Stress and hardness nondestructive quantitative evaluation method based on reconstructed hysteresis parameters
By preparing and processing magnetic Barkhausen signals, reconstructing hysteresis loops and establishing mapping models, the problems of insufficient physical interpretability of models and difficulty in separating stress-hardness coupling in traditional methods are solved, realizing rapid, accurate and non-destructive evaluation of material stress and hardness.
Patent Information
- Application Number
- CN202511266930.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-05
- Publication Date
- 2025-11-11
AI Technical Summary
Traditional magnetic Barkhausen analysis methods lack in-depth exploration of hysteresis characteristics, resulting in insufficient physical interpretability and generalization ability of the model. They are difficult to effectively distinguish the coupled effects of material stress and hardness, and the measurement process is complex and not suitable for rapid on-site testing.
By preparing a series of test blocks, controlling the hardness through heat treatment, conducting standardized tensile tests, collecting and noise-reducing magnetic Barkhausen signals, reconstructing hysteresis loops, establishing a calibration database, and using machine learning algorithms to train a mapping model, the mapping from reconstructed hysteresis parameters to stress and hardness is realized.
It enables non-destructive and rapid assessment of material stress and hardness, with simple hysteresis loop reconstruction, separable coupling effects of stress and hardness, and stable measurement results suitable for on-site testing.
Smart Images
Figure CN120927489A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of magnetic nondestructive testing technology, and in particular to a nondestructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters. Background Technology
[0002] In materials science and engineering, accurate measurement of material stress state and hardness is crucial for evaluating material properties, predicting service life, and ensuring structural safety. Traditional stress measurement and hardness testing techniques face certain limitations. First, stress measurement methods such as X-ray diffraction and ultrasonic testing rely on precision instruments, have cumbersome and time-consuming procedures, and some techniques require material pretreatment. Second, hardness testing methods such as Brinell hardness and Rockwell hardness are not only complex to operate and have long testing cycles, but also leave irreversible indentations on the material surface, affecting material integrity and subsequent performance. Therefore, developing a rapid, non-destructive, and highly accurate stress and hardness testing technology is of significant engineering importance.
[0003] Magnetic Barkhausen noise technology, as an emerging non-destructive testing method, measures the noise signal generated by the motion of magnetic domain walls in ferromagnetic materials under an alternating magnetic field, reflecting the material's microstructure, stress state, and hardness. Because the magnetic Barkhausen signal is sensitive to the magnetoelastic and magnetomechanical effects of materials, it is widely used for stress, hardness, and microstructure assessment. However, traditional magnetic Barkhausen analysis methods typically rely on empirical statistical parameters such as RMS values and peak amplitudes. These parameters have a weak physical correlation with the material's mechanical properties, limiting the stability and universality of the measurement results. The main problems are as follows: (1) The mapping relationship between magnetic Barkhausen signal and mechanical properties is unclear: Most existing studies are based on the empirical relationship between the amplitude, energy and other statistical characteristics of magnetic Barkhausen signal and stress and hardness, lacking in-depth exploration of hysteresis characteristics, resulting in insufficient physical interpretability and generalization ability of the model. (2) Difficulty in reconstructing hysteresis loop: Hysteresis loop can intuitively reflect the magnetization process of materials, but traditional methods require the use of high-precision fluxmeters or vibrating sample magnetometers (VSMs), etc. The measurement process is complicated and not suitable for rapid on-site detection. (3) The coupling effect of stress and hardness is difficult to separate: In practical applications, the stress and hardness changes of materials often affect the magnetic Barkhausen signal at the same time. Existing methods are difficult to effectively distinguish the contributions of the two, resulting in a decrease in measurement accuracy. Summary of the Invention
[0004] To address the above technical problems, this invention proposes a non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters, comprising the following steps: Step 1: Prepare a series of test blocks with different stress-hardness characteristics, adjust the hardness parameters of each series of test blocks through heat treatment process, and conduct standardized tensile tests on each series of test blocks. Step 2: Acquire magnetic Barkhausen signals and perform noise reduction processing on the magnetic Barkhausen signals; Step 3: Reconstruct the hysteresis loop using the noise-reduced Barkhausen signal and establish a calibration database for material stress and hardness; Step 4: Establish a mapping model from reconstructed hysteresis parameters to material stress and hardness using the calibration database.
[0006] Furthermore, a series of test blocks with different stress-hardness characteristics are prepared, the hardness parameters of each series of test blocks are controlled by heat treatment, and standardized tensile tests are performed on each series of test blocks, including the following steps: Step 1: Select a material consistent with the material to be tested as the base material to prepare a test block. Based on the heat treatment window of the material, the hardness is continuously graded and controlled within a large range by precisely controlling the quenching and tempering process parameters. Based on the linear relationship between the yield strength and hardness of the material to be tested, a quantitative correlation model of stress-hardness is established. Step 2: The hardness parameters of the base material test blocks are controlled by heat treatment to obtain a series of test blocks with varying hardness gradients. The elastic modulus E and yield strength σ of the test blocks with varying hardness gradients are determined by standardized tensile testing. _ Furthermore, the material state was verified by fracture morphology analysis, and the machining parameters of the prepared specimens of the basic material were adjusted to introduce a residual stress field on the surface to establish the initial stress. The N sets of stress sequences with gradient step size Δσ are used to construct an N×N two-dimensional experimental matrix of stress-hardness combinations.
[0007] Furthermore, the magnetic Barkhausen signal is acquired through a noise reduction processing detection system and denoised using a wavelet transform algorithm. The noise reduction processing detection system includes a magnetic Barkhausen noise excitation module, a signal conditioning module, a data acquisition module, and a signal processing module. The magnetic Barkhausen noise excitation module is an optimized U-shaped magnetic yoke excitation module with a stacked design of a specific width-to-thickness ratio. The signal conditioning module includes a preamplifier and a bandpass filter to amplify the effective signal and suppress out-of-band interference. The data acquisition module digitally acquires the conditioned signal at a sampling frequency conforming to Nyquist's theorem. The signal processing module includes a wavelet transform-based noise suppression algorithm to perform multi-scale decomposition of the original magnetic Barkhausen signal to separate noise components and retain effective signal characteristics.
[0008] Furthermore, the establishment of the hysteresis loop reconstruction and stress / hardness calibration database is characterized by the following steps: Step 1: After noise removal The magnetic flux density is obtained by performing time-domain integration. Step 2: Drawing and verifying the hysteresis loop; Step 3: Extract features from the magnetic Barkhausen signal and calculate its hysteresis parameters such as reconstructed coercivity, reconstructed remanence, reconstructed hysteresis loss, and reconstructed maximum permeability. Step 4: Construct a three-tier data structure system for the Magnetic Backhausen database.
[0009] Furthermore, based on the mapping model established and applied as described in step 4 above, the mapping model is trained and optimized through machine learning algorithms, and can accurately characterize the nonlinear relationship between the reconstruction parameters and stress and hardness, including the following steps: Step 1: Data Preparation and Preprocessing First, data extraction and integration: First, the test block's unique number is used as the primary key. The test block's basic information and hysteresis parameters are internally connected through the test block number. Second, the hysteresis parameters and the original signal are left-connected through the signal ID. Finally, the environmental parameters are time-stamped with the original signal through the timestamp. Key fields are extracted from the basic information of the test block, including material grade code, heat treatment process parameters, geometric dimensions, reference stress value, and reference hardness value; coercivity, remanence, maximum permeability, hysteresis loss, and confidence index of each parameter are extracted from the hysteresis parameters; and the ambient temperature and humidity and excitation conditions at the time of acquisition are extracted from the original signal. Second, data quality assessment and cleaning: Calculate the missing rate of each field and draw a missing value matrix; Third, dataset partitioning: A stratified sampling strategy was adopted to stratify the datasets according to material grade and heat treatment process to ensure consistent distribution across subsets; Step 2, Model Selection: Use a multi-output gradient boosting tree model; Step 3, Model Training: The model training architecture adopts a design of shared feature extraction and independent output heads. It uses the shared underlying tree structure to capture common stress-hardness features and uses independent output heads to process material-specific features.
[0010] This invention discloses a non-destructive quantitative assessment method for stress and hardness based on reconstructed hysteresis parameters. The method involves preparing a series of stress and hardness test blocks, obtaining test blocks of different hardness levels through heat treatment, conducting tensile tests on each block, and measuring the corresponding magnetic Barkhausen signals. The measured magnetic Barkhausen signals are used to reconstruct the hysteresis loop, obtaining the reconstructed hysteresis parameters, and forming a calibration dataset with the corresponding stress and hardness. This calibration dataset is then used to establish a mapping model from the reconstructed hysteresis parameters to stress and hardness. Ultimately, given a material to be tested, this method only requires measuring the reconstructed hysteresis loop using the magnetic Barkhausen method, solving for the reconstructed hysteresis parameters, and substituting these parameters into the mapping model to obtain the true stress and hardness of the material. This non-destructive quantitative assessment method for stress and hardness based on reconstructed hysteresis parameters achieves non-destructive measurement based on the mapping relationship between reconstructed hysteresis parameters and true stress and hardness. The mapping relationship between the MBN signal and mechanical properties is clear, the hysteresis loop reconstruction is simple, and the coupling effects of stress and hardness can be separated, making it suitable for the rapid assessment of material stress, hardness, and hysteresis parameters. Attached Figure Description
[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments; Figure 1 This is a flowchart of the non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters according to the present invention. Detailed Implementation
[0012] This invention provides a non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters, comprising the following steps: Step 1: Prepare a series of test blocks with different stress-hardness characteristics, adjust the hardness parameters of each series of test blocks through heat treatment process, and conduct standardized tensile tests on each series of test blocks. Step 2: Acquire magnetic Barkhausen signals and perform noise reduction processing on the magnetic Barkhausen signals; Step 3: Reconstruct the hysteresis loop using the noise-reduced Barkhausen signal and establish a calibration database for material stress and hardness; Step 4: Establish a mapping model from reconstructed hysteresis parameters to material stress and hardness using the calibration database.
[0013] Furthermore, a series of test blocks with different stress-hardness characteristics are prepared, the hardness parameters of each series of test blocks are controlled by heat treatment, and standardized tensile tests are performed on each series of test blocks, including the following steps: Step 1: Select a material consistent with the material to be tested as the base material to prepare a test block. Based on the heat treatment window of the material, the hardness is continuously graded and controlled within a large range by precisely controlling the quenching and tempering process parameters. Based on the linear relationship between the yield strength and hardness of the material to be tested, a quantitative correlation model of stress-hardness is established. Step 2: The hardness parameters of the base material test blocks are controlled by heat treatment to obtain a series of test blocks with varying hardness gradients. The elastic modulus E and yield strength σ of the test blocks with varying hardness gradients are determined by standardized tensile testing. _y Furthermore, the material state was verified by fracture morphology analysis, and the machining parameters of the prepared specimens of the basic material were adjusted to introduce a residual stress field on the surface to establish the initial stress. N sets of stress sequences with gradient step size Δσ are used to construct an N×N two-dimensional experimental matrix of stress-hardness combinations. Specifically, the design of the basic material preparation specimen follows the ASTM E8 standard. A residual stress field is introduced into the surface layer by adjusting machining parameters (cutting speed v, feed rate f, cooling conditions) to establish the initial stress. N stress sequences with gradient step size Δσ (each group has n ≥ 3 parallel samples). Hardness gradient is achieved through a quench-temper process: reference hardness. The hardness was obtained by full quenching, followed by stepped tempering with a step size of ΔH, resulting in N sets of hardness distributions. Finally, an N×N two-dimensional experimental matrix of stress-hardness combinations was constructed, where the row direction represents the residual stress (…). ), column direction is hardness Each cell ( This represents a stress-hardness combination, covering N×N sets of experimental conditions. The elastic modulus E and yield strength σ of each set of specimens were determined through standardized tensile testing. _y The parameters were measured, and the material state was verified by fracture morphology analysis. This design, through independent control of two variables, can quantitatively analyze the synergistic mechanism of stress and hardness. A series of stress gradient specimens were prepared using tensile testing. The specific implementation plan is as follows: Quasi-static tensile tests were conducted using a WDW-100 electronic universal testing machine strictly following the ASTM E8 standard. The designed N×N stress-hardness combination matrix (stress gradient) was then analyzed. Hardness gradient (Each group has n=3 parallel samples), and the elastic modulus E, yield strength σ, and tensile strength of each sample are measured sequentially. The baseline mechanical parameters, such as elongation at break δ, were recorded, along with the corresponding experimental temperature and humidity data.
[0014] During the experiment, systematic errors in the equipment were eliminated by using the Latin square test sequence, and all valid data met the quality control requirement of less than 5% dispersion in three-repetition tests. SEM fracture morphology analysis of the fractured specimens was performed to exclude outliers and ensure the reliability of the data matrix. This standardized testing scheme ensures the acquisition of a high-confidence dataset demonstrating the synergistic effect of stress and hardness.
[0015] Furthermore, the magnetic Barkhausen signal is acquired through a noise reduction processing detection system and denoised using a wavelet transform algorithm. The noise reduction processing detection system includes a magnetic Barkhausen noise excitation module, a signal conditioning module, a data acquisition module, and a signal processing module. The magnetic Barkhausen noise excitation module is an optimized U-shaped magnetic yoke excitation module with a stacked design of a specific width-to-thickness ratio. The signal conditioning module includes a preamplifier and a bandpass filter, used to amplify the effective signal and suppress out-of-band interference. The data acquisition module digitally acquires the conditioned signal at a sampling frequency conforming to Nyquist's theorem. The signal processing module includes a wavelet transform-based noise suppression algorithm to perform multi-scale decomposition of the original magnetic Barkhausen signal to separate noise components and retain effective signal characteristics.
[0016] Step 1, the magnetic Barkhausen noise excitation module: an optimized U-shaped magnetic yoke excitation module, employing a stacked design with a specific width-to-thickness ratio, and its magnetic circuit length... With cross-sectional area A satisfy The relationship is given by K, where K is the material property coefficient. The excitation coil generates a periodic magnetization field by applying an alternating current, and the magnetization field strength generated by the excitation coil is... H ( t ) and excitation current I ( t ) satisfy Relationship, k is the magnetic circuit coefficient. N The number of coil turns. L For effective magnetic circuit length; the multi-layer shielded signal pickup module uses a special winding process for its detection coil to ensure magnetic flux. Φ Maximize the detection of the induced voltage signal containing Barkhausen noise, and detect the original signal output by the coil. Satisfies the law of electromagnetic induction: ,in Φ The magnetic flux passing through the detection coil; Step 2: The signal conditioning module includes a preamplifier and a bandpass filter, used to amplify the effective signal and suppress out-of-band interference, wherein the gain of the preamplifier is... ,in To acquire the minimum measurable voltage of the system, the transfer function of the bandpass filter is: ; in fc f is the cutoff frequency, and f is the system frequency; Step 3: The data acquisition module digitizes the conditioned signal at a sampling frequency that conforms to Nyquist's theorem, wherein the sampling frequency... It should satisfy ; Magnetic excitation current I ( t sampling frequency Should satisfy ,in For excitation frequency; Step 4: The signal processing module includes a wavelet transform-based noise suppression algorithm to perform multi-scale decomposition of the original Barkhausen magnetic signal V(t) to separate noise components and retain effective signal features. Because wavelets possess good time-frequency locality and orthogonality, they are suitable for transient signal analysis. This wavelet is selected as the basis function. The signal is decomposed into approximate coefficients at different scales using Discrete Wavelet Transform (DWT). and details ,in: ; Where j is the decomposition scale and k is the translation parameter. Subsequently, thresholding denoising is performed on the detail coefficients using an adaptive soft thresholding method, where the threshold λ is calculated based on the noise standard deviation σ. ; The denoised signal is reconstructed using inverse wavelet transform. ; ; in: and These are the coefficients after thresholding.
[0017] Furthermore, the establishment of the hysteresis loop reconstruction and stress / hardness calibration database is characterized by the following steps: Step 1: Analyze the noise-removed Barkhausen magnetic signal. The magnetic flux density is obtained by time-domain integration, i.e., after noise removal. The magnetic flux density is obtained by time-domain integration. ; ; To reduce discretization errors, this invention employs the trapezoidal integral method to obtain the magnetic flux density. ; ; in: This represents the sampling time interval.
[0018] To eliminate system drift while ensuring hysteresis loop closure, discrete DC bias is used for correction. Let... Let N be the integral value of the magnetic flux density at the nth sampling point, where N is the number of sampling points in a single integration period. The corrected magnetic flux density... for: ; In the formula: Let n be the integral value of the magnetic flux density at the nth sampling point. This is the corrected magnetic flux density.
[0019] The corrected accuracy meets the following requirements: ; In the formula, max(|B|) represents the integral value of the maximum absolute magnetic flux density.
[0020] Calculation and filtering of magnetic field strength H(t). Magnetic field strength is calculated based on the fundamental formula: ; In the formula The effective magnetic circuit length is represented by I(t), the loop excitation current is represented by N, and the number of sampling points in a single integration cycle is represented by N.
[0021] Current signal sampling: at the same sampling frequency as the magnetic Barkhausen sampling frequency Synchronous sampling frequency The excitation current signal is sampled. And it satisfies: ; The excitation current value at the nth sampling point is denoted as I[n]; ; In the formula, Δt represents the sampling interval. Where n is the sampling frequency and n is the number of sampling points.
[0022] The discrete magnetic field strength H[n] is used for calculation: ; In the formula: N represents the total number of turns of the excitation coil, Represented as the effective magnetic circuit length; Signal filtering: A low-pass filter is used to filter the magnetic field strength signal; ; Where h[k] are the low-pass filter coefficients and M is the filter order.
[0023] Step 2, Hysteresis Loop Plotting and Closure Verification: First, align the synchronously acquired discrete data B[n] and H[n] according to time to form a (H[n], B[n]) data point sequence, and then plot the hysteresis loop. Verify the closure of the plotted hysteresis loop using the following conditions: ; In the formula, max(|B|) represents the integral value of the maximum absolute magnetic flux density.
[0024] If this condition is not met, check for integral drift or synchronization error issues.
[0025] Step 3: Extract features from the Barkhausen magnetic signal and calculate its hysteresis parameters, including reconstructed coercivity, reconstructed remanence, reconstructed hysteresis loss, and reconstructed maximum permeability. First, the calculation of key hysteresis parameters: Coercivity The intersection of the hysteresis loop and the H-axis is calculated using the following formula: ; In the formula: max{H|B=0} and min{H|B=0} represent the maximum and minimum integral values of magnetic flux density when the magnetic flux density B=0, respectively.
[0026] remanence The intersection of the hysteresis loop and the B-axis is taken as... and The average value of the magnetic flux density B at that point, i.e.: ; In the formula: express The average value of the magnetic flux density B at that location. express The average value of the magnetic flux density B at that location.
[0027] Maximum permeability The maximum slope of the hysteresis loop, i.e.: ; Hysteresis loss The area enclosed by the hysteresis loop is calculated using the trapezoidal approximation method: ; In the formula, H[n] represents the discrete magnetic field strength. denoted as discrete magnetic flux density.
[0028] Secondly, feature extraction and construction are performed on the magnetic Barkhausen signal. To systematically evaluate the performance of magnetic materials, the following three dimensions of characteristics were constructed: magnetic parameter ratio, product characteristics, and composite characteristics.
[0029] Magnetic parameter ratio characteristics: The core performance of a material can be intuitively reflected by the ratio relationship between parameters. The ratio reflects the degree of magnetization; the larger the ratio, the stronger the material's resistance to demagnetization. It represents the energy conversion efficiency. The higher the value, the better the magnetic conductivity per unit loss and the higher the energy utilization efficiency.
[0030] Product characteristics: revealing performance correlations through parameter product forms. It seems to reflect the size of the magnetic energy product, and this indicator directly determines the ability of magnetic materials to store magnetic energy. This establishes the correlation between magnetic permeability and hysteresis loss.
[0031] Composite characteristics: As a comprehensive indicator, it integrates hysteresis characteristics and energy loss parameters to evaluate the model from multiple dimensions.
[0032] Step 4: Construct a three-tiered data structure for the Magnetic Barkhausen database: First, the three-tier data structure architecture design: Test block basic information layer: stores material property data, including test block number (using "S001"-"SNNN" coding rules), material grade (compliant with GB / T 221-2008 standard), heat treatment process parameters, geometric dimensions, and reference mechanical parameters.
[0033] Hysteresis parameter layer: Records characteristic parameters extracted from the MBN signal, including coercivity Hc, remanence Br, maximum permeability μmax, and hysteresis loss Wh. Each parameter is accompanied by a confidence index (95% confidence interval).
[0034] Raw signal layer: Stores the raw voltage waveform in time series format, and synchronously saves the ambient temperature, humidity and excitation conditions at the time of acquisition.
[0035] Cross-modal retrieval of data such as MBN signals and mechanical parameters can be achieved through the unique number of the test block.
[0036] Secondly, data standardization processing Standardized format: The unified timestamp format is ISO 8601, the physical quantity units are forcibly converted to SI International System of Units, and the string encoding adopts UTF-8.
[0037] Outlier filtering: Grubbs test (significance level α=0.01) was used to remove outlier data, and the remaining data were subjected to a second verification using the 3σ principle.
[0038] Parameter normalization: Non-magnetic parameters such as stress and hardness are normalized using a min-max method, while magnetic parameters are normalized using Z-score normalization, such as: ; In the formula: , They are respectively The mean and standard deviation.
[0039] Finally, database implementation The implementation steps, using a MySQL relational database, include: First, environment configuration: Install the appropriate version of MySQL and configure the server parameters according to the data requirements; Secondly, data migration: Experimental data is transformed and loaded using ETL tools, and a two-stage commit protocol is adopted to ensure data integrity. A fragmented storage strategy is used for the raw signal data; Secondly, quality verification: CHECK constraints are established using the built-in physical rule checker of the Jiles-Atherton model, automatically marking data records that violate ferromagnetic theory to ensure compliance. The physical rationality of the data; Finally, the dynamic calibration function supports users to define new magnetic parameter calculation formulas, and the system automatically generates a versioned storage scheme.
[0040] Furthermore, the mapping model is trained and optimized through machine learning algorithms, and can accurately characterize the nonlinear relationship between the reconstruction parameters and stress and hardness, including the following steps: Step 1: Data Preparation and Preprocessing First, data extraction and integration: First, the test block's unique number is used as the primary key. The test block's basic information table and the hysteresis parameter table are joined internally by the test block number. Second, the hysteresis parameter table and the original signal table are joined left by the signal ID. Finally, the environmental parameter table is time-scaled and aligned with the original signal table using timestamps. Key fields were selected as follows: Extracted from the basic information layer of the test block: material grade code conforming to GB / T 221-2008 standard, heat treatment process parameters, geometric dimensions, reference stress value, and reference hardness value; extracted from the hysteresis parameter layer: coercivity. ,remanence Maximum permeability Hysteresis loss And the confidence index of each parameter; extract from the original signal layer: environmental temperature and humidity, excitation conditions, etc. during acquisition.
[0041] Second, data quality assessment and cleaning: Calculate the missing value rate for each field and plot a missing value matrix. For key features (such as...) , If any feature (such as temperature or humidity) is missing, the sample must be removed directly. If minor features (such as temperature or humidity) are missing, the median of the sample is used to fill the gaps. Finally, samples with a missing percentage greater than X% are removed.
[0042] Using the 3σ principle, values exceeding the mean are excluded. Data was calculated at multiples of the standard deviation. Simultaneously, a reasonable range was set based on the material's physical properties, and data that clearly did not conform to the material's magnetic properties was discarded.
[0043] Inconsistent data processing: First, the data samples are time-series aligned to ensure that the hysteresis parameters match the time scale of the original signal. Second, all parameters are converted to international standard units.
[0044] Third, dataset partitioning: A stratified sampling strategy is adopted to divide the dataset into sub-subsets according to material grade and heat treatment process to ensure that the distribution of each subset is consistent.
[0045] Step 2, Model Selection: Since the relationship between the magnetic Barkhausen signal and stress hardness exhibits piecewise nonlinearity, and different material grades show varying feature importance, this invention employs a multi-output gradient boosting model.
[0046] First, this model can simultaneously model the joint distribution of stress and hardness, capturing the intrinsic correlation between mechanical properties, improving accuracy by 5-15% compared to modeling them separately. Second, through automatic feature interaction learning via a tree structure, it can reasonably match the complex nonlinear relationship between hysteresis parameters and stress and hardness. Finally, the model has built-in regularization, maintaining good generalization ability under moderate sample data conditions.
[0047] Step 3, Model Training: The model training architecture adopts a design of shared feature extraction and independent output heads. It uses the shared underlying tree structure to capture common stress-hardness features and uses independent output heads to process material-specific features.
[0048] This model employs an improved multi-output gradient boosting tree architecture, achieving the prediction of stress and hardness from magnetic Barkhausen signals through three-stage progressive training: First, a shared feature extraction layer is constructed in the basic training stage, and a composite loss function is used to learn global feature relationships; the composite loss function is: ; Where: stress loss term The absolute error of direct constraint stress prediction requires a higher weight because stress is more sensitive to changes in hysteresis parameters. ; In the formula This represents the actual stress of the i-th sample. This represents the predicted stress of the i-th sample.
[0049] Hardness loss item It is more robust to outliers, preventing hardness prediction from being dominated by extreme values. The relationship between hardness and magnetic parameters is more nonlinear, requiring flexible constraints. (Threshold X is adjusted for HRC scaling characteristics); ; In the formula: This represents the actual Rockwell hardness of the i-th sample. This represents the predicted Rockwell hardness of the i-th sample.
[0050] Physical regularization term The forced hardness-stress gradient is close to the material's elastic modulus E. The Jacobian matrix of the predicted value is calculated by automatic differentiation to balance data fitting and physical rationality.
[0051] ; In the formula This represents the derivative of the predicted Rockwell hardness with respect to the predicted stress, and this value should be close to the material's elastic modulus E.
[0052] Next, in the fine-tuning stage, a dynamic sample weight and feature masking mechanism is introduced to freeze the shared layer parameters and focus on optimizing the task-specific layer to adapt to the magnetomechanical response characteristics of different materials. Finally, the physical rule reinforcement layer applies material strength limits and mechanical relationship constraints to the prediction results to ensure that the output conforms to physical laws.
[0053] Secondly, model evaluation and optimization were conducted. In verifying the magnetomechanical coupling effect, the ratio of coercivity to remanence was used. By calculating the correlation coefficient between this ratio and the predicted stress value, if the correlation coefficient is greater than a threshold X, it indicates that the model has successfully captured the magnetomechanical coupling effect; if the correlation coefficient is lower than the threshold, the model's learning ability for the coupling effect is strengthened by adding physical constraint terms related to magnetomechanical coupling. When evaluating the physical rationality of the model, the ratio of maximum permeability to hysteresis loss is used. If this feature shows the expected negative correlation with hardness, the physical mechanism of the model is reliable; if the correlation is inconsistent, check the parameter settings related to hysteresis loss and hardness in the model, or adjust the network structure to enhance feature interaction capabilities. In the key feature analysis of the stress prediction model, the product of coercivity and remanence... The model should have a high SHAP value (SHAP stands for Shapley Additive Explanations, which quantifies the contribution of each feature to the model's prediction; this feature has a significant impact on the model's output), and the prediction error in the high-value range should be significantly lower than in other ranges. If the SHAP value is low or the error in the high-value range does not converge, a feature selection algorithm should be used to remove redundant features, or data augmentation methods should be used to expand the samples in the high-value range. Simultaneously, the model's regularization parameters should be adjusted to reduce overfitting and improve feature discriminative power. For testing the model's ability to fit complex correlations, the product of the maximum permeability and the hysteresis loss ratio should be used. If the observed nonlinear relationship with hardness matches expectations, it indicates that the model has a good fitting ability; if the fitting effect is poor, a nonlinear activation function is introduced to enhance the model's expressive ability.
[0054] Finally, using composite features Perform outlier detection.
[0055] In the formula: For maximum permeability, This represents the hysteresis loss ratio.
[0056] If the model fails to output low confidence for eigenvalues exceeding the theoretical range of the material, a density-based outlier detection algorithm (Local Outlier Factor, LOF) is used to assist in the judgment, or the model output layer structure is adjusted. Simultaneously, uncertainty estimation is introduced to dynamically calibrate the confidence level, improving the model's robustness to outlier data. For the material under test, only its magnetic Barkhausen signal needs to be acquired and the hysteresis loop reconstructed; the stress and hardness values of the material can then be quickly and accurately predicted using a mapping model.
[0057] This invention discloses a non-destructive quantitative assessment method for stress and hardness based on reconstructed hysteresis parameters. The method involves preparing a series of stress and hardness test blocks, obtaining test blocks of different hardness levels through heat treatment, conducting tensile tests on each block, and measuring the corresponding magnetic Barkhausen signals. The measured magnetic Barkhausen signals are used to reconstruct the hysteresis loop, obtaining the reconstructed hysteresis parameters, and forming a calibration dataset with the corresponding stress and hardness. This calibration dataset is then used to establish a mapping model from the reconstructed hysteresis parameters to stress and hardness. Ultimately, given a material to be tested, this method only requires measuring the reconstructed hysteresis loop using the magnetic Barkhausen method, solving for the reconstructed hysteresis parameters, and substituting these parameters into the mapping model to obtain the true stress and hardness of the material. This non-destructive quantitative assessment method for stress and hardness based on reconstructed hysteresis parameters achieves non-destructive measurement based on the mapping relationship between reconstructed hysteresis parameters and true stress and hardness. The mapping relationship between the magnetic Barkhausen signal and mechanical properties is clear, the hysteresis loop reconstruction is simple, and the coupling effects of stress and hardness can be separated. It is suitable for the rapid assessment of material stress, hardness, and hysteresis parameters.
[0058] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not limited to the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters, characterized in that, By acquiring Barkhausen magnetic signals, a hysteresis loop is reconstructed, and a mapping model from the reconstructed hysteresis parameters to material stress and hardness is established. This includes the following steps: Step 1: Prepare a series of test blocks with different stress-hardness characteristics, adjust the hardness parameters of each series of test blocks through heat treatment process, and conduct standardized tensile tests on each series of test blocks. Step 2: Acquire magnetic Barkhausen signals and perform noise reduction processing on the magnetic Barkhausen signals; Step 3: Reconstruct the hysteresis loop using the noise-reduced Barkhausen signal and establish a calibration database for material stress and hardness; Step 4: Establish a mapping model from reconstructed hysteresis parameters to material stress and hardness using the calibration database.
2. The non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters as described in claim 1, characterized in that, Step 1, which establishes a quantitative correlation model between stress and hardness through heat treatment and standardized tensile testing, includes the following steps: Step 1: Select a material consistent with the material to be tested as the base material to prepare a test block. Based on the heat treatment window of the material, the hardness is continuously graded and controlled within a large range by precisely controlling the quenching and tempering process parameters. Based on the linear relationship between the yield strength and hardness of the material to be tested, a quantitative correlation model of stress-hardness is established. Step 2: The hardness parameters of the base material test blocks are controlled by heat treatment to obtain a series of test blocks with varying hardness gradients. The elastic modulus E and yield strength σ of the test blocks with varying hardness gradients are determined by standardized tensile testing. y Furthermore, the material state was verified by fracture morphology analysis, and the machining parameters of the prepared specimens of the basic material were adjusted to introduce a residual stress field on the surface to establish the initial stress. The N sets of stress sequences with gradient step size Δσ are used to construct an N×N two-dimensional experimental matrix of stress-hardness combinations.
3. The non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters as described in claim 1, characterized in that: In step 2, the magnetic Barkhausen signal is acquired through a noise reduction processing detection system, and the signal is denoised using a wavelet transform algorithm. The noise reduction processing detection system includes a magnetic Barkhausen noise excitation module, a signal conditioning module, a data acquisition module, and a signal processing module. The magnetic Barkhausen noise excitation module is an optimized U-shaped magnetic yoke excitation module with a stacked design of a specific width-to-thickness ratio. The signal conditioning module includes a preamplifier and a bandpass filter. The signal conditioning module is used to amplify the effective signal and suppress out-of-band interference. The data acquisition module digitally acquires the conditioned signal at a sampling frequency that conforms to Nyquist's theorem. The signal processing module includes a wavelet transform-based noise suppression algorithm to perform multi-scale decomposition on the original magnetic Barkhausen signal to separate noise components and retain effective signal features.
4. The non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters as described in claim 1, characterized in that, Step 3 involves constructing a three-layer structured database, from the original Barkhausen magnetic signal layer to the reconstructed hysteresis parameter layer and then to the material stress and hardness mapping layer, to achieve non-destructive quantitative characterization of material stress and hardness. This includes the following steps: Step 1: Analyze the noise-removed Barkhausen magnetic signal. The magnetic flux density is obtained by performing time-domain integration. Step 2: Drawing and verifying the hysteresis loop; Step 3: Extract features from the magnetic Barkhausen signal and calculate its reconstructed coercivity, reconstructed remanence, reconstructed hysteresis loss, reconstructed maximum permeability, and other reconstructed hysteresis parameters. Step 4: Construct a three-tier data structure system for the Magnetic Backhausen database.
5. The non-destructive quantitative evaluation method for stress and hardness based on reconstructed hysteresis parameters as described in claim 1, characterized in that: In step 4, a mapping model from reconstructed hysteresis parameters to material stress and hardness is established using the calibration database. This mapping model is trained and optimized using machine learning algorithms and can accurately characterize the nonlinear relationship between reconstructed hysteresis parameters and stress and hardness. The steps include: Step 1: Data Preparation and Preprocessing First, data extraction and integration: First, the test block's unique number is used as the primary key. The test block's basic information and hysteresis parameters are internally connected through the test block number. Second, the hysteresis parameters and the original signal are left-connected through the signal ID. Finally, the environmental parameters are time-stamped with the original signal through the timestamp. Key fields are extracted from the basic information of the test block, including material grade code, heat treatment process parameters, geometric dimensions, reference stress value, and reference hardness value; coercivity, remanence, maximum permeability, hysteresis loss, and confidence index of each parameter are extracted from the hysteresis parameters; and the ambient temperature and humidity and excitation conditions at the time of acquisition are extracted from the original signal. Second, data quality assessment and cleaning: Calculate the missing rate of each field and draw a missing value matrix; Third, dataset partitioning: A stratified sampling strategy is adopted to stratify the datasets according to material grade and heat treatment process to ensure that the distribution of each subset is consistent. Step 2, Model Selection: Use a multi-output gradient boosting tree model; Step 3, Model Training: The model training architecture adopts a design of shared feature extraction and independent output heads. It uses the shared underlying tree structure to capture common stress-hardness features and uses independent output heads to process material-specific features.