Steel corrosion layer depth inversion method based on LIBS spectral correlation analysis

Abstract

The invention relates to the technical field of corrosion layer inversion, in particular to a steel corrosion layer depth inversion method based on LIBS spectral correlation analysis, which comprises the following steps: S1, acquiring a spectral sequence from a covering corrosion layer to a metal matrix, performing layer-by-layer variation rate analysis on each wave band spectral line, and constructing a spectral depth variation path set changing along with the denudation process; s2, identifying positions with sudden change or stability turning of spectral line features, marking the corresponding denudation step number as a transition position between corrosion layers, and constructing a transition index set comprising the transition position, the mutation rate sudden enhancement degree and the spectral line group linkage degree dimension according to the transition position, the mutation rate sudden enhancement degree and the spectral line group linkage degree dimension; and S3, inputting the transition index set into a trained structure mapping model, and outputting the total corrosion layer depth, the thickness estimation value of each sub-layer and the confidence level of the target corrosion area in combination with the multi-type corrosion sample library. The method improves the detection resolution and process adaptability, and can be widely applied to industrial scenes sensitive to the corrosion layered structure.

Description

Technical Field

[0001] This invention relates to the field of corrosion layer inversion technology, and in particular to a method for inverting the depth of steel corrosion layers based on LIBS spectral correlation analysis. Background Technology

[0002] During the service life of steel, the surface often forms a multi-layered and complex corrosion layer structure due to moisture, oxidation, chemical corrosion, etc. Especially in marine, industrial high humidity or thermal stress environments, the corrosion layer often exhibits composite characteristics of multi-layer stacking, blurred transitions, and uneven morphology. Traditional corrosion layer depth assessment methods, such as metallographic sections, electrochemical impedance spectroscopy, or X-ray detection, can provide a certain degree of accuracy, but most of them have problems such as strong operational destructiveness, time-consuming process, difficulty in layer identification, or inability to be applied online, which limit their widespread deployment in scenarios such as structural residual life assessment, manufacturing quality monitoring, and maintenance intervention.

[0003] In recent years, laser-induced breakdown spectroscopy has been gradually introduced into the fields of material characterization and corrosion detection due to its non-destructive, rapid, and multi-element response capabilities. However, most existing LIBS technologies are limited to qualitative judgment based on a single spectrum or overall corrosion level assessment. They lack modeling and utilization of spectral line evolution behavior along the depth of the corrosion layer and cannot effectively identify complex layered interfaces. In particular, when faced with structural information such as nonlinear abrupt changes in spectral response and elemental linkage changes, there is a lack of quantifiable extraction and in-depth reasoning mechanisms. Summary of the Invention

[0004] This invention provides a method for inverting the depth of steel corrosion layers based on LIBS spectral correlation analysis. It integrates multi-layer spectral variation analysis, structural feature identification and intelligent mapping model to achieve non-contact, hierarchical and quantifiable depth inversion of composite corrosion layers of steel.

[0005] A method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis includes the following steps: S1: Perform multiple laser ablation operations sequentially in the target corrosion area to obtain the spectral sequence covering the corrosion layer to the metal substrate; perform layer-by-layer variation rate analysis on the spectral lines of each band to construct a set of spectral depth variation paths that change with the ablation process; S2: Based on the set of spectral depth variation paths, identify the locations where there are abrupt changes or instability transitions in spectral features, mark the corresponding erosion steps as the transition locations between corrosion layers, and construct a set of transition indexes including transition locations, abrupt increase in variation rate, and degree of linkage of spectral line groups. S3: Input the transition index set into the trained structure mapping model, combine it with a multi-type corrosion sample library, and output the total corrosion layer depth, estimated thickness of each sublayer, and confidence level of the target corrosion area.

[0006] Optionally, S1 specifically includes: S11: At the same location point in the target corrosion area, perform multiple consecutive laser ablation cycles with fixed laser energy and focusing conditions; S12: Arrange the spectra collected in multiple cycles according to the erosion order to form a spectral-depth sequence data from the surface of the corrosion layer to the metal substrate; S13: For at least one preset characteristic band spectral line in the spectral-depth sequence data, calculate the variation rate between adjacent erosion layers; S14: Summarize the variation trajectory of the variation rate of all preset characteristic band spectral lines with the number of erosion cycles into a set of spectral depth variation paths.

[0007] Optionally, the laser ablation cycle includes one laser pulse ablation and one spectral acquisition, until the spectral characteristics acquired in real time indicate that the ablation has reached the uncorroded metal substrate.

[0008] Optionally, the variation rate is characterized by changes in spectral line intensity, profile, or correlation with adjacent spectral lines.

[0009] Optionally, S2 specifically includes: S21: Perform inflection point detection on each variation path in the set of spectral depth variation paths, and initially mark the detected peak points of sudden increase in variation rate or the points where the variation trend changes as candidate transition points. S22: Based on the candidate transition points of all preset characteristic band spectral lines, the common erosion step interval is determined by density clustering analysis, and the erosion step number corresponding to the center of each interval is confirmed as the inter-corrosion layer transition position. S23: For each confirmed transition position, construct multi-dimensional transition indicators; S24: Arrange the multidimensional transition indices of all transition positions in the order of erosion to form a set of transition indices characterizing the layering characteristics of the corrosion layer structure.

[0010] Optionally, the multi-dimensional transition indicators specifically include: The transition position dimension is the erosion cycle number corresponding to the transition position; The mutation rate abrupt increase magnitude dimension is the average abrupt increase magnitude of the mutation rate of all characteristic band spectral lines at the said transition position; The degree of linkage of spectral lines is the synchronous correlation coefficient of the variation trend of spectral lines in different characteristic bands at the transition position.

[0011] Optionally, the density clustering analysis is based on a preset neighborhood search radius and minimum number of points to perform high-density regional clustering of transition points and identify several regions with concentrated erosion steps; the center position of the erosion steps in each cluster region is taken as the transition position between corrosion layers.

[0012] Optionally, S3 specifically includes: S31: Construct and train a structure mapping model, which is a supervised machine learning model trained on a multi-type corrosion sample library; S32: Using the transition index set as the input features of the structure mapping model, and the corresponding true values ​​of corrosion depth and layer thickness as the training targets, the structure mapping model is trained and validated. S33: Input the set of transition indices of the target corrosion region to be tested into the trained structure mapping model, and directly output the estimated total corrosion layer depth and the estimated thickness of each sublayer of the target corrosion region. S34: Based on the structure mapping model, when outputting the depth estimate, the confidence level through the uncertainty measure is given simultaneously.

[0013] Optionally, each sample in the sample library includes the true values ​​of the total corrosion layer depth and layer thickness calibrated by metallography, as well as the extracted set of transition indices.

[0014] Optionally, S32 specifically includes: S321: The transition index set of each corrosion sample is used as the input feature of the structure mapping model. The transition index set includes the position of each transition point, the intensity of the mutation rate abrupt increase, and the degree of linkage of spectral line groups. S322: Using the total corrosion depth and the thickness of each sublayer of the corresponding sample obtained by metallographic calibration as the training target, the structure mapping model is trained and validated through supervised learning, so that the model learns the mapping relationship between transition index features and the layered structure of the corrosion layer.

[0015] The beneficial effects of this invention are: This invention acquires multi-layer LIBS spectral sequences covering the corrosion layer to the metal substrate under fixed laser ablation conditions. By combining layer-by-layer spectral variation rate analysis and inflection point detection, it extracts physically meaningful transition feature positions and multi-dimensional index expressions. Compared with traditional methods based on single-layer spectral features or surface response, this approach achieves layer-by-layer perception and parametric modeling of the multi-layer structure inside the corrosion layer. It can estimate the corrosion depth layer by layer without destructive cutting, improving detection resolution and process adaptability. It is especially suitable for scenarios with complex composite corrosion morphology and alternating deformation and oxide layers.

[0016] This invention constructs a structural mapping model based on the alignment of transition indices and metallographic labels. It employs a deep learning network with uncertainty modeling capabilities, which can not only output the predicted results of the total thickness of the corrosion layer and the thickness of each sublayer, but also simultaneously provide uncertainty evaluations for each output item. This forms a reliable corrosion structure identification mechanism with predicted values ​​and confidence levels, improving the reliability and controllability of the intelligent model in engineering applications. It can be widely applied to industrial scenarios sensitive to corrosion delamination structures, such as weld failure prediction and high-precision component life assessment. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the logical framework of an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0020] like Figures 1-2 As shown, a method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis includes the following steps: S1: Perform multiple laser ablation operations sequentially in the target corrosion area to obtain the spectral sequence covering the corrosion layer to the metal substrate; perform layer-by-layer variation rate analysis on the spectral lines of each band to construct a set of spectral depth variation paths that change with the ablation process; S1 specifically includes: S11, Laser ablation and spectral acquisition cycle: A fixed laser energy is set at the same spatial location point in the target corrosion area. The laser ablation cycle is executed multiple times, with the laser spot focusing conditions including focal length f and beam diameter d; each cycle... include: Single laser pulse ablation; The spectral signal of laser-induced plasma emission is acquired in real time to obtain the spectrum. .

[0021] The termination condition is: when the spectrum acquired in a certain cycle... The intensity of the preset matrix element spectral lines exceeds the intensity threshold. This means that the erosion has reached the uncorroded metal substrate.

[0022] in, _i_laser_pulse_energy_, ranging from 20-100 mJ / pulse, is sufficient to induce LIBS plasma while avoiding excessive ablation; this is a typical LIBS energy range for metal ablation. _f_laser_focusing_focal_length_, ranging from 50-200 mm, determines the spot size and focusing depth, and depends on the optical system configuration. _d_laser_beam_diameter_, ranging from 50-300 µm, affects the ablation area and spatial resolution; smaller spots are beneficial for depth stratification. _N_laser_ablation_cycles_, ranging from 10-100, depends on the thickness of the etched layer and must be sufficient to penetrate the entire etched layer to the metal substrate. _i_current_ablation_cycle_number_. The matrix strength threshold is set at 30%-80% to determine whether the matrix elements are sufficiently exposed, at which point the data collection is terminated.

[0023] This step acquires spectral evolution information along the depth direction, penetrating the entire corrosion layer to the substrate, without moving the detection position. This process relies on laser-induced breakdown spectroscopy, where each laser pulse ablates a certain thickness of material and excites plasma on the newly exposed surface, thereby capturing its characteristic spectrum.

[0024] The fixed laser parameters in this step ensure consistent ablation depth and comparable spectral responses for each operation, avoiding spectral drift caused by uneven ablation. Simultaneously, immediate spectral acquisition after each laser irradiation allows for complete recording of material compositional information as it changes with depth. This is particularly helpful for identifying different corrosion products or interface transition zones, especially for structures with distinct layers and complex compositions like corrosion layers. The termination condition is set when a clear metallic matrix feature appears in the spectrum and its intensity reaches a certain threshold, indicating that ablation has reached the underlying layer. This design aims to avoid excessive ablation of the substrate while ensuring complete data coverage of the corrosion layer area, providing a reliable data foundation for subsequent spectral variation analysis and deep structure identification.

[0025] S12, Spectral-Depth Sequence Construction: Arrange the spectral data obtained from N iterations in the ablation order to form a spectral-depth sequence, represented as: ; in, It is the wavelength after the i-th layer is etched. The spectral line intensity distribution below, where i is the index of the corresponding ablation depth, and the ablation depth is... , The depth of a single ablation pass ranges from 0.1 to 5 µm, depending on the material type, laser power, and focusing conditions. It reflects the depth resolution of the spectrum. The wavelength is 200-900nm.

[0026] This step organizes the acquired laser ablation spectral data from multiple ablation operations into a depth-meaning spectral sequence according to the ablation order, constructing a spectral-depth sequence. Although LIBS technology only acquires the spectrum after each single-layer ablation, if each ablation is accurately considered as a layer corresponding to a certain depth, a spectral-depth mapping relationship reflecting the changes in the material's internal structure can be obtained as a whole. This data structure is extremely important for analyzing the component distribution of materials along the depth direction, identifying the transition interface between the corrosion layer and the matrix, and extracting the evolution trajectory of characteristic spectral lines. Compared to single-spectrum analysis, it is more hierarchical and trend-oriented, serving as the fundamental data form for subsequent layer identification and model inversion.

[0027] S13, Calculation of characteristic band variation rate: [For...] Preset feature band set in all spectral frames Calculate the variation rate between adjacent eroded layers. Specifically, it includes: S131, Intensity Variation Rate: ; S132, spectral profile variation rate: ; S133, Variation rate of correlation between spectral lines: ; in, To pre-define a set of characteristic bands, covering the main components of the corrosion layer and the matrix spectral lines, in order to analyze the layered structure, For the i-th layer at wavelength The spectral line intensity, ranging from 0 to 65535, characterizes the spectral line response at different ablation depths and is used to calculate the variability rate. For the i-th layer at wavelength The full width at half maximum (FWHM) of a spectral line, ranging from 0.1 to 5 nm, characterizes changes in the spectral line profile and can be used to determine the depth of laser-material interaction and changes in physical state. For the band in the i-th layer spectrum and The Pearson correlation coefficient, with a value range of 100. It reflects the interconnected changes between spectral lines of different elements, which helps in analyzing the coordinated migration or distribution of corrosive substances. This represents the normalized variability of spectral line intensity, ranging from 0 to 2. It reflects the degree of abrupt change in the intensity of spectral lines between adjacent layers; a larger change indicates a more significant difference in interlayer structure. The depth of a single laser ablation is expressed in units of 1. The value range is 0.1-5. , The correlation variation rate between spectral lines, ranging from 0 to 2, characterizes the change in the coupling relationship between elements in adjacent layers. Abrupt changes often indicate the interface between material layers.

[0028] This step quantitatively compares multi-layer data in the spectral sequence, extracts the response variation characteristics of each spectral line at different ablation depths, introduces the concept of variation rate, and numerically models the variation trend of the spectral line response of each preset characteristic band between adjacent ablation layers, considering three types of variation: Intensity variability: By comparing the intensity changes of a spectral line in a certain band between two adjacent layers, it is possible to identify whether the spectral line shows a sudden increase or decrease. This change usually reflects a dramatic shift in composition.

[0029] Spectral line profile variation rate: This mainly involves calculating the shape characteristics of spectral lines, such as their full width at half maximum (FWHM), to analyze changes in plasma state or emission environment, indirectly reflecting differences in material density and phase structure. This is of great significance for identifying corrosion layers with abrupt changes in physical state.

[0030] Inter-spectral correlation variation rate: By calculating the changes in correlation between different spectral lines in a certain layer, this method captures whether a transition occurs in the co-occurrence relationship between different elements. It can reveal whether certain components exist in a cooperative form and is often used to analyze complex structures where multiple corrosion products are mixed and intertwined.

[0031] In summary, this approach not only focuses on the intensity of spectral lines themselves but also considers the interactions and evolutionary patterns between them. This allows the analysis to go beyond single-point mutations and systematically reveal the complexity and hierarchy of the entire corrosion layer structure. Furthermore, these variation rate indices provide a trajectory data foundation for subsequently constructing spectral depth variation paths, contributing to the formation of highly regular and interpretable intermediate expressions.

[0032] S14, Spectral Depth Variation Path Construction: For each band The variation rate sequence formed during the erosion process is used to form a spectral depth variation path set, which describes the nonlinear trajectory of the spectral line response as a function of depth. This serves as the basis for subsequent transition detection and structural feature extraction, and is expressed as: ; ; in, For band The depth variation path is recorded, and the variation rate sequence of this band during the erosion process is recorded to identify transition points. This is a set of spectral depth variation paths, summarizing the variation paths across all spectral bands.

[0033] S2: Based on the set of spectral depth variation paths, identify the locations where there are abrupt changes or instability transitions in spectral features, mark the corresponding ablation steps as the transition locations between corrosion layers, and construct a set of transition indexes including transition locations, abrupt increase in variation rate, and degree of linkage of spectral line groups. S2 specifically includes: S21, Inflection Point Detection and Candidate Transition Point Extraction: Analyzing the Set of Spectral Depth Variation Paths For each mutation path in the data, perform inflection point detection; for any band... mutation path Calculate its first-order and second-order difference sequences, expressed as: ; ; If any of the following conditions are met, then point i is marked as a candidate transition point. : : Sudden peak value; Significant turning point.

[0034] in, Is the i-th layer in band The intensity variation rate, Let be the first-order difference of the i-th layer, i.e., the change in the variability rate, representing the rate of change of the spectral line. The second-order difference of the i-th layer, i.e., the curvature, is used to detect inflection points in the trend of spectral line changes, which helps to identify stability transitions. This is the threshold for the intensity of spectral line abrupt changes, ranging from 0.2 to 0.5. Values ​​exceeding this indicate significant transitions in spectral line intensity, often associated with corrosion layer boundaries. The curvature abrupt change threshold is set between 0.1 and 0.3. Curvature abrupt changes represent trend inflection points and help identify layered interfaces.

[0035] This step identifies potential spectral abrupt change locations at the corrosion layer interface from the spectral depth variation path, serving as candidate points for subsequent transition location confirmation.

[0036] For each characteristic band spectral line, the variation path is extracted, and its variation rate trend along the ablation depth direction is obtained. By analyzing the spectral response changes between adjacent ablation layers, locations of sudden increases or trend reversals are identified. Sudden increases are typically manifested as abrupt changes in spectral line intensity or morphology during continuous ablation, while trend reversals are manifested as significant changes in the direction or rate of spectral line change.

[0037] To enhance robustness, this step performs differential analysis on the variation trend of each spectral line, identifying two types of feature points: local abrupt changes and curvature changes. Erosion locations that meet the preset abrupt change threshold or transition sensitivity threshold are preliminarily marked as candidate transition points.

[0038] Ultimately, the candidate transition points for all bands will serve as the basis for subsequent cluster analysis to identify the collective transition behavior of spectral lines in the depth direction, thereby inferring the location of the structural interface of the corrosion layer.

[0039] S22, Cluster analysis of transition intervals and confirmation of transition locations: The set of candidate transition points for all bands is summarized as follows: ; The density-based clustering method DBSCAN is used to cluster the erosion step positions in set C, identifying several clustering intervals, denoted as: ; Take each cluster interval The center point is used as the confirmed jump position, denoted as: ; Where C is the set of candidate transition points for all bands, and G is the set of candidate transition intervals after clustering. Clustering merges candidate transition points from multiple spectral lines to reduce redundancy and accidental misjudgments. This is the l-th candidate transition interval. To confirm the interlayer transition locations, L is the final number of confirmed transition locations, ranging from 1 to 5. The corrosion layer typically consists of 2 to 5 layers. For band The Middle One candidate jump point location.

[0040] This step categorizes and integrates the candidate transition points initially identified from all characteristic spectral lines, identifies their clustering trends along the erosion depth direction, and confirms the actual interlayer transition locations based on this. Candidate transition points extracted from all preset characteristic band spectral lines are collected; these transition points are typically represented by erosion cycle numbers or corresponding depth positions. Since abrupt changes in multiple spectral lines on the same erosion layer interface may not completely overlap but occur within a few adjacent erosion steps, density analysis of these locations is required using a clustering algorithm. A density clustering method is used to cluster all candidate transition points along the erosion step dimension. This method identifies high-density clustering intervals of transition points by setting appropriate erosion step neighborhood distance thresholds and minimum point count thresholds, thereby effectively eliminating interference from isolated transition points or occasional disturbances.

[0041] For each identified transition interval, the average number of erosion steps within it is extracted as the representative transition position of that cluster region. This central position is considered the key depth at which significant transitions occur collectively in the spectral lines, possessing high structural layering confidence. The final set of transition positions, arranged in the erosion sequence, serves as the interface locations between eroded layers.

[0042] S23, Construction of Multidimensional Transition Indicators: For each confirmed transition position The abrupt change characteristics of each characteristic spectral line at the corresponding erosion layer are extracted, and the dimensions of transition position, mutation rate abrupt increase intensity, and spectral line group linkage are constructed to finally form a transition index vector, which specifically includes: S231, Transition Position Dimension: Directly Use Position Index express; The transition position indicates the number of laser ablation cycles in the ablation sequence during which the variation behavior of multiple spectral lines exhibits a concentrated abrupt change or trend reversal. Since each ablation represents a stable and controllable depth increment, each ablation sequence number can be regarded as an equally spaced depth sampling point.

[0043] In this case, use the erosion cycle number. Using it as a location index for the transition layer is reasonable and direct. It includes spatial information such as depth, which can effectively represent the relative position of the transition interface in the entire erosion layer structure, and is also convenient for subsequent sequential modeling in the structure mapping model. In addition, compared with using absolute depth values, the location index is more universal and adaptable to different samples.

[0044] S232, mutation rate spurious increase in strength dimension, expressed as the average: ; The mutation rate burst strength dimension is used to measure the mutation rate at a certain transition position. The intensity index of whether multiple spectral lines in the vicinity show a common and drastic change is suitable for reflecting the overall degree of change of multiple spectral lines at this position, and suppressing the interference of individual abnormal spectral lines on the overall judgment. S233, the dimension of spectral line group linkage, represented by the correlation coefficient: ; in, The abrupt increase intensity at the l-th transition position, The degree of linkage between spectral lines at the l-th transition point reflects the synchronicity of abrupt changes between spectral lines. Represents the set of characteristic bands. Indicates band With band In position Correlation of variation rate at location, Let Pearson correlation coefficient or covariance function be used. A value close to 1 indicates that most spectral lines undergo a strong abrupt change at that position, suggesting a clear physical boundary; if A value approaching 0 indicates disordered or inconsistent spectral line changes, suggesting it may be merely an occasional perturbation, thus its reliability is low; in summary, It is a mathematical quantification of transition synchronization, which can effectively improve the robustness and reliability of transition judgment, and is especially suitable for complex corrosion samples with many noise disturbances and asynchronous spectral responses. S234, ultimately forming the transition index vector, is expressed as: ; S24, Construction of the transition indicator set: Constructing all transition indicators Arranged in the order of erosion, a complete set of transition indices is formed, represented as: ; This set of indicators serves as an intermediate representation of the layered structure of the corrosion layer. It is then input into the subsequent structure mapping model to support the inference and inversion of the thickness and layer of the corrosion layer.

[0045] S3: Input the transition index set into the trained structure mapping model, combine it with a multi-type corrosion sample library, and output the total corrosion layer depth, estimated thickness of each sublayer, and confidence level of the target corrosion area. S3 specifically includes: S31, Constructing and Training the Structure Mapping Model: Constructing the structure mapping model Its essence is a supervised machine learning model used to learn the mapping relationship from transition indicators to the depth structure of the erosion layer.

[0046] The training samples come from a multi-type corrosion sample library. , Where N is the total number of samples, and its value ranges from 1 to 2. To ensure the generalization ability and robustness of the structure mapping model, It is the set of transition indices for the nth sample. , Let l be the transition index vector. , These are the true values ​​of the total corrosion layer depth and the thickness of each sublayer as determined by metallographic methods. The metallographic method specifically includes the following steps: (1) Sample preparation: Obtain different types of steel corrosion samples from actual corrosion environments, and use a cutting tool to vertically cut the target detection area to obtain a complete sample cross section including the corrosion layer and the metal substrate.

[0047] (2) Mounting and grinding: The sliced ​​sample is cold-mounted with epoxy resin. After curing, it is then subjected to multi-stage precision grinding to ensure that the sample cross-section surface is flat and smooth, and the corrosion layer structure can be clearly displayed.

[0048] (3) Etching and development: Select a suitable chemical etching solution according to the type of steel and perform etching treatment on the surface of the polished sample to reveal the differences in structure between different corrosion product layers, oxide zones, interface transition zones and metal substrates.

[0049] (4) Optical microscopic observation: The developed sample is placed under a metallographic microscope for observation. Different magnifications are used to image the corrosion layer structure and identify the layer boundaries and thickness changes of each sublayer. Polarizing and dark field enhancement methods can be used to help distinguish tissue details.

[0050] (5) Image analysis and layer thickness measurement: The captured images are digitized and the corrosion layer is labeled using image processing software. Based on the gray scale changes, interface clarity or tissue morphology differences, the boundaries of each sub-layer are segmented and the thickness of each layer is measured. Finally, the true structural data including the total corrosion layer thickness and the thickness of each sub-layer are obtained.

[0051] (6) Labeling results output: The above measurement results are matched one-to-one with the corresponding LIBS spectral test positions and used as the standard output labels for training samples to ensure that the subsequent machine learning model can learn the effective mapping relationship between spectral features and actual hierarchical structure.

[0052] S32, Model Training and Validation: As input features, As a supervisory label, for the model Supervised training is represented as: ; in, For the prediction function of the structure mapping model, The loss function is structure-weighted sublayer error, which serves as the objective function for training the structure mapping model. It consists of two parts: a total depth error term and a hierarchical structure error term. The former constrains the predicted total thickness to be close to the true value, while the latter penalizes situations where the sublayer thickness does not match the proportion of the true structure. A sublayer error weighting mechanism is introduced, assigning weights to the thickness prediction error of each layer based on its proportion in the total depth. This strengthens the supervision of sublayers that dominate structural features while suppressing error instability caused by overfitting extremely thin sublayers. During model training, the loss function not only penalizes the numerical deviation between the predicted and true values ​​but also emphasizes the relative consistency between the predicted structure and the true erosion layer structure. This gives the model strong hierarchical structure recovery capabilities, making it suitable for diverse complex erosion scenarios. The model parameter set is designed to enable the model to capture the nonlinear mapping relationship between transition indices and the actual corrosion hierarchy.

[0053] S33, Target Region Inversion: For the target region to be tested, its transition index set is obtained through S1 and S2. This is input into the trained model and represented as: ; in, It is a set of transition indicators for the target region. To estimate the total corrosion layer depth, For the estimated first The thickness of the corroded sublayer must meet the requirements. The model output is the structural layering estimation result of the corroded area, realizing non-contact depth inversion.

[0054] S34, Confidence Rating: The structural mapping model employs a machine learning architecture with uncertainty modeling capabilities, enabling a Monte Carlo dropout deep regression network to assess the confidence level of the output predictions. Specifically, this includes: S341, Model Structure Adjustment: A Dropout layer is introduced into the traditional deep network, and the Dropout layer remains active during the testing phase. A set of prediction distributions is obtained through multiple random forward propagations. The target transition index set is then predicted M times to obtain multiple output results, as shown below: ; Where M represents the number of Monte Carlo samplings, the number of forward predictions performed with Dropout activation for each input, ranging from 30 to 50. Smaller sampling numbers cannot stably estimate the distribution, while larger numbers will increase inference time. 30 is a commonly used equilibrium point. This is the estimated total corrosion layer depth output from the m-th sampling, representing the result of a single Dropout inference. Its value ranges from 10 to 300 μm, and it characterizes the model's prediction of the overall corrosion thickness for that sampling. The m-th sample output The estimated sublayer thickness is the result of a single Dropout inference, and the value ranges from 5 to 100 μm.

[0055] S342, Uncertainty Quantification: Calculate the mean and standard deviation of each predictor as the final output value and uncertainty estimate, including: (1) Total corrosion layer depth: ; (2) Sublayer thickness: ; in, This is the predicted value for the total corrosion layer depth. The standard deviation of the total depth prediction is used to quantify the uncertainty of the model's overall thickness estimate. This represents the predicted standard deviation of the thickness of each layer, used to assess the reliability of the predicted thickness of that sublayer. For the first Predicted values ​​for sublayer thickness.

[0056] S343, Confidence Level Classification: Based on Uncertainty The magnitude of the value determines the prediction results into three confidence levels, including: (1) The confidence level is high, the model prediction is stable, and there are similar samples in the training set; (2) At the confidence level, the model predicts moderate volatility and some structural differences. (3) The confidence level is low, the model prediction fluctuates greatly, and there may be unknown structures or areas with scarce samples. in, This is the upper limit of the confidence level, fixed at 5μm. Samples with extremely small prediction fluctuations are classified as high confidence levels, suitable for direct use or automatic judgment. This is the lower limit of confidence level, with a value range of 15-20μm, used to distinguish highly variable samples. Please note that manual verification or additional comments are required. S344, Output Structure: The final structure mapping model output is represented as follows: ; in, This represents the predicted value for the depth and structure of the corrosion layer. As a quantification of uncertainty, The set of standard deviations of all sublayer thicknesses. For confidence level labels, according to uncertainty The range is divided into high, medium, and low; the result not only provides a predicted value, but also simultaneously quantifies its reliability, which facilitates users to make decisions based on the confidence level in subsequent process evaluation, corrosion treatment or quality classification.

[0057] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0058] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis, characterized in that, Includes the following steps: S1: Perform multiple laser ablation operations sequentially in the target corrosion area to obtain the spectral sequence covering the corrosion layer to the metal substrate; perform layer-by-layer variation rate analysis on the spectral lines of each band to construct a set of spectral depth variation paths that change with the ablation process; S2: Based on the set of spectral depth variation paths, identify the locations where there are abrupt changes or instability transitions in spectral features, mark the corresponding erosion steps as the transition locations between corrosion layers, and construct a set of transition indexes including transition locations, abrupt increase in variation rate, and degree of linkage of spectral line groups. S3: Input the transition index set into the trained structure mapping model, combine it with a multi-type corrosion sample library, and output the total corrosion layer depth, estimated thickness of each sublayer, and confidence level of the target corrosion area.

2. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 1, characterized in that, S1 specifically includes: S11: At the same location point in the target corrosion area, perform multiple consecutive laser ablation cycles with fixed laser energy and focusing conditions; S12: Arrange the spectra collected in multiple cycles according to the erosion order to form a spectral-depth sequence data from the surface of the corrosion layer to the metal substrate; S13: For at least one preset characteristic band spectral line in the spectral-depth sequence data, calculate the variation rate between adjacent erosion layers; S14: Summarize the variation trajectory of the variation rate of all preset characteristic band spectral lines with the number of erosion cycles into a set of spectral depth variation paths.

3. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 2, characterized in that, The laser ablation cycle includes one laser pulse ablation and one spectral acquisition, until the spectral characteristics acquired in real time indicate that the ablation has reached the uncorroded metal substrate.

4. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 2, characterized in that, The rate of variation is characterized by changes in spectral line intensity, profile, or correlation with adjacent spectral lines.

5. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 1, characterized in that, S2 specifically includes: S21: Perform inflection point detection on each variation path in the set of spectral depth variation paths, and initially mark the detected peak points of sudden increase in variation rate or the points where the variation trend changes as candidate transition points. S22: Based on the candidate transition points of all preset characteristic band spectral lines, the common erosion step interval is determined by density clustering analysis, and the erosion step number corresponding to the center of each interval is confirmed as the inter-corrosion layer transition position. S23: For each confirmed transition position, construct multi-dimensional transition indicators; S24: Arrange the multidimensional transition indices of all transition positions in the order of erosion to form a set of transition indices characterizing the layering characteristics of the corrosion layer structure.

6. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 5, characterized in that, The multi-dimensional transition indicators specifically include: The transition position dimension is the erosion cycle number corresponding to the transition position; The mutation rate abrupt increase magnitude dimension is the average abrupt increase magnitude of the mutation rate of all characteristic band spectral lines at the said transition position; The degree of linkage of spectral lines is the synchronous correlation coefficient of the variation trend of spectral lines in different characteristic bands at the transition position.

7. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 5, characterized in that, The density clustering analysis is based on a preset neighborhood search radius and minimum number of points. It performs high-density regional clustering on the transition points and identifies several regions with concentrated erosion steps. The center position of the erosion steps in each cluster region is taken as the transition position between corrosion layers.

8. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 1, characterized in that, S3 specifically includes: S31: Construct and train a structure mapping model, which is a supervised machine learning model trained on a multi-type corrosion sample library; S32: Using the transition index set as the input features of the structure mapping model, and the corresponding true values ​​of corrosion depth and layer thickness as the training targets, the structure mapping model is trained and validated. S33: Input the set of transition indices of the target corrosion region to be tested into the trained structure mapping model, and directly output the estimated total corrosion layer depth and the estimated thickness of each sublayer of the target corrosion region. S34: Based on the structure mapping model, when outputting the depth estimate, the confidence level through the uncertainty measure is given simultaneously.

9. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 8, characterized in that, Each sample in the sample library includes the true values ​​of the total depth and layer thickness of the corrosion layer as determined by metallographic methods, as well as the extracted set of transition indices.

10. The method for inverting the corrosion layer depth of steel based on LIBS spectral correlation analysis according to claim 8, characterized in that, Specifically, S32 includes: S321: The transition index set of each corrosion sample is used as the input feature of the structure mapping model. The transition index set includes the position of each transition point, the intensity of the mutation rate abrupt increase, and the degree of linkage of spectral line groups. S322: Using the total corrosion depth and the thickness of each sublayer of the corresponding sample obtained by metallographic calibration as the training target, the structure mapping model is trained and validated through supervised learning, so that the model learns the mapping relationship between transition index features and the layered structure of the corrosion layer.

Images