5 keV, while the high-momentum region covers the range of 511 keV ± (2.5-5.0) keV. The boundary delineation of these two regions is based on the calibration of the electron momentum distribution width of the metallic component material. The boundary parameters differ for different alloy systems; the upper limit deviation of the low-momentum region for ferritic steel and austenitic steel can reach 0.2 keV. The low-lifetime component τ1 of the positron annihilation lifetime spectrum data corresponds to matrix free annihilation events and contributes the main peak in the low-momentum region. The long-lifetime component τ3 corresponds to vacancy cluster capture events, and its corresponding electron momentum distribution is significantly narrowed, making its contribution ratio in the low-momentum region higher than that of τ1. The superposition ratio of the two components in the low-momentum distribution varies with the degree of damage. The dynamic change of the superposition ratio is cross-validated with the measured results of the count distribution in the low-momentum region of the Doppler broadening data, ensuring that the extraction results of the low-momentum distribution reflect the current defect state of the metallic component. The high momentum distribution is obtained by integrating the two wing regions of the Doppler broadened data. The core electron momentum distribution is sensitive to the type of alloying element. The high momentum distribution exhibits characteristic differences in metal components with different alloy compositions. The low momentum distribution responds to the changes in vacancy-type defect concentration in the opposite direction to the high momentum distribution. The low momentum distribution expands with increasing vacancy concentration, while the high momentum distribution relatively contracts.

[0027] For example, the step of calculating the ratio deviation based on the low momentum distribution and the high momentum distribution to generate a momentum ratio sequence includes: performing 511keV annihilation peak asymmetry correction on the low momentum distribution and the high momentum distribution to generate a normalized momentum set; calculating the ratio of the integrated area of ​​the low momentum region to the total area at each detection point using the normalized momentum set to generate an original ratio sequence; identifying abnormal jump points using the original ratio sequence to generate a jump point mask; and performing jump-preserving smoothing filtering based on the jump point mask and the original ratio sequence to generate a momentum ratio sequence.

[0028] Asymmetric correction was applied to the 511 keV annihilation peak of both low-momentum and high-momentum distributions to generate a normalized momentum set. In actual measurements, the energy scale drift of the detection system affects the center position of the 511 keV annihilation peak, which can shift by 0.1–0.3 keV. This shift causes an asymmetric misalignment between the lower boundary of the low-momentum distribution and the upper boundary of the high-momentum distribution relative to the peak center. When the peak center shifts 0.2 keV towards the higher energy side, the systematic underestimation of the integral area of ​​the low-momentum distribution can reach 1.2%–2.5%, directly affecting the accuracy of the ratio calculation for the low-momentum to high-momentum distributions. Asymmetric correction uses the measured center position of the annihilation peak in the Doppler broadening data at each detection point as a benchmark. The estimated peak center parameters are extracted by fitting a Gaussian model to the 511 keV peak region. The integration window boundaries of the low-momentum and high-momentum distributions are synchronously shifted to a standard position with the measured peak center as the axis of symmetry. The shift accuracy is better than 0.05 keV. The shift amount at different detection points is calculated independently to account for the slow drift of detector gain with temperature or time. After asymmetric correction, the low-momentum and high-momentum distributions are normalized to a total spectral count area of ​​1.0. The normalization operation eliminates the influence of differences in radiation source intensity attenuation and measurement duration at each detection point on the absolute count. The corrected and normalized low-momentum and high-momentum distributions are combined to form a normalized momentum set. The two types of distributions at each detection point in the normalized momentum set are stored in the form of normalized probability densities. The normalized momentum set participates in the integration with the independent probability density distribution of each detection point in subsequent ratio calculations, ensuring consistency of the normalization benchmark among detection points.

[0029] The original ratio sequence is generated by calculating the ratio of the integrated area of ​​the low momentum region to the total area at each detection point using the normalized momentum set. The cumulative area of ​​the low momentum distribution at each detection point within the 511keV±1.5keV integration window is defined as the integrated area of ​​the low momentum region at that detection point, S_low. The ratio of S_low to the total area of ​​the normalized spectrum (1.0) is the original ratio, R_raw. The calculation of R_raw is performed sequentially for each detection point in the normalized momentum set, and the results are arranged in order of detection cycle to form the original ratio sequence. The sensitivity of the integration window width to R_raw in the normalized momentum set varies at different damage stages. In the stage with higher defect concentration, the count increment of low momentum distribution within the range of 511keV±1.5keV is significant. The influence of a 0.1keV deviation in the selection of the integration window width on R_raw is approximately 0.3%-0.8%. To suppress the systematic error introduced by the window boundary selection deviation, the normalized momentum set performs trapezoidal weighted smoothing on the boundary region of low momentum distribution from 511keV±1.4keV to ±1.6keV before integration. The smoothing width is 0.1keV, so that the count contribution of the boundary region is gradually included in S_low instead of being hard truncated. The trapezoidal weight decreases linearly from 1.0 at 0.1keV inside the boundary to 0.0 outside the boundary. The statistical uncertainty of each R_raw value in the original ratio sequence is calculated by the original count level propagation of the detection points corresponding to the standardized momentum group. The uncertainty propagation takes the Poisson statistical error of S_low and the total count area of ​​the spectrum as input. The R_raw uncertainty of low count points can reach 15%-25% of the mean, while that of high count points is usually less than 5%. The uncertainty difference is recorded synchronously with the original ratio sequence. The original ratio sequence and the uncertainty sequence together constitute the dual input for jump point identification. The detection period with higher uncertainty automatically relaxes the judgment threshold when judging the significance of jump points to avoid noise-dominated pseudo-jumps entering the jump point mask.

[0030] Anomaly jump points are identified using the original ratio sequence to generate a jump point mask. Positions in the original ratio sequence where the absolute value of the difference in R_raw between adjacent detection cycles exceeds twice the global standard deviation of the sequence are identified as candidate anomaly jump points. These jumps are typically caused by radioactive source position shifts, temporary detector gain fluctuations, or abrupt changes in the surface state of the metal component's detection area. All three causes are unrelated to defect evolution itself, and their common characteristic is that the change in R_raw significantly exceeds the normal evolution rate of adjacent detection cycles and is not accompanied by synchronous anomalies in the positron annihilation lifetime spectrum data. Before including candidate anomaly jump points in the jump point mask, reasonable rapid changes caused by genuine damage mutations must be excluded. The exclusion criteria are whether the defect evolution curve shows a synchronous and significant change in the κ value during the corresponding detection cycle. When the synchronous change in the κ value exceeds 1.8 times the expected slope of the defect evolution curve at that stage, it is considered a genuine damage mutation, and this candidate position is not included in the jump point mask but retains the original R_raw value. Candidate positions where no synchronous and significant change in the κ value occurs are considered pseudo-jumps caused by the instrument or operation and are included in the jump point mask. The transition point mask is aligned with the original ratio sequence in Boolean sequence form, with transition point positions marked as True and normal positions marked as False. When the uncertainty of a detection period in the original ratio sequence exceeds 40% of the absolute value of R_raw, that detection period and the 2-standard-deviation criterion are incorporated into the transition point mask using union logic. This prevents wide fluctuations of high-uncertainty nodes from being mixed into the mean calculation of normal nodes during transition-preserving smoothing filtering. The number of True entries in the transition point mask typically accounts for 3%-8% of the total length of the original ratio sequence in typical detection datasets. If the proportion exceeds 10%, it indicates a systematic stability problem in the acquisition process of this batch of Doppler broadening data, requiring backtracking and checking the detector gain drift records and the status archives of the radiation source fixation device.

[0031] Momentum ratio sequences are generated using a transition-preserving smoothing filter based on a transition point mask and the original ratio sequence. The transition-preserving smoothing filter is based on sliding window mean filtering on the original ratio sequence. For positions marked "True" in the transition point mask, a breakpoint crossing prohibition principle is applied. When the sliding window crosses a transition point, the R_raw values ​​on both sides of the transition point are not mixed in the mean calculation; instead, the sliding mean is performed independently on both sides of the transition point. The width of the sliding window on each side of the breakpoint is set to 3 detection cycles. When the number of nodes at the end of the window is less than 3, the mean is performed using the actual number of available nodes, ensuring that the smoothing of the beginning and end segments of the original ratio sequence is consistent with that of the middle segment. After the R_raw value at the jump point mask mark is independently smoothed on both sides, it is replaced by the arithmetic mean of the moving average of the nearest normal nodes to the left and right of the jump point. The interpolation range is limited to one detection cycle before and after the jump point. For consecutive jump points (both adjacent detection cycles are True), the interpolation is extended inward and outward by the smoothed value of the nearest non-jump point, ensuring the continuity of the momentum ratio sequence in the interpolation segment. The original R_raw values ​​of the jump points in the original ratio sequence are retained in the jump point mask and are not included in the output of the momentum ratio sequence. After the jump-preserving smoothing filter is completed, the momentum ratio sequence is stored with the R_smoothed value (dimensionless) of each detection cycle and the corresponding uncertainty. The uncertainty is calculated by combining the in-window variance introduced by the smoothing operation with the original statistical uncertainty. The design of smoothing on both sides of the jump point breakpoint ensures that the momentum ratio sequence remains continuous on both sides of the data quality abnormality segment without numerical aliasing across the breakpoint.

[0032] Transition node markers are generated based on the momentum ratio sequence, identifying nodes where the slope of the momentum ratio sequence changes from positive to negative. The point-by-point slope of the momentum ratio sequence is calculated using the difference in R_smoothed between adjacent detection cycles. Positions where the slope of two or more consecutive detection cycles is negative, and the slope of the preceding cycle is positive, are considered candidate transition nodes where the slope changes from positive to negative. The requirement for consecutive negative slopes excludes false transitions caused by noise fluctuations in a single detection cycle. The physical meaning of a slope changing from positive to negative is that the momentum ratio changes from a relatively expanded low-momentum distribution to a relatively contracted one before and after that detection cycle. This corresponds to the transition from a stage dominated by vacancy cluster capture to a stage of vacancy cluster-micropore coexistence or micropore dominance within the metal component. The position of the candidate transition node is the time anchor point of this transition. In the momentum ratio sequence, the slope at the location of the high uncertainty marker node is replaced by the slope of the three-point moving average centered on that node instead of the original point-by-point slope. The three-point moving average is calculated based on the difference between the mean R_smoothed values ​​of the current node and the nodes before and after it. The original R_smoothed value of the high uncertainty node is reduced to 0.4 in the moving average calculation, ensuring that the interference of the high uncertainty node on the slope sign judgment is effectively suppressed without affecting the slope accuracy of the normal nodes. Candidate transition nodes must pass significance verification. The verification condition is that the difference between the mean values ​​of R_smoothed in the two detection periods before and after the transition exceeds 0.8 times the global standard deviation of the momentum ratio sequence. Candidate nodes that fail the significance verification will not be included in the transition node labeling. Candidate nodes that pass the verification are labeled as transition nodes by recording two pieces of information: the detection period number and the corresponding R_smoothed slope transition amplitude. The slope transition amplitude is defined as the sum of the absolute values ​​of the last positive slope value before the transition and the first negative slope value after the transition. The larger the transition amplitude, the more drastic the switching of the ratio relationship between low momentum distribution and high momentum distribution.

[0033] An annihilation parameter correction factor is formed by performing directional weighted correction of transition nodes based on transition node markers and momentum ratio sequences. The defect type transition time located by the transition node marker corresponds to the segment in the momentum ratio sequence where the sign of the R_smoothed slope changes. In this segment, the ratio of low momentum distribution to high momentum distribution is in a state of coexistence of mixed defect types. The contribution ratios of vacancy clusters and micropores to the low and high momentum distributions respectively continuously change in this segment, resulting in a decrease in the discriminative significance of R_smoothed. The annihilation parameter correction factor performs directional weighted compression on this segment of the momentum ratio sequence with the transition node marker as the anchor point. Directional weighting assigns attenuation weights to the R_smoothed values ​​in one detection cycle before and after the transition node marking. The attenuation weights are linearly mapped to the inverse of the slope change amplitude of R_smoothed at the transition node. A larger slope change amplitude corresponds to lower reliability in identifying the ratio of low-momentum distribution to high-momentum distribution, resulting in a smaller attenuation weight. The weights in the second detection cycle before and after the transition node marking linearly recover to 0.8 based on the attenuation, forming a weight distribution pattern that gradually recovers from the transition node to both sides, avoiding abrupt weight changes at the boundary of the attenuation zone. The annihilation parameter correction factor carries the weighting coefficients corresponding to each detection cycle, indexed by the detection cycle. The weighting coefficients are 0.3-0.6 in the transition node segment and 1.0 in the stable defect type stage. The coefficient distribution pattern determines the window width contraction ratio of each detection cycle in the damage-sensitive time window. The adaptive detection window width corresponding to the transition segment shrinks to 40%-65% of the original width of the damage-sensitive time window after the annihilation parameter correction factor is applied.

[0034] An adaptive detection window is generated by applying an annihilation parameter correction factor to the defect type transition stage of the damage-sensitive time window for window narrowing correction. The window narrowing correction uses the weighting coefficient corresponding to each detection cycle within the damage-sensitive time window as a multiplier. The original window width of that detection cycle within the damage-sensitive time window is multiplied by the weighting coefficient to obtain the corrected width. The window width of the transition section shrinks to 40%-65% of the original width of the damage-sensitive time window after correction. Continuous sections within the damage-sensitive time window with a weighting coefficient lower than 0.5 are identified as strong transition zones. When the span of a strong transition zone is no less than two detection cycles, a unified width strategy for the section is adopted. The corrected window width for each detection cycle within the section is uniformly taken as the width corresponding to the minimum weighting coefficient of that section. Local low-weight nodes with a span of only a single detection cycle do not adopt the unified strategy but are still calculated according to the cycle-by-cycle multiplier rule. This avoids the window width fluctuations within the strong transition zone interfering with the continuity of the detection scheme, and also prevents a single low-weight point from being misjudged as an entire strong transition, resulting in excessive window shrinkage. The weighting coefficient of the damage-sensitive time window remains at 1.0 during the stable defect type phase, and the window width for the corresponding detection cycle remains unchanged from the original width of the damage-sensitive time window. The comparison of window widths between the stable and transition phases of the defect type forms an alternating distribution of wide and narrow windows within the adaptive detection window. The wide window segment covers the stable damage accumulation period dominated by a single defect type, while the narrow window segment precisely locks the critical transition period of defect type switching. After calibration for all detection cycles, the adaptive detection window is stored as a window width sequence and corresponding detection cycle number. The width value of each detection cycle in the window width sequence corresponds one-to-one with the weighting coefficient of the annihilation parameter correction factor.

[0035] Step S130: Adapt the detection scheme to the adaptive detection window to form a detection configuration table. Identify the defect capture path dominated by low-intensity long-life components according to the detection configuration table. Calculate the annihilation rate gradient of the defect capture path to obtain the annihilation lag time. Determine the damage level based on the annihilation lag time to generate a damage layer configuration.

[0036] Specifically, a detection configuration table is formed by adapting the detection scheme to the adaptive detection window. After window narrowing correction, the width sequence of the adaptive detection window exhibits an alternating distribution of wide and narrow windows. The window width of the detection cycle corresponding to the narrow window segment is typically only 40%-65% of that of the wide window segment. The detection scheme adaptation assigns differentiated single-point acquisition statistics, acquisition dwell time, and component fitting accuracy requirements to each detection cycle based on the difference in width and narrow distribution. Due to the compression of detection time and the reduction in the number of detectable events that can be accumulated within a single cycle, the single-point dwell time needs to be increased or the number of spatial sampling points within a single cycle needs to be increased to maintain the statistical accuracy of the positron annihilation lifetime spectrum data in the narrow window segment. The wide window segment is executed with a conventional dwell time and standard sampling density. The difference in acquisition configuration between the two types of segments is distinguished in the detection configuration table by the ratio of the number of accumulated events at a single point. The lower limit of the number of accumulated events in the narrow window segment is usually set to 1.4-2.0 times that of the wide window segment. The rows in the detection configuration table are indexed by each detection period of the adaptive detection window. Each row carries four parameters for that detection period: the lower limit of the cumulative event count at a single point, the single-point acquisition duration, the minimum accuracy requirement for component fitting, and the number of spatial sampling points in the detection area. The combination of these four parameters ensures that each detection period can meet the minimum statistical quality requirements for lifetime spectrum component parameter extraction under the constraints of the adaptive detection window. After the parameters for each detection period are assigned, the detection configuration table performs a global consistency check. The check ensures that the change in the lower limit of the cumulative event count between adjacent detection periods does not exceed 0.3 times the baseline value. If this change is exceeded, a transition configuration row is inserted between the two detection periods. The lower limit of the cumulative event count in the transition configuration row is taken as the linear interpolation of the two adjacent periods to avoid large jumps in the acquired statistics during continuous detection, which could cause discontinuities in the lifetime spectrum fitting accuracy between periods.

[0037] In some embodiments, identifying the defect capture path dominated by low-intensity long-lifetime components according to the detection configuration table includes: extracting the lifetime spectrum component intensity of each detection region in the detection configuration table to generate a component intensity distribution; comparing the long-lifetime component intensity through the component intensity distribution to generate a component intensity deviation sequence; filtering the expected dominant but actually weakened segments based on the component intensity deviation sequence to generate a suppressed weak component set; and performing trap competition effect analysis on the suppressed weak component set to locate the defect capture path.

[0038] The lifetime spectrum component intensities of each detection area in the detection configuration table are extracted to generate a component intensity distribution. The lifetime spectrum component intensities of each detection area in the detection configuration table are recorded using the Ii value of the multi-component fitting result, including the intensity ratios of the three components τ1, τ2, and τ3. In the spatial distribution of the metal component detection area, I3 typically reaches 12%-20% in high-damage areas, while in low-damage areas far from stress concentration zones, it is typically below 3%. The absolute value of I3 in high-damage areas is 10-18 percentage points higher than that in low-damage areas; this difference constitutes the most significant distinguishing signal in the spatial gradient of the component intensity distribution. The component intensity distribution is organized into a two-dimensional intensity matrix using the spatial coordinates of the metal component detection area as an index. Each spatial detection point in the matrix corresponds to the intensity value of the three components τ1, τ2, and τ3. The spatial distribution shape of the τ3 intensity matrix visually presents the spatial distribution pattern of damage in the metal component, and the connectivity of the high I3 region reflects the expansion direction of defects within the metal component. When extracting component intensity, the Ii value of the detection points marked with low statistics in the detection configuration table is marked with uncertainty. Detection points with uncertainty exceeding 20% ​​of the absolute value of Ii are completed in the spatial matrix of component intensity distribution by interpolation. The interpolation is taken as the weighted average of the Ii values ​​of four adjacent detection points. The weight is determined by the reciprocal of the distance from each neighboring point to the target point. After interpolation completion, the spatial matrix of component intensity distribution covers all spatial detection points in the detection area of ​​the metal component, forming a complete intensity distribution map without any missing points. The interpolation mark of each detection point in the intensity distribution map is saved synchronously with the matrix.

[0039] A component strength deviation sequence is generated by comparing the long-life component strength distribution. The spatial reference benchmark is determined by the mean I3 value of the undamaged area under the same service time for similar metal components. The reference benchmark is extracted from the low-damage section far from the high-stress concentration zone in the metal component's inspection area. The median of the I3 values ​​of each inspection point in this section is taken as the spatial reference benchmark. The median has better resistance to outlier interference than the mean, and can maintain the stability of the reference benchmark even when there are a few local abnormal inspection points in the low-damage section. The long-life component strength comparison defines the difference between the I3 value of each spatial inspection point in the component strength distribution and the spatial reference benchmark as the component strength deviation ΔI3. When ΔI3 is positive, it indicates that the τ3 component strength of the inspection point is higher than the reference level. The spatial distribution of ΔI3 is arranged according to the inspection point coordinates to form a component strength deviation sequence. The spatial gradient of ΔI3 in the component strength deviation sequence is characterized by the absolute value of the difference of ΔI3 between adjacent detection points. The gradient abrupt change location usually corresponds to the inhomogeneous transition zone of the material microstructure in the detection area of ​​the metal component. The component strength deviation sequence records the ΔI3 value and the corresponding spatial gradient value of each detection point simultaneously. The two values ​​together support the screening judgment of the suppressed weak component set. The component strength deviation sequence is calculated separately for the τ2 and τ3 components. The sign combination of the deviation sequence ΔI2 of the τ2 component and ΔI3 of the τ3 component at the same detection point reflects the defect type characteristics of that point. Detection points where ΔI2 is positive and ΔI3 is close to zero correspond to the early damage state dominated by single vacancies. Detection points where ΔI3 is significantly positive and ΔI2 is relatively declining correspond to the middle and late damage state dominated by vacancy clusters and micropores (the strength of the τ2 position is intercepted by the τ3 position due to competitive capture, resulting in a decline in strength). The spatial distribution of the two types of sign combinations clearly presents the spatial process of damage evolution in the component strength deviation sequence.

[0040] Suppressed weak component sets are generated by screening expected dominant but actually weakened segments based on component intensity deviation sequences. Segments where the expected τ3 component should dominate are determined by the damage stage of the defect evolution curve. Within the detection period corresponding to the accelerated growth segment of the defect evolution curve, the τ3 component should, according to physical laws, produce a significant positive deviation at the detection point in the high stress concentration area. The expected value of ΔI3 is estimated based on the standard damage curve of the same material, typically 1.5-3.0 times the reference baseline. Detection points where the measured ΔI3 is 60% lower than the expected value are identified as candidate points where the τ3 component is suppressed. The physical mechanisms of the measured weakening include two types: one is that competitive capture by adjacent defect types leads to the positrons of the τ3 component being intercepted at the τ2 position; the other is that a high-density dislocation network exists locally at the detection point, forming a dispersed capture center that inhibits the formation of vacancy clusters. The former is manifested in the component intensity deviation sequence as a higher ΔI2 and a lower ΔI3, while the latter is manifested as both ΔI2 and ΔI3 being simultaneously lower than expected. The characteristic combinations of these two mechanisms are simultaneously labeled during the screening of suppressed weak component sets to distinguish the physical causes. The suppressed weak component set includes all detection point records where the measured ΔI3 is 60% lower than the expected value. Each record carries five attributes: spatial coordinates of the detection point, measured ΔI3 value, expected ΔI3 value, suppression ratio, and ΔI2 value. The suppression ratio is defined as the ratio of measured ΔI3 to expected ΔI3. The lower the suppression ratio, the stronger the competitive capture effect of the τ3 component at that detection point. The size of the suppressed weak component set is 15%-30% of the total number of detection points on a typical moderately damaged metal component. When the size is less than 10%, it indicates that the defect capture path has not yet formed a clear spatial concentration trend, and the component strength deviation calculation needs to be re-executed after extending the detection cycle interval.

[0041] Trapping competition effect analysis is performed on the suppressed weak component set to locate the defect capture path. The defect capture path described in this step is defined as the spatially connected band in the metal component detection area where positron annihilation events are most significantly intercepted by abnormal defect locations, reflecting the spatial concentration trend of defect type competitive capture. The suppression ratio of each detection point in the suppressed weak component set is combined with the ΔI2 value for trap competition effect analysis. The spatial distribution of competitive capture suppression points in the metal component detection area typically exhibits a continuous band-like trend, with the direction of the band parallel to the principal stress direction inside the metal component. The connected band of competitive capture suppression points is the spatial segment with the highest interception density of τ3 component positrons, and this connected band corresponds to the core location target of the defect capture path. The trap competition effect analysis constructs a spatial adjacency graph using the spatial coordinates of each detection point in the suppressed weak component concentration as nodes. The edge weight between adjacent nodes in the adjacency graph is assigned as the reciprocal of the average suppression ratio of the two nodes. The lower the suppression ratio, the higher the edge weight between adjacent nodes, reflecting the more intense the defect capture competition in that path segment. A maximum weight path search is performed on the spatial adjacency graph. The starting point of the path is the detection point with the lowest suppression ratio in the suppressed weak component concentration, and the ending point is the detection point extending along the principal stress direction to the opposite edge of the metal component detection area with a similarly low suppression ratio. The maximum weight path is identified as the defect capture path. The defect capture path is described by three data items: the spatial coordinate sequence of each detection point along the path, the suppression ratio sequence, and the ΔI2 sequence. Detection points with a suppression ratio below 0.4 on the path are marked as strong capture nodes. The spatial density of strong capture nodes on the defect capture path is typically 3-5 times that of non-path areas. When multiple maximum weight path candidates exist in the spatial adjacency graph, the average density of strong capture nodes on each candidate path is used as the criterion, and the candidate path with the highest average density of strong capture nodes is selected as the final defect capture path.

[0042] In some embodiments, the step of calculating the annihilation rate gradient to obtain the annihilation lag time for the defect capture path includes: calculating the annihilation rate of each region based on the defect capture path to generate an annihilation rate spatial distribution; performing non-uniform spacing weighted difference calculation on the annihilation rate spatial distribution to generate an annihilation rate gradient sequence; identifying segments that maintain a high gradient after a sudden change based on the annihilation rate gradient sequence to generate damage front markers; and performing lag analysis based on the damage front markers and the defect capture path to obtain the annihilation lag time.

[0043] The annihilation rate spatial distribution is generated by calculating the annihilation rate of each region based on the defect capture path. The annihilation rate of each detection point on the defect capture path is converted into an equivalent annihilation rate λ_eff using the lifetime spectrum component parameters of the three-state capture model. During the conversion, λ_bulk and λ_trap calibrated by the same material non-destructive specimen are used as the benchmark. λ_trap is selected by weighting among three characteristic values ​​of vacancy monomers, vacancy clusters, and micropores based on the intensity ratio of τ2 and τ3 at the current detection point. The λ_eff values ​​of each detection point are arranged with spatial coordinates as the index to form the annihilation rate spatial distribution along the defect capture path. The annihilation rate of the metal component detection area outside the defect capture path is calculated simultaneously. The λ_eff values ​​of the detection points outside the path and the detection points inside the path together constitute the annihilation rate spatial distribution matrix covering the entire detection area. The comparison of λ_eff of the detection points inside and outside the path in the matrix reveals the degree of annihilation rate concentration along the defect capture path direction. Spatial locations where the difference in λ_eff between inside and outside the path exceeds 1.5 times the standard deviation of the mean of the entire matrix are marked as high salience areas of the path. The connectivity of the high salience areas of the path verifies the spatial rationality of the defect capture path positioning. The λ_eff of each detection point in the annihilation rate spatial distribution matrix includes the uncertainty calculated by the goodness-of-fit propagation of the lifetime spectrum component parameters, and the goodness-of-fit χ. 2 / ν (where ν is the statistical degree of freedom of the multi-component fitting) The λ_eff uncertainty of the detection point corresponding to the high-uncertainty segment can reach 8%-15% of the mean. Spatial interpolation smoothing is performed on the detection point with high uncertainty. The interpolation is taken as the weighted mean of the 3×3 neighborhood centered on the detection point. The weight is determined by the reciprocal of the λ_eff uncertainty of each neighborhood point, to ensure that the high uncertainty detection point does not form an isolated abnormal peak in the annihilation rate spatial distribution and affect the gradient calculation.

[0044] An annihilation rate gradient sequence is generated by non-uniform spacing weighted difference calculation on the spatial distribution of annihilation rate. The spacing between adjacent detection points in the λ_eff sequence unfolded along the defect capture path is not uniform. The spacing between detection points in the high stress concentration zone is smaller than that in the weld heat-affected zone, with the difference reaching 2-3 times. Directly using the difference between adjacent λ_eff points as the gradient will introduce systematic bias due to the non-uniform spacing. Non-uniform spacing weighted difference calculation divides the difference between adjacent detection points' λ_eff points by the actual spatial spacing between the two points to obtain the change in annihilation rate per unit distance. The gradient calculation formula is G(i)=(λ_eff(i+1)-λ_eff(i)) / d(i,i+1), where d(i,i+1) is the actual spatial spacing (in mm) between the i-th and i+1-th detection points, and G(i) is in nanoseconds. -1 / mm, the smaller the spacing between adjacent points, the higher the spatial representativeness of the resulting gradient value. Non-uniform spacing weighted difference is applied sequentially to all adjacent detection point pairs along the defect capture path direction in the spatial distribution of annihilation rate. The G(i) values ​​are arranged according to the spatial position of the path direction to form an annihilation rate gradient sequence. The spatial resolution of the gradient sequence in the high stress concentration section is better than that in the weld heat-affected zone. The resolution difference between the two types of sections is marked in the spatial position index of the annihilation rate gradient sequence with the actual coordinates of the detection points. After the annihilation rate gradient sequence is calculated, a gradient reliability mark is added to the high uncertainty interpolation section. The absolute value of the gradient in the reliability mark section is compressed due to the interpolation smoothing, and the reliability is lower than that of the non-interpolation section calculated with the original λ_eff. When the reliability mark continuously covers more than two adjacent detection point pairs in the annihilation rate gradient sequence, the gradient reliability of the continuous segment is further reduced. In the damage stratification configuration stage, a low reliability note is added to the damage level classification result of this spatial region.

[0045] Damage front markers are generated by identifying sections that maintain a high gradient level after a sudden gradient change based on the annihilation rate gradient sequence. The spatial distribution of the annihilation rate gradient sequence along the defect capture path is not monotonic. The gradient value at the boundary between the high stress concentration zone and the weld heat-affected zone typically exhibits a step-like characteristic of a steep rise followed by a plateau. The step amplitude in typical mid-term fatigue-damaged metal components is 2.3-4.5 times the mean of the gradient sequence. This step-like characteristic is the core morphological basis for identifying candidate sections of the damage front. A gradient abrupt change is identified when the gradient value at a certain spatial location in the annihilation rate gradient sequence increases by more than 1.8 times the standard deviation of the overall sequence gradient mean compared to the previous location. Confirmation of the abrupt change initiation requires that the gradient values ​​of the two consecutive detection point pairs preceding this location are both less than 1.0 times the overall sequence mean, ensuring that the gradient jump at the abrupt change initiation originates from a stable low-gradient region rather than an occasional peak in local fluctuations. A segment is considered a candidate damage front when the gradient values ​​of three or more consecutive adjacent detection point pairs after the abrupt change remain above 1.2 times the average gradient of the entire sequence. The maintenance endpoint of a candidate segment is determined by the location of the detection point pair where the gradient value first falls below 1.0 times the average gradient of the entire sequence. If the gradient rises above 1.2 times again after the endpoint and continues for three consecutive points, a new independent labeling segment is initiated. When the extension length of a candidate damage front exceeds 30% of the total path length, it is split into multiple independent labels based on the interval where the gradient falls below 1.0 times the average gradient as the internal boundary. Short segments with fewer than three gradient maintenance detection points are merged into the adjacent longest candidate segment to preserve its front location information. Damage front labels are recorded using two attributes: the start and end spatial coordinates of the candidate segment and the number of gradient maintenance detection points. The start and end spatial coordinates determine the spatial coverage of the damage front, while the number of gradient maintenance detection points reflects the statistical reliability of the labeled segment; a higher number indicates stronger physical evidence of high gradient maintenance.

[0046] Hysteresis lag time is obtained by hysteresis analysis based on damage front markers and defect capture paths. The spatial location of the high-gradient segment located by the damage front markers on the defect capture path differs spatially from the actual peak position of the defect concentration. The offset direction is along the damage propagation direction of the path. This spatial offset is converted into a time delay by combining the load cycle frequency of the metal component and the material defect propagation rate. The conversion is based on the number of load cycles corresponding to a unit millimeter spatial offset divided by the number of load cycles per unit detection cycle. For typical steel components under room temperature conditions, the time delay corresponding to a unit millimeter spatial offset is approximately 0.5-1.5 detection cycles. For austenitic stainless steel components, due to the higher vacancy migration activation energy, the conversion factor increases to 1.2-2.5 detection cycles per millimeter. The distance between the spatial starting point of each damage front marker segment on the defect capture path and the spatial centroid of the concentrated area of ​​strong capture nodes is defined as the characteristic spatial offset of that marker segment. The spatial centroid of the concentrated area is obtained by weighting the coordinates of the strong capture nodes according to the inverse of the suppression ratio of each node. The characteristic spatial offset of each marker segment is converted to obtain the corresponding estimated time delay value. The weighted average of the estimated time delay values ​​of multiple damage front marker segments is taken as the comprehensive annihilation lag time. The weights are determined by the number of gradient maintenance detection points in each marker segment. The annihilation lag time is calculated separately for each metal component detection area. The annihilation lag time in the high stress concentration area is usually 0.5-1.5 detection cycles. Due to the tortuous defect propagation path caused by material inhomogeneity in the weld heat-affected zone, the annihilation lag time can be extended to 2-4 detection cycles. When the annihilation lag time of a certain detection area exceeds 4 detection cycles, it indicates that the defect evolution rate in that area is significantly lower than the statistical law of similar components. It is necessary to retrospectively check the load measurement records and material composition files of that area to confirm whether there is local material strengthening or load underestimation.

[0047] Damage levels are determined based on annihilation lag time, generating a damage stratification configuration. The annihilation lag times of each metal component's inspection area are ranked numerically and assigned to three damage layers: areas with lag times less than one inspection cycle are classified as high-sensitivity damage layers, characterized by rapid defect evolution, concentrated capture paths, and significant damage front advancement; areas with lag times between one and three inspection cycles are classified as medium-sensitivity damage layers, where vacancy clusters and micropores compete for capture, and the damage process is in a transitional state of alternating acceleration and stability; and areas with lag times exceeding three inspection cycles are classified as low-sensitivity damage layers, typically located at the edge of the weld heat-affected zone or in low-gradient sections far from the principal stress direction, where the response delay of defect concentration to external loads is most significant. The boundaries of the three damage layers are primarily determined by the aforementioned lag time thresholds (1, 3). Adjacent inspection areas with a transition zone width less than 0.3 inspection cycles near the boundary are assigned to the nearest layer based on strong capture node density, with higher density areas preferentially assigned to the high-sensitivity layer direction. In the damage stratification configuration, each detection area record carries the damage level label, annihilation lag time value, defect capture path number, and spatial coordinate range of the detection area. The statistical mean and standard deviation of the annihilation lag time of each partition are written simultaneously. The smaller the standard deviation, the more uniform the damage evolution rate of each detection area within the damage layer. When the standard deviation exceeds 40% of the layer mean, it indicates that there is local material inhomogeneity or load distribution deviation within the layer. The correspondence between the lifetime spectrum component parameters and the defect capture path needs to be individually reviewed for detection areas with large standard deviations. After the damage stratification configuration is completed with the labeling of the three types of damage layers, spatial connectivity verification is performed. Detection areas within the same damage layer should form connected sub-regions in the space of the metal component. The stratification of isolated detection points is corrected based on the damage layer with the largest area among the surrounding adjacent areas.

[0048] Step S140: Perform cross-state hysteresis consistency evaluation on the damage stratification configuration to determine the optimal excitation mode, detect the optimal component decomposition window data of the optimal excitation mode, perform fluctuation priority sorting based on the optimal component decomposition window data to form a detection execution sequence, and perform joint analysis of damage degree based on the detection configuration table and the detection execution sequence to output fatigue damage detection results.

[0049] In some embodiments, the step of determining the optimal excitation mode by performing cross-state hysteresis consistency assessment on the damage stratification configuration includes: determining the hysteresis time of each excitation mode under the initial state, service state, and failure state of the damage stratification configuration to generate a cross-state delay sequence group; calculating the consistency difference through the cross-state delay sequence group to generate a consistency deviation coefficient; identifying excitation modes whose deviations monotonically converge with the deepening of damage based on the consistency deviation coefficient to generate a robust excitation candidate set; and using the robust excitation candidate set to perform optimal excitation matching to determine the optimal excitation mode.

[0050] A cross-state time delay sequence group is generated by determining the lag times of each excitation mode under the initial, service, and failure states in the damage stratification configuration. In the damage stratification configuration, the high-sensitivity, medium-sensitivity, and low-sensitivity damage layers correspond to the failure, service, and initial states in the fatigue damage process of the metal component, respectively. The statistical mean of the annihilation lag times of the three types of damage layers constitutes the state benchmark for evaluating the lag times of the excitation modes. The accuracy of the benchmark value directly determines the ability of the cross-state time delay sequence group to distinguish different excitation modes. The lag time of each excitation mode for the three damage states is characterized by the mean annihilation lag time of that excitation mode in the detection area of ​​the corresponding damage layer. The excitation modes include three timing configurations: continuous beam injection, pulsed beam injection, and variable energy beam injection. The difference in lag time between the three types of excitation modes in the high-sensitivity damage layer is typically 0.3-0.8 detection cycles, while the difference in the low-sensitivity damage layer can expand to 1.2-2.5 detection cycles. The earlier the damage state, the greater the difference in lag time between excitation modes. The degree of material inhomogeneity in the detection area of ​​the low-sensitivity damage layer in the damage stratification configuration is the main factor affecting the magnitude of this difference. The cross-state delay sequence group constructs a delay matrix with the excitation mode type as the row index and the three damage states (initial state, service state, and failure state) as the column index. Each element in the matrix is ​​the mean annihilation lag time of the corresponding excitation mode in the corresponding damage state. The standard deviation of the lag time of each damage layer in the damage layer configuration is synchronously written into the cross-state delay sequence group as the uncertainty attribute of each element. Elements with larger standard deviations have their weight reduced to 0.6 when calculating the consistency difference, so as to prevent the discrete values ​​of the non-uniform region inside the damage layer from interfering with the consistency evaluation results of the excitation mode. The delay matrix and uncertainty matrix of the cross-state delay sequence group together constitute the complete dual input for the consistency difference calculation.

[0051] For example, the step of calculating the consistency difference through the cross-state delay sequence group to generate a consistency deviation coefficient includes: calculating the mean lag time of each state in the cross-state delay sequence group to generate a state mean group; performing cross-state lag slope trend analysis through the state mean group to generate a state slope monotonicity distribution; identifying non-monotonic excitation modes based on the state slope monotonicity distribution to generate exclusion markers; and performing screening based on the exclusion markers and the cross-state delay sequence group to generate a consistency deviation coefficient.

[0052] The mean lag time of each state in the cross-state delay sequence group is calculated to generate a state mean group. The delay matrix of the cross-state delay sequence group calculates the mean of all excitation mode delay values ​​under the same damage state column in the initial state, service state, and failure state columns. These three mean values ​​constitute the three components of the state mean group. When averaging columns, each delay element uses the reciprocal of the corresponding uncertainty in the cross-state delay sequence group as a weighting coefficient for summation and normalization. Elements with larger standard deviations have lower weights to avoid excessive amplification of discrete values ​​in the non-uniform regions within the damage layer during mean calculation. The mean of the initial state column is typically 0.8-2.0 detection cycles higher than the mean of the failure state column. This difference reflects the overall decrease in annihilation lag time as the metal component evolves from a damage-free state to near failure. Typical values ​​for steel components range from 1.5-3.5 detection cycles for the initial state and 0.5-1.2 detection cycles for the failure state. The larger the absolute difference between the two components, the more effectively the damage stratification configuration distinguishes the three types of damage layers. The three components of the state mean group are accompanied by the total uncertainty calculated by combining and propagating the uncertainties of each element in the corresponding column of the cross-state delay sequence group. The total uncertainty is combined in the form of the square root of the sum of squares of the uncertainties of each element, and the contribution terms of the elements that have undergone weight reduction are reduced synchronously. The total uncertainty of the initial state component is usually higher than that of the failure state component because the standard deviation of the annihilation lag time corresponding to the low-sensitivity damage layer in the initial state is larger in the damage stratification configuration. The distribution of the total uncertainty in the three components reflects the consistency of the annihilation response in the detection area within each damage layer of the damage stratification configuration. The systematic decrease of uncertainty as the damage state deepens is an indication of good uniformity of the detection area within the damage layer. Conversely, if the uncertainty of the failure state component exceeds that of the initial state component, it indicates that there is an abnormal distribution of local defects within the high-sensitivity damage layer, which requires triggering a backtracking check.

[0053] Cross-state hysteresis slope trend analysis is performed using state mean groups to generate a monotonic distribution of the inter-state slope. The three components of the state mean group are arranged according to the initial state, service state, and failure state and denoted as μ_initial, μ_service, and μ_failure, respectively (μ represents the statistical mean of the annihilation hysteresis time under the corresponding damage state, with the dimension being the detection period). The difference in mean between adjacent states is defined as the inter-state hysteresis slope. The slopes Δτ1 = (μ_service - μ_initial) and Δτ2 = (μ_failure - μ_service) characterize the rate of change of annihilation hysteresis time when the damage evolves from the initial state to the service state and from the service state to the failure state, respectively. The signs and relative magnitudes of the two slopes together characterize the cross-state response trend of the excitation mode. Inter-state hysteresis slope trend analysis calculates the corresponding Δτ1 and Δτ2 for each excitation mode in the trans-state time delay sequence group. The sign and magnitude relationship of Δτ1 and Δτ2 for each excitation mode form the slope characteristic pair of that excitation mode. The slope characteristic pair of typical monotonically convergent modes satisfies Δτ1<0 and Δτ2<0 and |Δτ1|>|Δτ2|, that is, the annihilation hysteresis time continuously decreases as the damage deepens and the rate of decrease slows down in stages. The slope characteristic pair of non-monotonic modes shows that Δτ1 and Δτ2 have opposite signs or |Δτ1|<|Δτ2|. The former indicates that the annihilation response direction of the excitation mode is reversed in a certain damage state interval, and the latter indicates that the deepening of damage accelerates the change of hysteresis time instead of tending to stabilize. The inter-state slope monotonicity distribution records the slope feature pairs of each excitation mode. The ratio of monotonically convergent to non-monotonic types is usually 1:1 to 2:1 in the evaluation of the three typical excitation modes. The slope feature log values ​​of each excitation mode in the distribution are stored synchronously with the uncertainty of the state mean group. The slope feature pairs corresponding to the state mean group with higher uncertainty have lower confidence in the judgment. Slope feature pairs with confidence below 0.7 are marked with low confidence in the inter-state slope monotonicity distribution.

[0054] Excitation modes with non-monotonic slopes are identified and exclusion markers are generated based on the monotonicity distribution of slopes between states. The slope characteristics of each excitation mode in the monotonicity distribution of slopes between states are checked one by one according to the monotonic convergence judgment rule. The judgment rule is that Δτ1<0 and Δτ2<0 and |Δτ1|-|Δτ2|>0.1 detection cycles. Excitation modes that do not meet all three conditions are judged as having non-monotonic slopes. The physical meaning of non-monotonic slopes is that the annihilation response of the excitation mode shows a reversal or acceleration at a certain stage of damage evolution, and cannot reliably track the monotonic damage process of the metal component from the initial state to the failure state. The identification of non-monotonic slopes covers two typical forms: Δτ1 and Δτ2 with opposite signs correspond to a reversal of response direction in damage evolution. This reversal usually originates from the oversensitivity of the excitation mode to moderately damaged regions, causing the time delay of the moderately sensitive damage layer to deviate abnormally from the damage layers at both ends. Δτ1 and Δτ2 with the same sign, but |Δτ1| < |Δτ2|, correspond to accelerated divergence of the response, reflecting the deepening of damage and the expansion of the lag time change amplitude rather than stabilization. In non-monotonic excitation modes, cases where Δτ1 and Δτ2 have opposite signs are marked as strong exclusion candidates, while weakly monotonic cases where |Δτ1| - |Δτ2| is between 0 and 0.1 are marked as weak exclusion candidates. Strong exclusion candidates are unconditionally removed during the screening process of cross-state time delay sequence groups, while weak exclusion candidates are retained and marked with a weak monotonic warning in the extreme case where the robust excitation candidate set is empty. When all excitation modes are marked as strong exclusion candidates, the decision threshold |Δτ1|-|Δτ2| is relaxed from 0.1 detection cycles to 0.05 detection cycles, and the exclusion marker generation is re-executed. During the regeneration stage, the previously strongly excluded modes are checked a second time according to the relaxed threshold. Modes that pass the second check are downgraded from strong exclusion to weak exclusion. The exclusion marker is recorded with three pieces of information: excitation mode type, slope feature logarithm value, and exclusion intensity level. The complete record of exclusion markers covers all excitation modes that are judged to have non-monotonic slopes in the slope monotonicity distribution between states.

[0055] Consistency deviation coefficients are generated by screening based on exclusion markers and cross-state delay sequence groups. Excitation modes with strong exclusion markers are removed entirely from the delay matrix of the cross-state delay sequence group. Excitation modes with weak exclusion markers are removed synchronously when the robust excitation candidate set is not empty. When the robust excitation candidate set is empty, the row with weak exclusion markers is retained and a low reliability weight of 0.5 is added in subsequent calculations. This low reliability weight reduces the contribution of the delay value of the weak exclusion marker excitation mode to the CV calculation, avoiding excessive interference from its unstable response characteristics on the consistency deviation coefficient. After the elimination labeling process, the time delay matrix of the cross-state time delay sequence group retains only the row of excitation modes with monotonically convergent slopes. The coefficient of variation (CV) across the three damage states is calculated for each retained row. Let μ_avg (dimension: detection period) be the arithmetic mean of the annihilation lag time of the excitation mode in the initial, service, and failure states, and σ (dimension: detection period) be the standard deviation of the time delay value of the three states relative to μ_avg. The coefficient of variation is defined as CV = σ / μ_avg (dimensionless). The elimination process reduces the set of excitation modes participating in the CV calculation from the original three categories to a monotonically convergent subset. The range of CV values ​​for each excitation mode in the subset is narrowed as a whole because modes with unstable response directions are excluded. The uncertainty attribute of each excitation mode in the retained row participates in the CV calculation in the form of weights. The weight of the state corresponding to the element with high uncertainty is reduced to 0.6 when calculating the standard deviation σ, ensuring that the time delay discreteness of the high uncertainty damage state is not excessively amplified to the CV numerator. After the consistency deviation coefficient is removed, a sequence of CV values ​​of each excitation mode is retained. Each CV value in the sequence is accompanied by the CV uncertainty calculated by the uncertainty propagation of the corresponding row uncertainty of the transstate time delay sequence group. Excitation modes with CV uncertainty exceeding 30% of the CV value itself are marked with high uncertainty in the consistency deviation coefficient sequence. The participation weight of the excitation mode marked with high uncertainty is reduced to 0.7 in the robust excitation candidate set screening stage.

[0056] Robust excitation candidate sets are generated based on the consistency deviation coefficient, which identifies excitation modes whose deviations monotonically converge with increasing damage. The consistency deviation coefficient characterizes the time delay dispersion of each excitation mode across the three damage states: initial state, service state, and failure state. Low dispersion does not necessarily mean that the excitation mode has a tendency to converge in the direction of increasing damage; therefore, the generation of robust excitation candidate sets is based on the low dispersion criterion plus the monotonically convergent criterion. The physical meaning of monotonically convergent is that the rate of decrease in time delay from the initial state to the failure state of the excitation mode gradually slows down with increasing damage and tends to saturate near the failure state, so that the detection results of the mode in the most severely damaged area are not affected by response rate fluctuations. The quantitative index of monotonically convergent is the difference between |Δτ1| and |Δτ2| for each excitation mode. The larger the |Δτ1|-|Δτ2|, the slower the rate of decrease in lag time from the service state to the failure state compared to the initial state-service state stage. Excitation modes with a consistency deviation coefficient lower than the mean CV of all participating excitation modes and a |Δτ1|-|Δτ2| difference exceeding 0.1 detection cycles are included in the robust excitation candidate set. In the dual parallel criterion, the consistency deviation coefficient reflects the overall cross-state dispersion being below the mean level, while |Δτ1|-|Δτ2| reflects the significant stage-by-stage convergence characteristics in the direction of damage deepening. Each record in the robust excitation candidate set carries three attributes: excitation mode type, consistency deviation coefficient value, and |Δτ1|-|Δτ2| difference. Among the three attributes, the consistency deviation coefficient serves as the primary sorting key, and the |Δτ1|-|Δτ2| difference serves as the secondary sorting key in the descending order of the optimal excitation matching stage. The size of the robust excitation candidate set is usually 1-2 in the typical evaluation results of the three types of excitation modes. In the extreme case of zero size, the |Δτ1|-|Δτ2| threshold is relaxed to 0.05 detection cycles for re-screening.

[0057] The optimal excitation mode is determined by optimal excitation matching using a robust excitation candidate set. The consistency deviation coefficients of each excitation mode in the robust excitation candidate set are sorted in ascending order within the set, with the excitation mode having the smallest consistency deviation coefficient being the preferred optimal excitation mode. When the robust excitation candidate set contains only one member, it is directly determined as the optimal excitation mode. When the robust excitation candidate set contains two or more members and the difference in consistency deviation coefficients is less than 0.03, the mean standard deviation of the annihilation lag time of each damage layer in the damage layer configuration is introduced as an auxiliary criterion. Excitation modes with a smaller mean standard deviation value correspond to more uniform annihilation responses in the detection areas within the damage layer and are given priority as optimal excitation modes. Excitation modes in the robust excitation candidate set with a mean delay of less than 0.8 detection cycles in highly sensitive damage layers receive additional priority during optimal excitation matching. This additional priority is used in the sorting process by subtracting 0.02 from the consistency deviation coefficient, ensuring that the response sensitivity of the optimal excitation mode is guaranteed in the most severely damaged areas. When the additional priority adjustment causes a change in the ranking within the candidate set, the adjusted coefficient values ​​are re-sorted in descending order, and the excitation mode corresponding to the minimum value is taken as the final result. The optimal excitation mode is recorded using three parameters: excitation type, corresponding three-state delay mean sequence, and consistency deviation coefficient. The excitation type parameter serves as the basis for excitation configuration during the optimal component decomposition window data detection phase, and the three-state delay mean sequence is used to correct the time base offset of the optimal component decomposition window data under each damage state.

[0058] The optimal component decomposition window data of the optimal excitation mode is detected. Positron injection and coincidence timing units are driven by the excitation type parameters of the optimal excitation mode. The injection mode is implemented according to the optimal type matched among three configurations: continuous beam, pulsed beam, or variable energy beam, ensuring that the statistical characteristics of the annihilation events acquired in this study are consistent with the cross-state robust response characteristics calibrated by the optimal excitation mode. Lifetime spectrum data is re-acquired according to the lower limit of the single-point cumulative event count, single-point acquisition duration, and component fitting accuracy requirements for each detection cycle in the detection configuration table. The re-acquired lifetime spectrum data is divided into periodic windows along the width sequence of the adaptive detection window. Wide window segments are completed with a normal dwell time, while narrow window segments maintain statistical accuracy by increasing the single-point dwell time or increasing the spatial sampling points, avoiding component separation failure due to insufficient event count during the defect type transition phase of the optimal component decomposition window data. Multi-component decomposition is performed on the lifetime spectrum within each window, outputting lifetime values ​​τ1, τ2, and τ3 and their corresponding intensity ratios I1, I2, and I3. The goodness of fit χ² of the component decomposition is calculated. 2The statistical errors of / ν and each component parameter are recorded synchronously as uncertainty attributes. The decomposition results are offset-corrected using the mean time delay sequence of the three states of the optimal excitation mode as the time base. The component parameters of each detection cycle in the optimal component decomposition window data are shifted to a unified reference system according to the mean time delay corresponding to their respective damage states. At the same time, the S-parameters and W-parameters of the Doppler broadening data are aligned with the same period to keep the component parameters and broadening parameters of each detection cycle in time synchronized. The corrected component parameters and broadening parameters are encapsulated into optimal component decomposition window data indexed by the detection cycle number. Each index position carries the lifetime value and intensity ratio of τ1, τ2, and τ3, the S-parameters, the W-parameters, and the statistical uncertainties of each item. The index records of all detection cycles cover the complete evolution process of the damage-sensitive time window and the stable defect type stage.

[0059] In some embodiments, the step of performing fluctuation priority sorting based on optimal component decomposition window data to form a detection execution sequence includes: determining the annihilation characteristic change amount based on the optimal component decomposition window data to generate a characteristic change sequence; identifying segments whose change amount is continuously lower than the background fluctuation based on the characteristic change sequence to generate abnormal silence markers; performing saturation capture stability scoring on each segment in the characteristic change sequence based on the abnormal silence markers to generate a priority scoring table; and performing score descending mapping through the priority scoring table to form a detection execution sequence.

[0060] The annihilation characteristic variation sequence is determined based on the optimal component decomposition window data. The optimal component decomposition window data acquired under the optimal excitation mode includes the component decomposition results of the positron annihilation lifetime spectrum data within each detection cycle. The gradual change in the intensity ratio of the τ2 and τ3 components with the detection cycle constitutes the main signal source of the annihilation characteristic variation. The change in the intensity I3 of the τ3 component between adjacent detection cycles, δI3(t) = I3(t+1) - I3(t) (distinguished from the spatial deviation ΔI3 in S130; this step uses lowercase δ to represent the time difference along the detection cycle direction), and the synchronous change in the S-parameter, δS(t), are taken after normalization of their respective sequence standard deviations. The arithmetic mean is combined to form the annihilation characteristic change for the detection period: δ_combined(t) = [δI3(t) / σ_δI3 + δS(t) / σ_δS] / 2, where σ_δI3 and σ_δS are the sample standard deviations of the δI3 and δS sequences, respectively, over the entire detection period. The selection of the normalization benchmark ensures that δ_combined(t) maintains a balanced contribution even when the magnitudes of the δI3 and δS parameters differ significantly. The combined merging of δI3 and δS allows the characteristic change sequence to simultaneously carry dual evolutionary information of defect concentration changes and defect type transitions. The annihilation characteristic changes for each detection period in the optimal component decomposition window data are arranged in the order of the detection periods to form a characteristic change sequence. The sequence length is consistent with the total number of detection periods covered by the optimal component decomposition window data, typically 20-60 elements in a full-lifetime detection scenario. Each element carries the combined change of δI3 and δS and its statistical uncertainty. The annihilation characteristic change of the detection period segment corresponding to the spatial region of the high-sensitivity damage layer is usually 2.5-4.0 times that of the low-sensitivity damage layer segment. The magnitude difference of each element in the characteristic change sequence shows a significant jump at the detection period of the damage layer boundary. In the characteristic change sequence, the detection period where the δI3 uncertainty exceeds 20% of the change itself is marked with a high uncertainty mark. The silence judgment threshold of the high uncertainty mark element is relaxed by 10% when identifying abnormal silent sections to prevent the underestimation of the change caused by measurement noise from triggering false silence marks.

[0061] An anomalous silence marker is generated by identifying segments where the change in characteristic variation sequence is consistently lower than the background fluctuation. The background fluctuation level of the characteristic variation sequence is determined by the lower quartile of the absolute value of δI3 throughout the sequence. The lower quartile corresponds to the normal small fluctuation level during the slow damage evolution stage. Multiplying this lower quartile by a coefficient of 0.6 defines the silence judgment threshold. When the absolute value of δI3 in three or more consecutive detection cycles in the characteristic variation sequence is lower than the silence judgment threshold, the segment is judged as an anomalous silence segment. The occurrence of anomalous silence segments in the characteristic variation sequence has two types of physical causes: one is that the damage evolution of the metal component in that detection cycle segment enters a brief plateau period, and the rate of increase in defect concentration slows down; the other is that the defect capture position reaches a saturation state, causing new defects to be unable to be effectively captured by positrons, thus artificially suppressing the characteristic variation. The plateau period corresponds to δS synchronously approaching zero, and saturation capture corresponds to δS showing a slight negative fluctuation while δI3 is close to zero. The distinction between the two types of characteristics is determined by the sign of the mean value of δS in the anomalous silence segment. The abnormal silence marker includes all segments that meet the condition of three consecutive points below the threshold. Each record carries two attributes: the start and end detection cycle number of the segment, and the sign of the mean δI3 and mean δS within the segment. When the detection cycle of a high uncertainty marker in the characteristic change sequence is covered by an abnormal silence segment, the mean δI3 of that segment is accompanied by an uncertainty correction description. The correction description records the proportion of high uncertainty elements within the segment. When the proportion exceeds 50%, the reliability rating of the abnormal silence marker is downgraded to low reliability. Low reliability markers are uniformly subject to an additional deduction of less than the benchmark score difference during the saturation capture stability scoring stage to reflect the downgrade in assessment credibility.

[0062] A priority scoring table is generated by scoring the saturation capture stability of each segment in the characteristic change sequence based on anomalous silence markers. The saturation capture stability score of each detection cycle segment in the characteristic change sequence is determined by superimposing a baseline score and three adjustment items. The baseline score is set in three levels according to the classification results of the anomalous silence markers: the highest baseline score for saturation capture type silence segments, the lowest baseline score for evolutionary plateau type silence segments, and the baseline score for non-silent segments is in between. The three levels of baseline scores are separated according to the above-mentioned high-low relationship, reflecting that saturation capture type segments have exceeded the effective identification range of the excitation mode and must be re-examined first to avoid missed detection. The saturation capture stability score is adjusted based on the baseline score for each segment using three methods: The first adjustment is based on the duration of the abnormal silence markers; a longer duration corresponds to a higher increment, reflecting the need for priority detection in long-term saturation states. The second adjustment is based on the mean δI3 of the segment immediately preceding the previous segment in the characteristic change sequence; a more significantly higher mean δI3 of the previous segment compared to the background fluctuation mean corresponds to a larger increment, reflecting the weighted contribution of the high activity level before the sudden drop in change to the current saturation state assessment. The third adjustment is based on the negative bias of the mean δS in the abnormal silence markers; a more significant negative bias of the mean δS corresponds to a larger increment. Negative δS bias is a core physical characteristic that distinguishes saturation capture from the evolutionary platform. After these three adjustments, the saturation capture stability scores for each segment are summarized into a priority scoring table. The priority scoring table uses the detection cycle segment number as an index and the adjusted total score as the score value. Each record in the table carries both the original baseline score and the sub-item values ​​of the three adjustment increments. Low-reliability marker segments are uniformly subject to additional deductions after the three adjustments, with the deduction amount being less than the baseline score difference.

[0063] A priority scoring table is used to perform a descending score mapping to form a detection execution sequence. Each detection cycle segment in the priority scoring table is arranged in descending order of score value. This descending order places the segment with the strongest saturation capture at the forefront of the execution sequence, ensuring that detection resources are prioritized for the detection cycle segments with the most urgent defect capture status. The descending order is primarily sorted by total score; when total scores are the same, the average δI3 value within the segment is used as a secondary sorting dimension, prioritizing segments with lower δI3 values ​​to reflect their deeper defect capture saturation. After descending sorting, adjacent segments are merged. Adjacent segments with a score difference less than 5% of the overall score range and consecutive detection cycle numbers are merged into one execution record. The score of the merged execution record is the weighted average of the two segment scores, with the weight determined by the number of detection cycles for each segment. This merging operation reduces frequent switching in the detection execution sequence caused by fine-grained segmentation, improving the continuous execution efficiency of the detection scheme. Low-reliability marked segments are not merged with non-low-reliability segments during the merging check, maintaining independent records so that their result reliability level can be separately marked during the execution phase of the detection sequence. In a full-lifetime detection scenario, the final number of entries in the detection execution sequence is usually 8-20. Each record carries three pieces of information: the corresponding detection period segment number, the merged score value, and the damage level attribution within the segment. The damage level attribution is determined by the corresponding level of the segment in the damage stratification configuration. Execution records with high-sensitivity damage level attribution are usually concentrated in the top 30% of the high-priority positions in the detection execution sequence.

[0064] The fatigue damage detection results are output through joint analysis of damage degree based on the detection configuration table and the detection execution sequence. The detection configuration table provides the lower limit of the cumulative number of events at a single point, the single-point acquisition time, and the component fitting accuracy requirements for each detection cycle. The detection execution sequence provides the priority score and damage level assignment for each detection cycle segment. The two types of inputs are matched cycle by cycle using the detection cycle number as the alignment key. Each detection cycle also carries acquisition condition constraints and execution priority weights. The joint analysis uses these dual attributes to drive the weight allocation of the life spectrum component parameters and the broadening feature parameters. The joint analysis of damage degree performs weighted fusion of the life spectrum component parameters and the broadening feature parameters with priority scores as weights. The weight coefficient for high-priority segments is 1.0-1.5, and for low-priority segments it is 0.6-0.8. The weighted fused parameters are mapped to three types of damage layers according to the damage level assignment. The three mapping results are summarized in all detection areas of the metal component to form a spatially continuous damage degree distribution map. Fatigue damage detection results are output in a structured report format. The field structure of each detection area record includes: detection area number and functional area category (high stress concentration area, weld heat-affected zone, or fatigue crack initiation prediction area), damage level classification, defect capture rate κ value, dominant defect type (determined based on the proportion of I2 and I3 strengths and S-parameters as single vacancy, vacancy cluster, or micropore), coordinate position of the damage front along the defect capture path, and annihilation lag time value. The distribution of κ values ​​for each detection area is attached to the report as a spatial heat map, and the position of the damage front is superimposed on the heat map with vector labels. The structured report and the heat map together constitute the complete output of fatigue damage detection results.

[0065] To implement the positron annihilation-based fatigue damage detection method for metal components corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, see [link to relevant documentation]. Figure 4 , Figure 4 The diagram shows a structural block diagram of a positron annihilation-based fatigue damage detection device 400 for metal components provided in an embodiment of this application, comprising: The signal acquisition module 401 is used to acquire positron annihilation lifetime spectrum data and Doppler broadening data of the detection area of ​​the metal component, and to perform defect capture rate correlation analysis based on the positron annihilation lifetime spectrum data and the Doppler broadening data to establish a defect evolution curve. The window correction module 402 is used to locate the inflection point of damage evolution based on the defect evolution curve to determine the damage-sensitive time window, obtain the defect identification momentum ratio parameter from the positron annihilation lifetime spectrum data and the Doppler broadening data to form an annihilation parameter correction factor, and use the annihilation parameter correction factor to perform window narrowing correction for the defect type transformation stage of the damage-sensitive time window to generate an adaptive detection window. Path analysis module 403 is used to adapt the detection scheme to the adaptive detection window to form a detection configuration table, identify the defect capture path dominated by low intensity and long life component according to the detection configuration table, calculate the annihilation rate gradient of the defect capture path to obtain the annihilation lag time, and define the damage level and generate damage layer configuration based on the annihilation lag time. The result output module 404 is used to perform cross-state hysteresis consistency evaluation on the damage stratification configuration to determine the optimal excitation mode, detect the optimal component decomposition window data of the optimal excitation mode, perform fluctuation priority sorting based on the optimal component decomposition window data to form a detection execution sequence, and perform joint analysis of damage degree based on the detection configuration table and the detection execution sequence to output fatigue damage detection results.

[0066] The aforementioned positron annihilation-based metal component fatigue damage detection device 400 can implement the positron annihilation-based metal component fatigue damage detection method described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0067] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. A fatigue damage detection method for metal components based on positron annihilation, characterized in that, include: Positron annihilation lifetime spectrum data and Doppler broadening data of the detection area of ​​the metal component are collected. Based on the correlation analysis of the defect capture rate of the positron annihilation lifetime spectrum data and the Doppler broadening data, a defect evolution curve is established. Based on the defect evolution curve, the damage evolution inflection point is located to determine the damage-sensitive time window. The defect identification momentum ratio parameter is obtained from the positron annihilation lifetime spectrum data and the Doppler broadening data to form an annihilation parameter correction factor. The annihilation parameter correction factor is used to perform window narrowing correction for the defect type transformation stage of the damage-sensitive time window to generate an adaptive detection window. The adaptive detection window is adapted to form a detection configuration table. Based on the detection configuration table, the defect capture path dominated by low-intensity long-life components is identified. The annihilation rate gradient of the defect capture path is calculated to obtain the annihilation lag time. Based on the annihilation lag time, the damage level is defined to generate a damage layer configuration. The optimal excitation mode is determined by cross-state hysteresis consistency evaluation of the damage stratification configuration. The optimal component decomposition window data of the optimal excitation mode is detected. The fluctuation priority sorting is performed based on the optimal component decomposition window data to form a detection execution sequence. The fatigue damage detection result is output by joint analysis of the damage degree based on the detection configuration table and the detection execution sequence.

2. The method according to claim 1, characterized in that, The step of determining the damage-sensitive time window by locating the inflection point of damage evolution based on the defect evolution curve includes: The annihilation rate time series of the defect evolution curve is obtained according to the detection cycle to generate an annihilation rate evolution sequence; The inflection point set of the second-order difference recognition curvature inversion node generation rate is calculated based on the annihilation rate evolution sequence. Based on the rate inversion inflection inflection point set, the defect concentration change trend is inversely calculated to generate a damage acceleration interval; The damage-sensitive time window is determined by boundary mapping using the damage acceleration interval.

3. The method according to claim 1, characterized in that, The step of obtaining the defect discrimination momentum ratio parameter from the positron annihilation lifetime spectrum data and the Doppler broadening data to form the annihilation parameter correction factor includes: Momentum spectral decomposition is performed on the positron annihilation lifetime spectrum data and the Doppler broadening data to generate low momentum distribution and high momentum distribution; Based on the ratio deviation between the low momentum distribution and the high momentum distribution, a momentum ratio sequence is generated. Based on the momentum ratio sequence, identify the node where the slope of the ratio sequence changes from positive to negative and generate a transition node marker; An annihilation parameter correction factor is formed by performing directional weighted correction of the transition node based on the transition node marker and the momentum ratio sequence.

4. The method according to claim 1, characterized in that, The step of identifying the defect capture path dominated by the low-intensity, long-life component according to the detection configuration table includes: Extract the lifetime spectrum component intensity of each detection region from the detection configuration table to generate a component intensity distribution; A component intensity deviation sequence is generated by comparing the long-life component intensity through the component intensity distribution; Based on the component intensity deviation sequence, segments that are expected to be dominant but actually weakened are selected to generate a set of suppressed weak components. Traps competition effect analysis is performed on the suppressed weak component set to locate the defect capture path.

5. The method according to claim 1, characterized in that, The step of calculating the annihilation rate gradient of the defect capture path to obtain the annihilation lag time includes: Based on the defect capture path, the annihilation rate of each region is calculated to generate the spatial distribution of the annihilation rate. The annihilation rate gradient sequence is generated by performing non-uniform spacing weighted difference calculation on the spatial distribution of the annihilation rate. Based on the annihilation rate gradient sequence, damage front markers are generated for segments that maintain a high gradient level after a sudden change. Hysteresis hysteresis time is obtained by performing hysteresis analysis based on the damage front marker and the defect capture path.

6. The method according to claim 1, characterized in that, The step of determining the optimal excitation mode by performing cross-state hysteresis consistency evaluation on the damage stratification configuration includes: The hysteresis time of each excitation mode under the initial state, service state and failure state is determined for the damage stratification configuration to generate a cross-state time delay sequence group; A consistency deviation coefficient is generated by calculating the consistency difference of the cross-state time delay sequence group. Based on the consistency deviation coefficient, a robust excitation candidate set is generated by identifying excitation modes whose deviations monotonically converge as the damage deepens. The optimal excitation mode is determined by using the robust excitation candidate set for optimal excitation matching.

7. The method according to claim 1, characterized in that, The step of performing fluctuation priority sorting based on the optimal component decomposition window data to form a detection execution sequence includes: Based on the optimal component decomposition window data, the annihilation characteristic change amount is determined to generate a characteristic change sequence; Based on the characteristic change sequence, segments where the change amount is consistently lower than the background fluctuation are identified, and abnormal silent markers are generated. Based on the abnormal silent markers, a priority scoring table is generated by performing saturation capture stability scoring on each segment of the characteristic change sequence. The detection execution sequence is formed by performing a descending score mapping using the priority scoring table.

8. The method according to claim 3, characterized in that, The step of generating a momentum ratio sequence based on the ratio deviation between the low momentum distribution and the high momentum distribution includes: The low momentum distribution and the high momentum distribution are corrected for 511 keV annihilation peak asymmetry to generate a normalized momentum set; The original ratio sequence is generated by calculating the ratio of the integrated area of ​​the low momentum region to the total area at each detection point using the standardized momentum set. The original ratio sequence is used to identify abnormal jump points and generate jump point masks; A momentum ratio sequence is generated by performing a transition-preserving smoothing filter based on the transition point mask and the original ratio sequence.

9. The method according to claim 6, characterized in that, The step of calculating the consistency deviation coefficient by means of the cross-state delay sequence group includes: For the cross-state delay sequence group, calculate the mean lag time of each state to generate a state mean group; Cross-state lag slope trend analysis is performed using the state mean group to generate a monotonic distribution of slope between states. Based on the slope monotonicity distribution between states, non-monotonic excitation modes are identified and exclusion markers are generated. Based on the exclusion marker and the cross-state delay sequence group, a consistency deviation coefficient is generated through screening.

10. A fatigue damage detection device for metal components based on positron annihilation, characterized in that, include: The signal acquisition module is used to acquire positron annihilation lifetime spectrum data and Doppler broadening data of the detection area of ​​the metal component, and to establish a defect evolution curve based on the correlation analysis of the defect capture rate of the positron annihilation lifetime spectrum data and the Doppler broadening data. The window correction module is used to locate the inflection point of damage evolution based on the defect evolution curve to determine the damage-sensitive time window, obtain the defect identification momentum ratio parameter from the positron annihilation lifetime spectrum data and the Doppler broadening data to form an annihilation parameter correction factor, and use the annihilation parameter correction factor to perform window narrowing correction for the defect type transformation stage of the damage-sensitive time window to generate an adaptive detection window. The path analysis module is used to adapt the detection scheme to the adaptive detection window to form a detection configuration table, identify the defect capture path dominated by low-intensity long-life components according to the detection configuration table, calculate the annihilation rate gradient of the defect capture path to obtain the annihilation lag time, and define the damage level and generate a damage layer configuration based on the annihilation lag time. The result output module is used to perform cross-state hysteresis consistency evaluation on the damage stratification configuration to determine the optimal excitation mode, detect the optimal component decomposition window data of the optimal excitation mode, perform fluctuation priority sorting based on the optimal component decomposition window data to form a detection execution sequence, and perform joint analysis of damage degree based on the detection configuration table and the detection execution sequence to output fatigue damage detection results.

Images