Titanium plate performance analysis method and system

Abstract

The application provides a titanium plate performance analysis method and system, relates to the technical field of metal material detection, and comprises the following steps: a three-dimensional structure of a titanium plate surface is established by using an optical scanner, and a test area is divided; stress, hardness and conductivity data of test points are synchronously collected by using a multi-probe cooperative detection device; a double-branch feature extraction network is constructed, including a surface topography feature branch and a physical feature branch; features in the test area are statistically integrated, distribution characteristics are calculated, and a weight coefficient is determined; a performance analysis result is generated through feature fusion and a multi-task output layer, and finally, a titanium plate performance grade is determined. The method realizes deep fusion of topography information and physical characteristics, and improves the accuracy and reliability of titanium plate performance evaluation.

Description

Technical Field

[0001] This invention relates to metal material testing technology, and more particularly to a method and system for analyzing the performance of titanium plates. Background Technology

[0002] Existing methods for analyzing titanium plate performance mainly rely on single physical quantity tests or simple surface morphology inspections, making it difficult to comprehensively reflect the overall performance characteristics of titanium plates. For example, stress testing cannot reflect changes in the material's microstructure, hardness testing has limited predictive value for material fatigue performance, and conductivity testing is easily affected by environmental factors. These single-parameter analysis methods lack effective correlation mechanisms, leading to biased analysis results and insufficient accuracy in predicting titanium plate performance, failing to meet the stringent requirements of high-end fields such as aerospace. While some existing methods attempt to combine multiple testing methods, they mostly involve simple superposition or averaging, failing to establish physical correlation models between parameters. Furthermore, due to the significant spatial non-uniformity of titanium plate performance, the overall uniform sampling strategy used in traditional methods often ignores the crucial impact of local anomalies on overall performance, resulting in significant biases in the evaluation results. In addition, the lack of an effective multi-scale feature fusion mechanism also limits the ability of existing methods to accurately assess and predict titanium plate performance. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for analyzing the performance of titanium plates, which can solve the problems in existing technologies.

[0004] A first aspect of the present invention provides a method for analyzing the performance of titanium plates, comprising:

[0005] An optical scanner is used to scan the surface of the titanium plate under test to establish a three-dimensional structure and divide it into multiple test areas. Multiple test points are set in a matrix distribution within each test area. A multi-probe collaborative detection device is used to perform performance testing on each test point. The multi-probe collaborative detection device includes stress probes, hardness probes, and conductivity probes uniformly arranged along the circumference, simultaneously collecting stress data, hardness data, and conductivity data of the test points. A dual-branch feature extraction network is constructed, including a planar feature branch for extracting surface morphology features and a physical feature branch based on materials science knowledge. The physical feature branch includes a physical constraint layer and a physical correlation layer, used to establish the correlation between stress data, hardness data, and conductivity data. The distribution characteristics of the surface morphology features and physical features in each test area are calculated, and corresponding feature weight coefficients are determined based on the distribution characteristics. The features are then weighted and fused. Based on the fused features, a multi-task output layer generates titanium plate performance analysis results, and the performance level of the titanium plate under test is determined based on the performance analysis results.

[0006] Optionally, the steps of using an optical scanner to scan the surface of the titanium plate to be tested to establish a three-dimensional structure, dividing it into multiple test areas, and setting multiple test points in a matrix distribution within each test area include:

[0007] Image data of the surface of the titanium plate under test is acquired using a dual-camera collaborative acquisition system. Initial three-dimensional point cloud data is obtained by phase measurement based on the image data. Vibration errors in the initial three-dimensional point cloud data are eliminated using the birefringence optical path difference method, and temperature drift errors in the initial three-dimensional point cloud data are compensated based on a pre-calibrated temperature-deformation curve to obtain compensated three-dimensional point cloud data. Local curvature distribution data of the surface of the titanium plate under test is calculated based on the compensated three-dimensional point cloud data. The surface of the titanium plate under test is divided into multiple test areas according to the local curvature distribution data, where the area of ​​each test area is inversely proportional to the local curvature value within that test area. A test point matrix is ​​set within each test area based on the local curvature distribution data, where the distance between adjacent test points within each test area is inversely proportional to the local curvature value at that test point location.

[0008] Optionally, the steps for performing performance testing at each test point using a multi-probe collaborative testing device include:

[0009] The stress release characteristic time of the titanium plate under test is obtained, and the measurement sequence of the stress probe, hardness probe, and conductivity probe is determined based on the stress release characteristic time. During the stress probe measurement, the load application rate is controlled to be proportional to the yield strength of the titanium plate under test to obtain stress data at the test point. During the conductivity probe measurement, the initial conductivity data of the test point is obtained, and the perturbation correction coefficient of the stress field on the conductivity is calculated based on the stress data. The initial conductivity data is corrected based on the perturbation correction coefficient to obtain conductivity data. During the hardness probe measurement, the initial hardness data of the test point is obtained, and the deformation compensation factor of the hardness measurement on the stress field is calculated based on the indentation depth. The initial hardness data is corrected based on the deformation compensation factor to obtain hardness data. The stress data and conductivity data are corrected based on the deformation compensation factor to obtain the corrected stress data and conductivity data of the test point.

[0010] Optionally, a dual-branch feature extraction network is constructed, including a planar feature branch for extracting surface morphology features and a physical feature branch constructed based on materials science knowledge; the physical feature branch includes a physical constraint layer and a physical correlation layer, and the step of establishing the correlation between stress data, hardness data, and electrical conductivity data includes:

[0011] The local curvature value and height distribution probability are calculated based on the height data of the titanium plate surface to be tested, and morphological entropy features are generated based on the local curvature value and height distribution probability. Height deviation data of the titanium plate surface to be tested are collected, and surface roughness parameters are calculated. The morphological entropy features and the surface roughness parameters are used to construct surface morphological features. The hardness variation law corresponding to the stress data is analyzed to establish the correspondence between stress and hardness. The variation law of electrical conductivity data with dislocation density is analyzed to establish the correspondence between electrical conductivity and dislocation density. The correspondence between stress and hardness, and the correspondence between electrical conductivity and dislocation density are used as physical constraint features. An interaction influence matrix of stress, hardness, and electrical conductivity is constructed. The interaction strength of the stress data, hardness data, and electrical conductivity data is calculated based on the interaction influence matrix. The stress data, hardness data, and electrical conductivity data are corrected based on the interaction strength, and the corrected data are used to construct physical correlation features.

[0012] Optionally, the step of calculating the distribution characteristics of the surface morphology features and physical features in each test area, determining the corresponding feature weight coefficients based on the distribution characteristics, and weighted fusion of features includes:

[0013] The morphology entropy spatial distribution of the surface morphology features in each test area is calculated, and the morphology entropy spatial distribution is obtained by calculating the negative logarithm of the probability of the surface morphology features in the test area; the multi-physical quantity correlation strength of the physical features in each test area is calculated, and the multi-physical quantity correlation strength is calculated based on the correlation coefficient matrix between physical quantities; the dispersion coefficient and anomaly detection index in the test area are calculated based on the morphology entropy spatial distribution and the multi-physical quantity correlation strength; the regional importance is obtained by weighting the dispersion coefficient with the multi-physical quantity correlation strength, and the feature reliability is calculated based on the anomaly detection index; the local weight coefficient of each test area is calculated based on the regional importance and the feature reliability, and the global weight coefficient is calculated based on the local weight coefficient; the surface morphology features and physical features in each test area are weighted based on the local weight coefficient to obtain local fusion features, and the local fusion features are weighted based on the global weight coefficient to obtain global fusion features.

[0014] Optionally, the step of calculating the dispersion coefficient and anomaly detection index within the test area based on the spatial distribution of the morphological entropy and the correlation strength of the multiple physical quantities includes:

[0015] The scale dispersion coefficient within the test area is calculated based on the spatial distribution of the morphological entropy, and the directional dispersion coefficient is calculated based on the correlation strength of the multiple physical quantities. The scale dispersion coefficient is obtained by weighting the ratios of the standard deviations to the means at multiple feature scales, and the directional dispersion coefficient is obtained by normalizing the deviations of the feature gradients from the mean. The scale dispersion coefficient and the directional dispersion coefficient are weighted and combined to obtain a comprehensive dispersion coefficient. Feature anomaly scores are calculated based on the spatial distribution of the morphological entropy and the correlation strength of the multiple physical quantities. Temporal weights are updated based on the temporal changes of the feature anomaly scores, and the feature anomaly scores are adjusted based on the temporal weights to obtain an anomaly detection index. The comprehensive dispersion coefficient and the anomaly detection index are combined to obtain the regional importance of each test area. The feature weights of the test areas are updated based on the regional importance to obtain local weight coefficients for the test areas. The local weight coefficients are weighted and combined with the feature similarity of adjacent test areas to obtain a global weight coefficient, where the weight coefficient of the global weight coefficient is proportional to the feature similarity of adjacent test areas.

[0016] Optionally, based on the fused features, the titanium plate performance analysis results are generated through a multi-task output layer, and the step of determining the performance level of the titanium plate under test based on the performance analysis results includes:

[0017] The fused features are input into a shared feature extraction layer and a task-specific layer. The shared feature extraction layer performs multi-scale hierarchical extraction on the fused features to generate multi-scale features. The task-specific layer includes a performance distribution prediction branch, a defect type identification branch, and a reliability assessment branch. In the performance distribution prediction branch, the spatial distribution data of the multi-scale features is calculated, the distribution interval of performance indicators is determined based on the spatial distribution data, and the distribution interval is dynamically updated to generate a performance distribution prediction result. In the defect type identification branch, the multi-scale features are matched layer by layer with preset defect features to determine the optimal matching feature, and the defect type and its corresponding confidence level are output based on the optimal matching feature. In the reliability assessment branch, the performance distribution prediction result and the defect type and its confidence level are used as evaluation criteria, the credibility of each evaluation criterion is calculated, and the evaluation criteria are weighted and combined based on the credibility to obtain a reliability assessment result. Based on a pre-set performance level determination standard, combined with the performance distribution prediction result, the defect type and its confidence level, and the reliability assessment result, the performance level of the titanium plate to be tested is determined through interval mapping.

[0018] Optionally, the steps of using the performance distribution prediction results and the defect types and their confidence levels as evaluation criteria, calculating the credibility of each evaluation criterion, and weighting and combining the evaluation criteria according to the credibility to obtain the reliability evaluation results include:

[0019] The probability distribution deviation and the prediction error of the defect type confidence level of the performance distribution prediction results are calculated, and an uncertainty matrix is ​​constructed based on the probability distribution deviation and the prediction error. The temporal fluctuation characteristics of each evaluation criterion are calculated according to the uncertainty matrix to obtain a stability index for the evaluation criterion. Based on the stability index, the historical evaluation results of the evaluation criterion are tracked and analyzed to calculate the accuracy change trend of the evaluation results, and dynamic evaluation weights are generated based on the accuracy change trend. The stability index and the dynamic evaluation weights are combined to obtain the credibility of each evaluation criterion. The performance distribution prediction results and the defect type and its confidence level are weighted and fused based on the credibility to output a reliability evaluation result. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the titanium plate performance analysis method according to an embodiment of the present invention;

[0021] Figure 2 A flowchart for dual-branch feature extraction in titanium plate performance analysis. Detailed Implementation

[0022] Figure 1 This is a flowchart illustrating the titanium plate performance analysis method of the present invention, as shown below. Figure 1 As shown, the method includes:

[0023] An optical scanner was used to scan the surface of the titanium plate under test to establish a three-dimensional structure and divide it into multiple test areas. Multiple test points were set in a matrix distribution within each test area.

[0024] A multi-probe collaborative testing device is used to perform performance testing on each test point. The multi-probe collaborative testing device includes stress probes, hardness probes and conductivity probes that are uniformly arranged along the circumference, and simultaneously collects stress data, hardness data and conductivity data of the test points.

[0025] A dual-branch feature extraction network is constructed, including a planar feature branch for extracting surface morphology features and a physical feature branch based on materials science knowledge. The physical feature branch includes a physical constraint layer and a physical correlation layer, which are used to establish the correlation between stress data, hardness data, and electrical conductivity data. The features extracted from multiple test points in each test area are statistically integrated to obtain the feature representation of the test area. The distribution characteristics of the surface morphology features and physical features in each test area are calculated, and the corresponding feature weight coefficients are determined according to the distribution characteristics. The features are then weighted and fused.

[0026] Based on the fused features, the performance analysis results of the titanium plate are generated through a multi-task output layer, and the performance level of the titanium plate under test is determined according to the performance analysis results.

[0027] For example, an optical scanner is used to scan the surface of the titanium plate under test to establish a three-dimensional structure. The established three-dimensional structure is then divided into regions, dividing the entire titanium plate into multiple test areas. A grid method is used for region division, uniformly dividing the 1m × 1m titanium plate into 10 × 10 grids, with each grid measuring 100mm × 100mm, forming 100 test areas. Each test area is identified by a corresponding number, such as A1 to J10. Multiple test points are set in a matrix distribution within each test area. For each 100mm × 100mm test area, a 5 × 5 matrix of test points is set, with a spacing of 20mm between test points, resulting in 25 test points per test area. The test points are identified by a combination of region number and point number, such as A1-1 to A1-25.

[0028] A multi-probe collaborative testing device was used to perform performance testing at each test point. The testing device was positioned above each test point using a precision positioning system. The support descended to bring the probe into contact with the titanium plate surface, simultaneously acquiring stress, hardness, and conductivity data for each test point. The acquired data was transmitted to a data processing unit via a high-speed data acquisition system, forming a multi-physical quantity dataset for each test point. For the 2500 test points on the entire titanium plate, a total of 7500 raw data values ​​were generated.

[0029] A dual-branch feature extraction network was constructed, comprising a planar feature branch for extracting surface morphology features and a physical feature branch based on materials science knowledge. The planar feature branch extracts surface morphology features from the 3D structural data. The extraction process first calculates the statistical characteristics of height data within each test region, including mean, variance, skewness, and kurtosis; then, it calculates the local curvature values ​​by fitting a quadratic surface to a 3×3 region around each test point, calculating the principal curvature and Gaussian curvature; finally, it generates the height distribution probability, dividing the height values ​​into 10 equally spaced intervals, and statistically analyzing the proportion of points in each interval to the total number of points as the probability of that interval. Morphology entropy features include curvature entropy and height entropy. Curvature entropy is calculated by weighted sum of the probabilities of the principal curvature distribution, reflecting the complexity of the surface curvature distribution; height entropy is calculated by weighted sum of the height distribution probabilities, reflecting the uniformity of the surface height distribution. For well-processed TC4 titanium plates, the curvature entropy is typically between 0.2 and 0.5, and the height entropy is between 0.8 and 1.2. For areas with processing marks or defects, the curvature entropy exceeds 0.8, and the height entropy exceeds 1.5. Simultaneously, height deviation data of the titanium plate surface is collected, and surface roughness parameters are calculated. Roughness parameters include arithmetic mean roughness Ra, maximum peak-to-valley height Rz, and profile peak density RPc. The arithmetic mean roughness Ra is the absolute mean of the height deviation, typically 0.8–2.5 μm; the maximum peak-to-valley height Rz is the height difference between the highest and lowest points, typically 5–15 μm; and the profile peak density RPc is the number of peaks per unit length, typically 30–60 peaks / cm. The morphological entropy features and surface roughness parameters are used to construct a surface morphological feature, forming a 7-dimensional feature vector that comprehensively characterizes the morphological properties of the titanium plate surface. The physical feature branch includes a physical constraint layer and a physical correlation layer, used to establish the correlation between stress data, hardness data, and electrical conductivity data. The physical constraint layer analyzes the hardness variation pattern corresponding to stress data, establishing a correlation between stress and hardness. For TC4 titanium alloy, the correlation between stress and hardness is based on the strain hardening theory of materials science. When a material undergoes plastic deformation under external force, the hardness value increases with increasing stress. Through statistical analysis of a large amount of test data, it was found that for TC4 titanium alloy, the hardness value increases by approximately 20 HV for every 100 MPa increase in stress. Specifically, the correlation can be expressed as follows: when the stress is in the range of 800-850 MPa, the hardness value is typically 330-340 HV; when the stress is in the range of 850-900 MPa, the hardness value is typically 340-350 HV; and when the stress is in the range of 900-950 MPa, the hardness value is typically 350-370 HV. This correlation reflects the relationship between the internal dislocation density and macroscopic mechanical properties of the material. The variation pattern of electrical conductivity data with dislocation density is analyzed, establishing a correlation between electrical conductivity and dislocation density. The physical correlation layer constructs an interaction matrix of stress, hardness, and electrical conductivity.The interaction matrix is ​​a 3×3 matrix, with diagonal elements equal to 1 and off-diagonal elements representing the degree of mutual influence between different physical quantities. Through analysis of extensive test data of TC4 titanium alloy, typical interaction matrix values ​​were determined: the effect of stress on hardness is 0.7, and the effect of stress on electrical conductivity is -0.4; the effect of hardness on stress is 0.5, and the effect of hardness on electrical conductivity is -0.3; the effect of electrical conductivity on stress is -0.2, and the effect of electrical conductivity on hardness is -0.1. Positive values ​​indicate positive correlation, negative values ​​indicate negative correlation, and the absolute value indicates the degree of influence.

[0030] The interaction strength of stress, hardness, and electrical conductivity data is calculated based on the interaction influence matrix. The interaction strength is obtained by multiplying the interaction coefficient between the physical quantities by the normalized value of each physical quantity. Normalization of a physical quantity is achieved by subtracting the mean from the original value and then dividing by the standard deviation. For TC4 titanium alloy, a typical normalization range is -2 to 2. The stress, hardness, and electrical conductivity data are then corrected based on the interaction strength. An iterative method is used for correction; in each iteration, the correction amount for the current physical quantity is calculated based on the influence of the other two physical quantities and the current value. This process is repeated 5-10 times until the correction amount changes very little. The corrected stress, hardness, and electrical conductivity data are then used to construct a physical correlation feature, forming a 9-dimensional feature vector.

[0031] Features extracted from multiple test points within each test region are statistically integrated to obtain the feature representation of that test region. For each test region's 25 test points, the mean, variance, maximum, minimum, and median of 7-dimensional surface morphology features and 15-dimensional physical features (including physical constraint features and physical correlation features) are calculated to form a comprehensive feature representation of the region. After statistical integration, the feature dimensions of each test region remain unchanged, but represent the average level and variation of the entire region.

[0032] Distribution characteristics include spatial distribution properties and numerical distribution properties. Spatial distribution properties are determined by calculating the gradient of feature changes between adjacent regions, while numerical distribution properties are determined by calculating the distribution pattern of feature values. Features with uniform spatial distribution and stable numerical distribution are assigned lower weights, while features with non-uniform spatial distribution and large numerical fluctuations are assigned higher weights. Feature weighting fusion uses a weighted averaging method, averaging surface morphology features and physical features according to weight coefficients to obtain a fused feature vector. Based on the fused features, the titanium plate performance analysis results are generated through a multi-task output layer.

[0033] Optionally, the steps of using an optical scanner to scan the surface of the titanium plate to be tested to establish a three-dimensional structure, dividing it into multiple test areas, and setting multiple test points in a matrix distribution within each test area include:

[0034] Image data of the surface of the titanium plate under test is acquired using a dual-camera collaborative acquisition system; initial three-dimensional point cloud data is obtained by phase measurement based on the image data.

[0035] The vibration error in the initial three-dimensional point cloud data is eliminated by the birefringence optical path difference method, and the temperature drift error in the initial three-dimensional point cloud data is compensated based on the pre-calibrated temperature-deformation curve to obtain the compensated three-dimensional point cloud data.

[0036] Based on the compensated three-dimensional point cloud data, the local curvature distribution data of the surface of the titanium plate to be tested is calculated. According to the local curvature distribution data, the surface of the titanium plate to be tested is divided into multiple test areas, wherein the area of ​​each test area is inversely proportional to the local curvature value within the test area.

[0037] Based on the local curvature distribution data, a test point matrix is ​​set in each test area, wherein the spacing between adjacent test points in each test area is inversely proportional to the local curvature value at the test point location.

[0038] For example, a dual-camera collaborative acquisition system is used to acquire images of the titanium plate surface. This dual-camera system includes two industrial-grade cameras with a baseline distance of 180 mm between them, an optical axis angle of 15 degrees, a focal length of 35 mm, and a field of view of 300 mm × 400 mm. The camera resolution is 4096 × 3072 pixels, and the frame rate is 60 frames per second.

[0039] During image acquisition, a diffuse reflection coating was applied to the titanium plate surface to enhance its texture. A dual-camera system used structured light illumination to project a grid pattern onto the titanium plate surface, with a grid period of 1 mm, continuously acquiring 30 frames of images at different angles. The acquired image data was processed using a phase unwrapping algorithm to calculate the correspondence between phase values ​​and actual height. A 360-degree change in phase value corresponds to a 0.5 mm change in actual height. Initial 3D point cloud data with a resolution of 0.05 mm was generated using phase measurement methods.

[0040] The initial point cloud data contains noise caused by equipment vibration and drift errors due to temperature changes. The vibration error is eliminated using the birefringence optical path difference method. Specifically, a birefringent crystal is added to the optical path, splitting the incident light into two mutually perpendicular polarized beams. The resulting optical path difference is proportional to the vibration displacement. By detecting the change in the optical path difference, the displacement error caused by vibration is calculated.

[0041] Temperature drift error compensation is based on a pre-calibrated temperature-deformation curve. During calibration, a high-precision temperature sensor records changes in ambient temperature while simultaneously monitoring the surface deformation of the titanium plate. Within a temperature range of 15-35 degrees Celsius, data points are taken at 2-degree intervals to establish the correlation between temperature and deformation. For example, the average expansion of the titanium plate surface height is 0.003 mm for every 1-degree Celsius increase in temperature. Based on this relationship, ambient temperature is measured in real time to compensate for temperature drift in the point cloud data, improving the accuracy of the compensated point cloud data to 0.003 mm.

[0042] The local curvature distribution of the titanium plate surface was calculated based on the compensated 3D point cloud data. The curvature calculation employed a neighborhood fitting method, where a spherical neighborhood with a radius of 5 mm was selected as the center of each point, and a quadratic surface was fitted to calculate the Gaussian curvature and mean curvature at that point. The Gaussian curvature value typically ranged from 0.001 to 0.1, and the mean curvature value ranged from 0.005 to 0.05. Based on the calculated curvature distribution data, the titanium plate surface was divided into multiple test regions using a curvature threshold segmentation method. A baseline curvature threshold of 0.01 was first set, and regions with curvature values ​​greater than this threshold were marked as high-curvature regions, while those less than the threshold were marked as low-curvature regions. For locations with curvature gradients greater than 0.005 / mm, region boundaries were established. The area of ​​the test region was inversely proportional to the average local curvature value within that region, with a specific proportionality coefficient of 10000 square millimeters. For example, when the average curvature of a region is 0.02, the area is set to 500 square millimeters; when the average curvature is 0.01, the area is set to 1000 square millimeters. This division method ensures denser test coverage in areas with large curvature variations.

[0043] Test points are set within each predefined test area. The layout of the test points is also related to the local curvature value; the spacing between test points is inversely proportional to the local curvature value at that point. A baseline spacing of 10 mm and a baseline curvature of 0.01 are set. When the local curvature value is 0.02, the spacing between adjacent test points is set to 5 mm; when the local curvature value is 0.005, the spacing between adjacent test points is set to 20 mm. For test points at the boundaries, the density is increased by 25% to better capture boundary features. For example, for a 300 mm × 300 mm titanium plate sample, it is generally divided into 15-25 test areas, with 16-64 test points set within each area, forming a matrix distribution.

[0044] This method employs a curvature-based adaptive region division and test point layout strategy, which significantly improves testing efficiency and data quality. It is more adaptable to the geometric features of titanium plate surfaces and can achieve denser sampling in complex structural regions while ensuring measurement accuracy.

[0045] Optionally, the steps for performing performance testing at each test point using a multi-probe collaborative testing device include:

[0046] The stress release characteristic time of the titanium plate to be tested is obtained, and the measurement sequence of the stress probe, hardness probe and conductivity probe is determined based on the stress release characteristic time.

[0047] During the stress probe measurement process, the load application rate is controlled to be proportional to the yield strength of the titanium plate under test, so as to obtain the stress data at the test point.

[0048] During the conductivity probe measurement process, the initial conductivity data of the test point is acquired, the perturbation correction coefficient of the stress field on the conductivity is calculated based on the stress data, and the initial conductivity data is corrected based on the perturbation correction coefficient to obtain the conductivity data.

[0049] During the hardness probe measurement process, the initial hardness data of the test point is acquired, the deformation compensation factor of the hardness measurement on the stress field is calculated based on the indentation depth, and the initial hardness data is corrected based on the deformation compensation factor to obtain the hardness data; the stress data and the conductivity data are corrected based on the deformation compensation factor to obtain the corrected stress data and conductivity data of the test point.

[0050] For example, the multi-probe collaborative detection device consists of a stress probe, a hardness probe, and a conductivity probe. These three probes are evenly distributed along the circumference, with an angle of 120 degrees between them, forming a three-probe detection unit. Each probe is connected to a central controller to achieve synchronous data acquisition and analysis. The stress probe uses a piezoelectric strain gauge sensor with a sensitivity of 0.5 microvolts / microstrain; the hardness probe uses the micro-indentation testing principle with a maximum load of 5 Newtons and a diamond Vickers indenter; the conductivity probe uses a four-point measurement method with a measurement range of 0.5-60 MSiemens / meter and a resolution of 0.01 MSiemens / meter.

[0051] Before the testing process begins, the stress release characteristic time of the titanium plate to be tested needs to be obtained. The stress release characteristic time refers to the time required for the stress value of the titanium plate to stabilize after a load is applied. This parameter is related to the material composition and heat treatment state of the titanium plate. The stress release characteristic time is determined through pre-testing: a standard load (usually 50% of the yield strength of the titanium plate) is applied to the titanium plate sample, and the stress value is recorded as a curve over time. The time interval from the stress value reaching its peak value to its stable value is the stress release characteristic time. For common TC4 titanium alloy plates, the stress release characteristic time is generally between 3 and 8 seconds; for high-strength titanium alloys that have undergone special heat treatment, this time is extended to 10-15 seconds.

[0052] Based on the measured stress release characteristic time, the measurement sequence of the three probes was determined. The principle of the measurement sequence design is: the stress probe measures first, and after the stress field stabilizes, the conductivity probe and hardness probe measure in sequence. Specifically, the stress probe measurement duration is 1.5 times the stress release characteristic time to ensure complete stress field stabilization; the conductivity probe starts immediately after the stress probe measurement, with a measurement duration of 2 seconds; the hardness probe starts after the conductivity probe measurement, with a measurement duration of 3 seconds. This sequence arrangement minimizes mutual interference between the measurements of each physical quantity.

[0053] During stress probe measurements, the load application rate (Newtons / second) is equal to the yield strength (MPa) × 0.05. For example, for TC4 titanium alloy with a yield strength of 900 MPa, the load application rate is set to 45 Newtons / second. The load is applied in a gradient manner: for the first 30% of the time, the load increases at 70% of the set rate; for the middle 40% of the time, it increases at 100% of the set rate; and for the last 30% of the time, it increases at 50% of the set rate, forming a smooth load curve. The raw data obtained from the stress probe measurements are acquired by a data acquisition system with a sampling frequency of 1000 Hz. High-frequency noise is removed using a filtering algorithm to obtain the stress data at the test points.

[0054] During conductivity probe measurement, the initial conductivity data at the test points is first acquired. This initial conductivity data is obtained using a four-point measurement method. Four equally spaced electrodes are arranged in the measurement area. A constant current (typically 5 mA) is applied to the two outer electrodes, while the potential difference is measured on the two inner electrodes. The resistance value is calculated using Ohm's law, and then converted to a conductivity value by combining this with the geometric parameters of the test area (electrode spacing of 2 mm and measurement depth of 0.5 mm). The original conductivity value is then corrected for temperature (reference temperature 20 degrees Celsius, correction factor is 0.4% for every 1 degree Celsius increase in conductivity) to obtain the initial conductivity data. Since the stress field can disturb the conductivity measurement, a correction factor for the stress field disturbance to conductivity needs to be calculated. The perturbation correction factor is related to the stress value and stress gradient. The calculation method is as follows: when the stress value at the measurement point is in the range of 0-300 MPa, the perturbation correction factor is 1 + stress value × 0.0002; when the stress value is in the range of 300-800 MPa, the perturbation correction factor is 1 + stress value × 0.0003; when the stress value is greater than 800 MPa, the perturbation correction factor is 1 + stress value × 0.0004. For example, when the stress value at the measurement point is 500 MPa, the perturbation correction factor is 1.15. Dividing the initial conductivity data by the perturbation correction factor yields the corrected conductivity data. For TC4 titanium alloy, the conductivity data before correction is generally between 0.58-0.62 MSiemens / m, and after correction, it is usually between 0.52-0.56 MSiemens / m.

[0055] The hardness probe measurement process employs a multi-stage loading method with loads of 2 Newtons, 3.5 Newtons, and 5 Newtons, each held for 0.5 seconds. The indentation depth under each load stage is recorded to calculate the initial hardness data. The initial hardness data calculation is based on the Vickers hardness measurement principle, calculating the hardness value by measuring the diagonal length of the indentation (using a built-in optical microscope system with 400x magnification and 0.5 micrometer resolution). Specifically, the applied load (in kilograms of force) is divided by the indentation surface area (in square millimeters), and then multiplied by a constant of 1.8544 to obtain the Vickers hardness value. For each test point, the average of the hardness values ​​calculated under three loads is taken as the initial hardness data. Since the hardness measurement process causes deformation of the local stress field, a deformation compensation factor needs to be calculated. The deformation compensation factor is closely related to the indentation depth. When the indentation depth is in the range of 0-20 micrometers, the deformation compensation factor is 1 + indentation depth × 0.015; when the indentation depth is in the range of 20-40 micrometers, the deformation compensation factor is 1 + indentation depth × 0.02; and when the indentation depth is greater than 40 micrometers, the deformation compensation factor is 1 + indentation depth × 0.025. Taking an indentation depth of 25 micrometers as an example, the deformation compensation factor is 1.5. Multiplying the initial hardness data by the deformation compensation factor yields the corrected hardness data. For TC4 titanium alloy, the hardness value before correction is typically between 320-350 HV, and after correction, it is between 480-525 HV.

[0056] Local deformation caused by hardness measurement not only affects the hardness value but also the measured stress and conductivity data. Therefore, secondary correction is needed for the stress and conductivity data. The correction method for stress data is: Corrected stress value = Original stress value × (1 - Deformation compensation factor × 0.1). The correction method for conductivity data is: Corrected conductivity value = Original conductivity value × (1 - Deformation compensation factor × 0.05). For a deformation compensation factor of 1.5, the correction coefficient for stress value is 0.85, and the correction coefficient for conductivity value is 0.925. Through this multi-physical quantity interactive correction mechanism, the corrected stress, hardness, and conductivity data for the test points are finally obtained.

[0057] The multi-probe collaborative detection method employed in this study achieves high-precision collaborative measurement of three physical quantities—stress, hardness, and conductivity—through a rationally designed measurement sequence. The measurement sequence design based on stress release characteristic time reduces measurement interference. By introducing disturbance correction coefficients and deformation compensation factors, an interactive correction mechanism among multiple physical quantities is established, significantly improving the accuracy and consistency of measurement data and providing a reliable data foundation for the comprehensive evaluation of titanium plate performance.

[0058] Optionally, a dual-branch feature extraction network is constructed, including a planar feature branch for extracting surface morphology features and a physical feature branch constructed based on materials science knowledge; the physical feature branch includes a physical constraint layer and a physical correlation layer, and the step of establishing the correlation between stress data, hardness data, and electrical conductivity data includes:

[0059] The local curvature value and height distribution probability are calculated based on the height data of the titanium plate surface to be tested, and the morphology entropy feature is generated based on the local curvature value and height distribution probability; the height deviation data of the titanium plate surface to be tested are collected, and the surface roughness parameter is calculated; the morphology entropy feature and the surface roughness parameter are used to construct the surface morphology feature.

[0060] Analyze the hardness variation pattern corresponding to the stress data to establish the correspondence between stress and hardness; analyze the variation pattern of electrical conductivity data with dislocation density to establish the correspondence between electrical conductivity and dislocation density; use the correspondence between stress and hardness and the correspondence between electrical conductivity and dislocation density as physical constraint features.

[0061] An interaction matrix of stress, hardness, and electrical conductivity is constructed. The interaction strength of the stress data, hardness data, and electrical conductivity data is calculated based on the interaction matrix. The stress data, hardness data, and electrical conductivity data are corrected based on the interaction strength, and the corrected data are constructed as physical correlation features.

[0062] Combination Figure 2 The flowchart for the dual-branch feature extraction process in titanium plate performance analysis is explained below: The dual-branch network adopts a parallel structure design. Features are extracted independently by the two branches and then merged in the fusion layer. The planar feature branch uses a multi-scale convolutional structure, containing three scale layers with kernel sizes of 3×3, 5×5, and 7×7, and each layer contains 16 convolutional filters. The physical feature branch includes a physical constraint layer and a physical association layer. The physical constraint layer uses a fully connected structure, with the number of input nodes equal to the number of physical quantity data points at the test points. The hidden layer has 128 nodes, and the output layer has 64 nodes. The physical association layer uses a graph convolutional structure with 32 nodes and an edge feature dimension of 16.

[0063] The feature extraction process for the planar feature branch begins by calculating the local curvature value and height distribution probability based on the height data of the titanium plate surface under test. The height data is derived from the aforementioned 3D point cloud data, with a sampling interval of 0.1 mm. The local curvature value is calculated using a surface fitting method. A circular region with a radius of 0.5 mm is taken as the center of each point, and a quadratic surface is fitted. The principal curvatures k1 and k2 at that point are solved, and the Gaussian curvature (k1×k2) and mean curvature ((k1+k2) / 2) are calculated. For a typical TC4 titanium plate surface, the Gaussian curvature value ranges from 0.0001 to 0.01, and the mean curvature value ranges from 0.001 to 0.05. The height distribution probability is obtained by statistically analyzing the distribution of height values ​​within the region. The height value range is divided into 10 equally spaced intervals, and the proportion of points in each interval to the total number of points is calculated to obtain a height distribution histogram.

[0064] The morphological entropy features include two parts: curvature entropy and height entropy. Curvature entropy is calculated by dividing the local curvature value range into eight intervals, counting the percentage of points within each interval, and then summing the results after multiplying the negative logarithm of each percentage value with the percentage value itself. For regions with uniform surface morphology, the curvature entropy value is low, typically between 0.5 and 1.5; for regions with complex morphological variations, the curvature entropy value is high, reaching 2.0 to 3.0. Height entropy is calculated based on the aforementioned height distribution histogram, using a similar method to curvature entropy, by multiplying the negative logarithm of each interval's percentage value with the percentage value itself and then summing the results. The height entropy value of a typical titanium plate surface ranges from 1.0 to 2.5.

[0065] The surface height deviation data of titanium plates refers to the difference between the height value at each point and the average height of the region. Surface roughness parameters include arithmetic mean roughness Ra, root mean square roughness Rq, and maximum height difference Rz. The arithmetic mean roughness Ra is the average of the absolute values ​​of the height deviations, the root mean square roughness Rq is the square root of the average of the squares of the height deviations, and the maximum height difference Rz is the height difference between the highest and lowest points in the region. For aerospace-grade TC4 titanium plates, Ra is typically between 0.8 and 1.5 micrometers, Rq between 1.0 and 1.8 micrometers, and Rz between 4.0 and 8.0 micrometers. The morphological entropy features (curvature entropy and height entropy) are combined with the surface roughness parameters (Ra, Rq, Rz) to form a 7-dimensional feature vector, constituting the surface morphology features. This feature vector is passed through a feature mapping layer, which uses a two-layer fully connected network structure with 32 hidden layer nodes and 16 output layer nodes. The activation function is ReLU, mapping the 7-dimensional feature vector into a 16-dimensional feature representation.

[0066] The physical characteristics branch first analyzes the hardness variation pattern corresponding to stress data, establishing a correlation between stress and hardness. Stress and hardness data at each test point are paired and analyzed to extract stress-hardness pairs. For TC4 titanium alloy, the hardness variation rate is approximately 0.2 HV / MPa when the stress is in the range of 0-300 MPa; approximately 0.3 HV / MPa when the stress is in the range of 300-600 MPa; and approximately 0.4 HV / MPa when the stress is in the range of 600-900 MPa. Based on these variation rates, a piecewise linear mapping relationship is established to map the stress value to the expected hardness value, calculating the deviation between the actual and expected hardness values. For titanium plates with stable performance, the deviation is typically less than 5%; for areas with defects or unstable performance, the deviation can reach 15%-25%.

[0067] In establishing the relationship between electrical conductivity and dislocation density, the relationship between dislocation density and material hardness is based on the dislocation strengthening theory in metallurgy. For TC4 titanium alloy, the relationship between dislocation density and hardness is determined by combining the Taylor relation and the Hall-Petch effect: when the material undergoes plastic deformation, the increase in hardness is proportional to the square root of the dislocation density. Specifically, the baseline hardness value (approximately 330 HV) and the corresponding baseline dislocation density (approximately 5 × 10⁻⁶) of TC4 titanium alloy in the standard annealed state are first measured. 10 ( / square meter), and then calculate the dislocation density increment based on the hardness increment. For example, when the hardness value increases from the baseline value by 50 HV to 380 HV, the corresponding dislocation density increases by approximately 6 × 10⁻⁶. 12 / square meter. The relationship between electrical conductivity and dislocation density follows Matthiessen's rule, that is, the increase in resistivity is proportional to the dislocation density. By measuring the electrical conductivity and corresponding dislocation density of TC4 titanium alloy in different deformation states, an empirical relationship was established: for every 1 × 10⁻⁶ increase in dislocation density... 13 / square meter, the conductivity decreases by approximately 4.8%. Based on this relationship, the expected conductivity value is calculated and compared with the measured conductivity to obtain the conductivity deviation. The conductivity deviation of a normal titanium plate is usually within ±2%; in areas with microscopic defects, the deviation can reach ±5% or higher. The deviations in the correlation between stress and hardness and the correlation between conductivity and dislocation density are combined into physical constraint features, forming a 4-dimensional feature vector.

[0068] The interaction matrix is ​​a 3×3 matrix, with diagonal elements having a value of 1 and off-diagonal elements representing the influence coefficients between the two physical quantities. For TC4 titanium alloy, the influence coefficient of stress on hardness is approximately 0.4, the influence coefficient of stress on conductivity is approximately 0.3, the influence coefficient of hardness on stress is approximately 0.2, the influence coefficient of hardness on conductivity is approximately 0.25, the influence coefficient of conductivity on stress is approximately 0.1, and the influence coefficient of conductivity on hardness is approximately 0.15. The construction of the interaction matrix is ​​based on the fitting of materials science theory and experimental data. The specific construction method is as follows: first, measurement data of hardness and conductivity under different stress states are collected, requiring at least 100 sets of standard sample data; then, multivariate regression analysis is used to determine the coefficient relationships between each physical quantity; finally, the coefficients are adjusted to minimize the prediction error through iterative optimization. This matrix is ​​predetermined before network training and serves as a fixed parameter of the physical constraint layer.

[0069] The interaction strength of the three physical quantities is calculated based on the interaction matrix. The calculation method is as follows: for physical quantities i and j, the interaction strength equals the corresponding element value in the interaction matrix multiplied by the normalized value of physical quantity j. Normalization maps each physical quantity to the range of 0-1. The normalization formula uses the minimum-maximum scaling method, i.e., (value - minimum value) / (maximum value - minimum value). For TC4 titanium alloy, the normalized range for stress values ​​is 0-1000 MPa, for hardness values ​​it is 300-600 HV, and for electrical conductivity values ​​it is 0.5-0.7 M Siemens / meter.

[0070] The stress, hardness, and conductivity data are corrected based on the interaction strength. The correction process uses an iterative method. In each iteration, the correction amount for the current physical quantity is calculated based on the influence coefficients of the other two physical quantities and the current value. This process is repeated 5-10 times until the correction amount is less than a threshold (usually set to 0.1% of the original value). When calculating the corrected value of the current physical quantity, its original value is first retained, and then the influence of each other physical quantity is accumulated. The influence of each physical quantity is equal to its normalized value multiplied by the corresponding interaction coefficient and then multiplied by an adjustment factor of 0.1. These accumulated influences are then multiplied by the original value to obtain the final corrected value. For example, when correcting the stress value, the influence of hardness and conductivity needs to be considered; when correcting the hardness value, the influence of stress and conductivity needs to be considered; and when correcting the conductivity value, the influence of stress and hardness needs to be considered. The difference between the corrected stress data and the original data is typically between 3% and 8%, the difference in hardness data is between 4% and 10%, and the difference in conductivity data is between 2% and 6%. The corrected stress, hardness, and conductivity data are used to construct a physical correlation feature, forming a 9-dimensional feature vector. This feature vector is then encoded through a physical constraint layer, outputting a 64-dimensional feature representation.

[0071] The physical constraint layer is a crucial step in how it actually encodes materials science knowledge. The weights of the physical constraint layer are not initialized randomly, but rather pre-defined based on the laws of materials physics. Specifically, the initial values ​​of the weight matrix of the fully connected layer are determined according to theoretical models of stress-hardness, hardness-dislocation density, and dislocation density-conductivity relationships. For example, the first three rows of the weight matrix correspond to the stress-hardness relationship, with initial values ​​set according to the aforementioned piecewise linear mapping; rows 4-6 correspond to the hardness-dislocation density relationship; and rows 7-9 correspond to the dislocation density-conductivity relationship. This initialization method ensures that the network contains materials science knowledge before training begins, greatly accelerating the training process and improving physical interpretability.

[0072] The physical correlation layer employs a graph convolutional network structure to represent multiple physical quantities and their relationships as a graph. In this graph, nodes represent different physical quantities (stress, hardness, electrical conductivity) and their derived features (such as stress gradient, rate of change of hardness, etc.), and edges represent the relationships between physical quantities. Graph convolution operations update the features of the current node by aggregating information from neighboring nodes, ensuring that the relationships between physical quantities are fully utilized during feature extraction. The convolution kernel parameters of the graph convolutional layer are also initialized according to the interaction matrix, ensuring that the network can learn feature representations that conform to physical laws. The physical correlation layer outputs a 32-dimensional feature vector, representing the complex interaction relationships between multiple physical quantities.

[0073] The fusion of surface morphology features and physical features employs an attention mechanism. Specifically, the 16-dimensional feature vector output from the planar feature branch and the 96-dimensional feature vector output from the physical feature branch (a concatenation of the 64-dimensional physical constraint layer and the 32-dimensional physical association layer) are input into the attention layer. The attention layer calculates the correlation score between the two sets of features by performing a softmax normalization on the dot product of the two sets of features. Then, the two sets of features are weighted and summed based on the correlation score to obtain the fused feature. The fused feature has a dimension of 112 (16+96), and it passes through a dimensionality reduction layer (fully connected layer, output dimension 64) to obtain the final feature representation.

[0074] The training process of the dual-branch feature extraction network is a crucial step. The training dataset consists of two parts: a labeled dataset containing 200 sets of titanium plate samples with known performance levels, each set including surface morphology data and physical quantity measurement data; and an unlabeled dataset containing a large number of unlabeled titanium plate test data sets, totaling 2000 sets. Training employs a semi-supervised learning method, combining supervised and self-supervised learning strategies. The supervised learning part uses labeled data to train the network, and the loss function includes three parts: feature reconstruction loss, classification loss, and physical consistency loss. The feature reconstruction loss measures the network's ability to reconstruct input features and is calculated using mean squared error; the classification loss measures the accuracy of the network's prediction of the titanium plate's performance level and is calculated using cross-entropy loss; the physical consistency loss measures the degree of agreement between the network's prediction results and physical laws, calculated as the sum of squared deviations between the predicted physical quantity relationships and the theoretical physical model. The self-supervised learning part uses unlabeled data to train the network and designs three self-supervised tasks: feature mask reconstruction, physical quantity relationship prediction, and region consistency maintenance. Feature masking reconstruction involves randomly masking some input features and having the network predict the masked feature values; physical quantity relationship prediction involves predicting other physical quantities based on some physical quantities; and region consistency maintenance involves ensuring a smooth transition in feature representations between adjacent regions. These self-supervised tasks help the network learn the intrinsic structure and physical laws of the data, improving its generalization ability.

[0075] The optimization algorithm employs the Adam optimizer with an initial learning rate of 0.001, dynamically adjusted using cosine annealing. To prevent overfitting, weight decay (coefficient of 0.0001) and Dropout (ratio of 0.3) regularization techniques are used. The training run consists of 200 epochs, with 32 samples per epoch. Training data is augmented through random pruning, rotation, and noise addition. During training, model performance is evaluated on the validation set every 10 epochs, and the learning rate is dynamically adjusted based on validation performance. The parameters of the best-performing model are saved.

[0076] Model validation employed a 5-fold cross-validation strategy, randomly dividing the 200 labeled datasets into 5 parts. Four parts were used as the training set and one part as the validation set in each iteration, with the average performance metric calculated after 5 iterations. Evaluation metrics included classification accuracy, root mean square error (RMSE), correlation coefficient, and physical consistency score. Classification accuracy measures the accuracy of the model's prediction of titanium plate performance grades; RMSE measures the deviation between the predicted and actual physical quantities; the correlation coefficient measures the correlation between the predicted and actual results; and the physical consistency score measures the degree of agreement between the predicted results and physical laws.

[0077] Feature distribution analysis is a crucial step in feature extraction. For each test region, the statistical distribution characteristics of surface morphology and physical features are calculated, including mean, standard deviation, skewness, and kurtosis. These statistical characteristics reflect the central tendency and dispersion of the features, helping to identify anomalous regions. For example, a skewness value exceeding 1.5 or a kurtosis value exceeding 5 in the physical feature distribution usually indicates the presence of defects in that region. Feature distribution characteristics are also used to calculate feature weights; features with more uniform distributions have higher weights, while those with more anomalous distributions have lower weights, ensuring that the final fused features are not dominated by outliers.

[0078] The dual-branch feature extraction network constructed in this method achieves deep integration of surface morphology features and physical features through the design of morphology entropy feature extraction, physical constraint layer, and physical correlation layer. The physical feature branch establishes the correlation between multiple physical quantities, making the feature extraction process conform to the laws of materials science and significantly improving the physical interpretability and accuracy of the features. The morphology entropy analysis and interaction influence matrix mechanism introduced in the feature extraction process enhance the sensitivity to local abnormal characteristics of titanium plates, providing a reliable feature basis for the accurate evaluation of titanium plate performance.

[0079] Optionally, the step of calculating the distribution characteristics of the surface morphology features and physical features in each test area, determining the corresponding feature weight coefficients based on the distribution characteristics, and weighted fusion of features includes:

[0080] The morphology entropy spatial distribution of the surface morphology features in each test area is calculated, and the morphology entropy spatial distribution is obtained by calculating the negative logarithm of the probability of the surface morphology features in the test area; the correlation strength of the physical features among multiple physical quantities in each test area is calculated, and the correlation strength of the multiple physical quantities is obtained based on the correlation coefficient matrix between physical quantities.

[0081] Based on the spatial distribution of morphological entropy and the correlation strength of the multiple physical quantities, the dispersion coefficient and anomaly detection index within the test area are calculated; the dispersion coefficient and the correlation strength of the multiple physical quantities are weighted to obtain the regional importance, and the feature reliability is calculated based on the anomaly detection index; the local weight coefficient of each test area is calculated according to the regional importance and the feature reliability, and the global weight coefficient is calculated according to the local weight coefficient; the surface morphological features and physical features within each test area are weighted according to the local weight coefficient to obtain the local fusion feature, and the local fusion feature is weighted according to the global weight coefficient to obtain the global fusion feature.

[0082] For example, the process of calculating the spatial distribution of surface morphology features' morphology entropy in each test area is based on information entropy theory. The spatial distribution of morphology entropy reflects the complexity and uncertainty of surface morphology features. Specifically, the calculation method involves transforming the surface morphology features (such as curvature entropy and height entropy) within the test area into probability distributions, and then calculating the negative logarithmic weighted sum of these probabilities. The method for transforming into probability distributions involves dividing the feature values ​​into several intervals (usually 10 equally spaced intervals) according to their numerical magnitude, and statistically analyzing the proportion of data points within each interval to the total number of data points as the probability of that interval. For a feature probability p in a given interval, its corresponding morphology entropy contribution is p multiplied by the negative logarithm p. The spatial distribution of morphology entropy in the test area is the sum of the morphology entropy contributions from all intervals. For TC4 titanium plate areas with uniform surface morphology, the spatial distribution value of morphology entropy is typically between 1.5 and 2.0; for areas with slight processing marks, the value is between 2.0 and 2.5; and for areas with obvious defects, the value exceeds 3.0.

[0083] The multi-physical quantity correlation strength reflects the degree of interrelation between physical quantities such as stress, hardness, and electrical conductivity. The correlation strength is calculated based on a correlation coefficient matrix between the physical quantities. This matrix is ​​a 3×3 square matrix, with diagonal elements equal to 1 and off-diagonal elements representing the correlation coefficients between the corresponding physical quantities. The correlation coefficient is calculated by dividing the inner product of the standardized data of two physical quantities by the number of samples. For TC4 titanium alloy, the correlation coefficient between stress and hardness is typically between 0.65 and 0.80, between stress and electrical conductivity between 0.50 and 0.65, and between hardness and electrical conductivity between 0.45 and 0.60. The multi-physical quantity correlation strength is the weighted average of all off-diagonal elements in the correlation coefficient matrix. The weights are related to the importance of the physical quantity; typically, stress has a weight of 0.4, hardness a weight of 0.35, and electrical conductivity a weight of 0.25. For regions of TC4 titanium plates with stable performance, the multi-physical correlation strength is typically between 0.60 and 0.75; for regions with unstable performance, the value is either below 0.50 or above 0.85.

[0084] The coefficient of dispersion measures the uniformity of feature distribution within a region. It is calculated as the ratio of the standard deviation to the mean of the morphological entropy spatial distribution. For titanium plate regions with good uniformity, the coefficient of dispersion is typically less than 0.2; for regions with local anomalies, the coefficient of dispersion reaches 0.4 or higher. The anomaly detection index is used to identify outliers within a region. It is calculated as the proportion of points whose statistical feature values ​​deviate from the region's mean by more than twice the standard deviation. The anomaly detection index for normal titanium plate regions is typically less than 0.05; for regions with obvious defects, this index exceeds 0.15.

[0085] Regional importance reflects the significance of the test area in the overall performance evaluation. It is calculated by multiplying the dispersion coefficient by 0.35 and the multi-physical quantity correlation strength by 0.65. For critical performance areas (such as high-stress zones), the weight can be appropriately increased. The regional importance of a typical TC4 titanium plate ranges from 0.4 to 0.8, with high-importance areas typically being parts with high mechanical performance requirements. Feature reliability is calculated based on anomaly detection indicators. Feature reliability equals 1 minus the anomaly detection indicator, reflecting the credibility of the feature data. The feature reliability of normal areas is typically higher than 0.95, while the feature reliability of defective areas is as low as 0.8.

[0086] Local weight coefficients for each test region are calculated based on regional importance and feature reliability. The calculation method involves multiplying the regional importance by the feature reliability, followed by normalization (making the sum of the weight coefficients for all regions equal to 1). Normalization divides the initial weight of each region by the sum of the initial weights of all regions. The local weight coefficient reflects the importance of a single region within its neighborhood. For a TC4 titanium plate with 100 test regions, the local weight coefficient is typically between 0.005 and 0.025, with more important regions having higher weight coefficients. Global weight coefficients are then calculated based on the local weight coefficients. This calculation considers the spatial correlation between regions, weighting the local weight of each region with the local weights of its neighboring regions. The weight of neighboring regions decreases with increasing distance, typically using a Gaussian decay function with a decay radius of 3 grid units. The global weight coefficient reflects the overall importance of a region within the entire titanium plate, with a distribution similar to the local weight coefficients but with a smoother spatial distribution.

[0087] Surface topography and physical features within each test region are weighted based on local weighting coefficients. For each test region, each component of the 7-dimensional surface topography feature vector is multiplied by the local weighting coefficient for that region to obtain the weighted surface topography features; similarly, each component of the 9-dimensional physical feature vector is multiplied by the local weighting coefficient to obtain the weighted physical features. The weighted surface topography and physical features are concatenated into a 16-dimensional vector as the local fusion feature for that region. The local fusion feature retains information from the original features while also reflecting the importance of the region. For important regions, their features have a greater weight in the fusion process; for regions with low reliability, their feature contribution is appropriately reduced.

[0088] The calculation of the global fusion feature involves weighting and summing the local fusion features of all test areas according to their corresponding global weight coefficients, resulting in a 16-dimensional feature vector representing the comprehensive performance characteristics of the entire titanium plate. The first seven components of the global fusion feature originate from surface morphology features, reflecting the surface quality of the titanium plate; the latter nine components originate from physical features, reflecting the mechanical and electrical properties of the titanium plate. The global fusion feature can be directly used for the classification and rating of titanium plates. For aerospace-grade TC4 titanium plates, the components of the global fusion feature are typically within specific ranges: surface morphology-related components are between 0.5 and 1.5, and physical feature-related components are between 0.7 and 1.3.

[0089] In practical applications, the feature weighted fusion process requires processing a large amount of test data. For a 1m x 1m TC4 titanium plate, divided into 100 test areas, approximately 1000 test points are collected in each area. The data processing flow is as follows: first, the surface morphology and physical characteristics of each test area are calculated; then, the feature distribution is analyzed, weighting coefficients are determined, and finally, feature fusion is performed. The entire process takes approximately 3-5 minutes on a computing device with 16GB of memory and an 8-core CPU, meeting the real-time requirements of industrial testing.

[0090] This method introduces two core indicators—morphological entropy spatial distribution and multi-physical quantity correlation strength—to achieve intelligent weighting of the surface morphological and physical characteristics of titanium plates. The introduction of the dispersion coefficient and anomaly detection index enhances the algorithm's sensitivity to abnormal regions. The dual-layer weighting system of local and global weights ensures both the preservation of local detailed features and the comprehensive evaluation of global performance, providing an efficient and reliable technical means for the accurate analysis and quality control of titanium plate performance.

[0091] Optionally, the step of calculating the dispersion coefficient and anomaly detection index within the test area based on the spatial distribution of the morphological entropy and the correlation strength of the multiple physical quantities includes:

[0092] The scale dispersion coefficient within the test area is calculated based on the spatial distribution of the morphological entropy, and the directional dispersion coefficient is calculated based on the correlation strength of the multiple physical quantities. The scale dispersion coefficient is obtained by weighting the ratio of the standard deviation to the mean at multiple feature scales, and the directional dispersion coefficient is obtained by normalizing the deviation of the feature gradient from the mean. The scale dispersion coefficient and the directional dispersion coefficient are then combined by weighting to obtain the comprehensive dispersion coefficient.

[0093] The feature anomaly score is calculated based on the spatial distribution of the morphological entropy and the correlation strength of the multiple physical quantities. The temporal weight is updated based on the temporal change of the feature anomaly score. The feature anomaly score is then adjusted based on the temporal weight to obtain the anomaly detection index.

[0094] The comprehensive dispersion coefficient is combined with the anomaly detection index to obtain the regional importance of each test region. The feature weights of the test regions are updated based on the regional importance to obtain the local weight coefficients of the test regions.

[0095] The local weight coefficients are weighted and combined with the feature similarity of adjacent test regions to obtain the global weight coefficients. The weight coefficients of the global weight coefficients are proportional to the feature similarity of adjacent test regions.

[0096] For example, the process of calculating the scale dispersion coefficient within a test area based on the spatial distribution of morphological entropy can be achieved through statistical characteristic analysis at multiple feature scales. For instance, for a typical TC4 titanium alloy sheet, the surface morphological features are analyzed at three different scales: microscale (5 μm × 5 μm), mesoscale (50 μm × 50 μm), and macroscale (500 μm × 500 μm). At each scale, the ratio of the standard deviation to the mean of the morphological entropy is calculated and denoted as the microscale ratio, mesoscale ratio, and macroscale ratio, respectively. For stable titanium alloy regions, these three ratios are typically 0.15, 0.12, and 0.08, respectively; for defective regions, the ratios reach 0.30, 0.25, and 0.20, respectively. The scale dispersion coefficient is calculated by weighting these three ratios, with weights allocated as follows: microscale 0.5, mesoscale 0.3, and macroscale 0.2, to highlight the importance of microstructure. Therefore, the scale dispersion coefficient for stable regions is approximately 0.13, while that for defective regions is approximately 0.27.

[0097] The directional dispersion coefficient is calculated based on the analysis of characteristic gradients in the multi-physical quantity correlation strength calculation. For each test region, the gradient values ​​of stress, hardness, and conductivity in different directions (horizontal, vertical, and diagonal) are calculated. The gradient values ​​in normal titanium alloy regions are relatively uniformly distributed, with small deviations from the mean in each direction; however, in regions with non-uniform performance, the gradient directionality is significant, with obvious deviations from the mean in each direction. The directional dispersion coefficient is obtained by calculating the difference between the gradient in each direction and the average gradient, followed by normalization. Normalization involves dividing the deviation in each direction by the maximum deviation value to keep the result between 0 and 1. For regions with uniform performance, the directional dispersion coefficient is typically between 0.05 and 0.15; for regions with significant directional defects, the coefficient reaches 0.3 to 0.4.

[0098] The overall dispersion coefficient is obtained by multiplying the scale dispersion coefficient by 0.6 and the directional dispersion coefficient by 0.4, thus balancing the effects of scale and directional factors. For a tested TC4 titanium alloy plate, the scale dispersion coefficient for a typical region is 0.18, the directional dispersion coefficient is 0.12, and the calculated overall dispersion coefficient is 0.156. The overall dispersion coefficient reflects the degree of non-uniformity of feature distribution within a region and is an important indicator for assessing the importance of a region. A higher value indicates a more non-uniform feature distribution within the region, requiring greater attention.

[0099] The feature anomaly score measures the degree of deviation of a feature from the normal pattern within a test area. It is calculated by comparing the spatial distribution of morphological entropy with a standard distribution, while considering variations in the correlation strength of multiple physical quantities. The comparison between the spatial distribution of morphological entropy and the standard distribution is achieved by calculating the Mahalanobis distance, which takes into account the covariance structure of the feature. For TC4 titanium alloy, the standard distribution is determined by the statistical properties of multiple defect-free samples. The variation in the correlation strength of multiple physical quantities is calculated by the difference from a reference value, which is determined by the average correlation strength of a large number of normal samples, typically 0.68. For a certain test area, the Mahalanobis distance of the spatial distribution of morphological entropy is 2.3, and the deviation of the correlation strength of multiple physical quantities is 0.15. The feature anomaly score is calculated by dividing the Mahalanobis distance by 10 to obtain a base score of 0.23, then adding the correlation strength deviation multiplied by a weighting coefficient of 0.5, i.e., 0.15 × 0.5 = 0.075. The final feature anomaly score is the base score minus the weighted deviation, calculated as 0.23 - 0.075 = 0.225. Mahalanobis distance reflects the degree of morphological anomaly, while correlation strength deviation reflects the degree of physical property anomaly. Only by combining the two can the anomaly status of a region be comprehensively assessed. The characteristic anomaly score of normal regions is usually below 0.2, while the score of regions with obvious defects exceeds 0.4.

[0100] Time-series weights reflect the stability of features over time. In continuous production or testing scenarios, multiple measurements are performed on the same area, and the time series of feature anomaly scores are recorded. If the feature anomaly scores fluctuate little, it indicates that the performance of that area is stable, and the time-series weight is low; if the fluctuations are large, it indicates that there are unstable factors in that area, and the time-series weight should be increased. The time-series weight is calculated by dividing the standard deviation of the time series feature anomaly scores by the mean, and then multiplying by an adjustment factor of 0.8. For stable areas, the time-series weight is usually between 0.1 and 0.3; for unstable areas, it should be above 0.5.

[0101] The anomaly score is adjusted based on temporal weights by multiplying the anomaly score by (1 + temporal weight). For a region with an anomaly score of 0.225 and a temporal weight of 0.2, the calculated anomaly detection index is 0.27. This anomaly detection index comprehensively considers both the degree of feature anomaly and temporal stability, providing an important basis for calculating region importance. The anomaly detection index for normal regions is typically below 0.25, while that for problematic regions exceeds 0.4.

[0102] The regional importance of each test area is obtained by combining the comprehensive dispersion coefficient and the anomaly detection index. The combination method is to multiply the comprehensive dispersion coefficient by 0.45 and add the anomaly detection index by 0.55. For a region with a comprehensive dispersion coefficient of 0.156 and an anomaly detection index of 0.27, the calculated regional importance is 0.219. Regional importance reflects the criticality of the test area; a higher value indicates a greater impact of the area on overall performance, and it should be given higher weight in subsequent analysis. The importance of normal areas is usually between 0.15 and 0.25, while that of critical areas reaches 0.3 to 0.4.

[0103] The feature weights of test regions are updated based on regional importance to obtain local weight coefficients for each test region. The update method converts regional importance into an adjustment factor for the feature weights, calculating the base feature weight value plus the adjustment factor. The base feature weight value is 1 / N, where N is the total number of test regions; the adjustment factor is the regional importance minus the average of all regional importance values, multiplied by a scaling factor of 0.5. For 100 test regions, the base feature weight value is 0.01; for a regional importance of 0.219 and an average regional importance of 0.2, the calculated adjustment factor is 0.0095, resulting in a local weight coefficient of 0.0195. The local weight coefficient determines the contribution of regional features in the subsequent fusion process; important regions receive higher weights, thus playing a greater role in feature fusion.

[0104] The global weight coefficient is obtained by weighting the local weight coefficient with the feature similarity of adjacent test regions. Feature similarity is calculated by taking the cosine distance between the feature vectors of the current region and its neighboring regions. The smaller the cosine distance, the higher the similarity. For each region, the feature similarity of its four adjacent regions (above, below, left, and right) is considered, and the average similarity is calculated. The global weight coefficient is calculated by multiplying the local weight coefficient by (1 + average similarity multiplied by 0.3). For a region with a local weight coefficient of 0.0195 and an average similarity of 0.85, the global weight coefficient is 0.022. The global weight coefficient considers both the importance of the region itself and the correlation between regions, making the weight allocation more reasonable. Regions with similar features to their neighbors receive higher global weights, which is beneficial for discovering continuously distributed defects or characteristics.

[0105] In the actual testing of TC4 titanium alloy plates, after calculating the global weight coefficients for 100 test areas, normalization is required to ensure that the sum of all weight coefficients is 1. The normalization method involves dividing the global weight coefficient of each area by the sum of the global weight coefficients of all areas. The normalized global weight coefficients are directly used in the subsequent feature weighting and fusion process. Areas with higher global weight coefficients have a greater impact on the fusion result, thus achieving the goal of differentiated weighting based on regional importance.

[0106] This method introduces a dual dispersion evaluation mechanism of scale dispersion and directional dispersion, combined with temporal stability analysis and regional similarity calculation, to achieve accurate assessment and weight allocation of the importance of the test area. This method effectively identifies and highlights key areas and anomalous features on titanium alloy surfaces, significantly improving the targeting and accuracy of feature fusion. It provides reliable technical support for the quality assessment and defect early warning of titanium alloy plates, and solves the technical problems of unreasonable weight allocation and insufficient consideration of regional correlation in traditional methods.

[0107] Optionally, based on the fused features, the titanium plate performance analysis results are generated through a multi-task output layer, and the step of determining the performance level of the titanium plate under test based on the performance analysis results includes:

[0108] The fused features are input into a shared feature extraction layer and a task-specific layer. The shared feature extraction layer performs multi-scale hierarchical extraction on the fused features to generate multi-scale features. The task-specific layer includes a performance distribution prediction branch, a defect type identification branch, and a reliability assessment branch.

[0109] In the performance distribution prediction branch, the spatial distribution data of the multi-scale features is calculated, the distribution interval of the performance index is determined based on the spatial distribution data, the distribution interval is dynamically updated, and a performance distribution prediction result is generated. In the defect type identification branch, the multi-scale features are matched layer by layer with preset defect features to determine the optimal matching feature, and the defect type and its corresponding confidence level are output based on the optimal matching feature. In the reliability assessment branch, the performance distribution prediction result and the defect type and its confidence level are used as assessment criteria, the credibility of each assessment criterion is calculated, and the assessment criteria are weighted and combined based on the credibility to obtain the reliability assessment result.

[0110] Based on the pre-set performance level determination criteria, combined with the performance distribution prediction results, the defect types and their confidence levels, and the reliability assessment results, the performance level of the titanium plate to be tested is determined through interval mapping.

[0111] For example, when the fused features are input into the shared feature extraction layer and the task-specific layer, the shared feature extraction layer adopts a cascaded structure to perform multi-scale hierarchical extraction of the fused features. For instance, for a typical TC4 titanium alloy sheet, the fused feature dimension is 16, and it first enters the shared feature extraction layer. The shared feature extraction layer contains three cascaded feature extraction modules, each consisting of a nonlinear activation unit and a normalization unit. The first-level module maps the 16-dimensional features to 32-dimensional features, the second-level module maps the 32-dimensional features to 64-dimensional features, and the third-level module maps the 64-dimensional features to 128-dimensional features. This cascaded structure can progressively extract abstract representations of features, generating multi-scale features. The multi-scale features include micro-scale features (corresponding to the output of the first-level module), meso-scale features (corresponding to the output of the second-level module), and macro-scale features (corresponding to the output of the third-level module). Different scale features can capture the performance characteristics of the titanium sheet at different spatial scales.

[0112] The task-specific layer includes a performance distribution prediction branch, a defect type identification branch, and a reliability assessment branch, each designed and optimized for a specific task. The performance distribution prediction branch consists of a feature space transformation module and a distribution parameter estimation module. The feature space transformation module employs an attention mechanism, weighting multi-scale features according to their importance to highlight key features. The distribution parameter estimation module contains two parallel processing units, which estimate the central tendency parameter and the dispersion parameter of the distribution, respectively. For TC4 titanium alloy sheet, the main performance indicators include tensile strength, yield strength, and elongation; the distribution parameters for each indicator are estimated by the corresponding processing unit.

[0113] In the performance distribution prediction branch, the spatial distribution data of multi-scale features are calculated. Spatial distribution data is obtained through the distribution characteristics of statistical features in the spatial domain, including mean distribution, variance distribution, and skewness distribution. For a 1m × 1m TC4 titanium alloy sheet, it is divided into a 10×10 grid, with each grid representing a spatial cell. Within each spatial cell, the statistics of multi-scale features are calculated to form a spatial distribution matrix. The mean distribution matrix reflects the average performance level of each spatial cell, the variance distribution matrix reflects the degree of performance fluctuation, and the skewness distribution matrix reflects the asymmetry of the performance distribution. Based on the spatial distribution data, the distribution range of performance indicators is determined, expressed using confidence intervals, typically set at 95% confidence level. For the tensile strength of TC4 titanium alloy, the typical distribution range is 920-980 MPa; the yield strength distribution range is 830-890 MPa; and the elongation distribution range is 10%-14%.

[0114] The distribution interval is dynamically updated, taking into account historical data and current measurement results. The update method uses an exponentially weighted average: the new lower limit of the distribution interval is equal to the original lower limit multiplied by 0.9 plus the current minimum measured value multiplied by 0.1; the new upper limit of the distribution interval is equal to the original upper limit multiplied by 0.9 plus the current maximum measured value multiplied by 0.1. This dynamic update mechanism can adapt to the changing trends of titanium plate performance, improving prediction accuracy. The process of generating performance distribution prediction results involves combining the updated distribution interval with statistical characteristic data to output complete performance prediction data. Specifically, the statistical characteristics of the spatial distribution matrix are used to calculate parameters such as the mean, standard deviation, skewness, and kurtosis of the distribution, and then the most suitable distribution type (such as normal distribution, Weibull distribution, or mixed distribution) is determined. For TC4 titanium alloy, the calculated tensile strength has a mean of 950 MPa and a standard deviation of 15 MPa, indicating a normal distribution; the yield strength has a mean of 860 MPa and a standard deviation of 12 MPa, also indicating a normal distribution; and the elongation has a mean of 12%, a shape parameter of 3.2, and a dimensional parameter of 13.5, indicating a Weibull distribution. The final performance distribution prediction results include these distribution parameters, distribution types, and predicted out-of-limit probabilities, collectively constituting the comprehensive prediction results for the titanium plate properties. For most TC4 titanium alloy plates, tensile strength and yield strength typically follow a normal distribution, while elongation is closer to a Weibull distribution.

[0115] In the defect type identification branch, multi-scale features are matched layer by layer with preset defect features. The preset defect feature library contains common defects of typical titanium alloy plates, such as surface cracks, delamination, porosity, and inclusions, each with corresponding feature descriptions. The layer-by-layer matching process adopts a hierarchical comparison strategy: first, coarse matching is performed at the macro scale to filter out candidate defect types; then, fine matching is performed at the meso scale to narrow down the candidate range; finally, precise matching is performed at the micro scale to determine the final defect type. The matching degree is calculated using cosine similarity; a higher cosine similarity indicates a higher degree of matching. The optimal matching feature is determined by selecting the preset defect feature with the highest cosine similarity to the sample features.

[0116] The defect type and its corresponding confidence level are output based on the best matching feature. The confidence level is calculated by subtracting the cosine similarity of the second best match from the cosine similarity of the best match, and then multiplying by an adjustment factor of 1.5. A higher confidence level is achieved when the best match differs significantly from other matches; a lower confidence level is achieved when multiple matching results are similar. For typical TC4 titanium alloy sheet defects, such as surface cracks, a confidence level above 0.85 can be achieved when the feature matching degree is high; for ambiguous defect features, the confidence level drops to below 0.6.

[0117] In the reliability assessment branch, performance distribution prediction results and defect types, along with their confidence levels, are used as the assessment criteria. The reliability assessment branch consists of a confidence calculation module and a weighted combination module. The confidence calculation module evaluates the reliability of each assessment criterion, with the calculation method varying depending on the criterion type. For performance distribution prediction results, confidence is inversely proportional to the width of the distribution interval and directly proportional to the sample size; for defect types and confidence levels, confidence is related to the typicality and stability of the defect characteristics. Confidence calculation employs an evidence-based approach, comprehensively considering multiple factors.

[0118] The reliability assessment result is obtained by weighting and combining the various assessment criteria according to their credibility. The weighted combination uses a method of multiplying the performance distribution credibility by 0.6 and the defect type credibility by 0.4, balancing the influence of both performance and defects. For TC4 titanium alloy plates with stable performance and few defects, the reliability assessment result is usually between 0.85 and 0.95; for plates with large performance fluctuations or obvious defects, the reliability assessment result is as low as below 0.7. The reliability assessment result is an important indicator for judging the overall quality of titanium plates and directly affects their final performance level.

[0119] The interval mapping process first establishes a correspondence table between each evaluation index and performance level. For the performance distribution prediction results, the distribution characteristics of three indices—tensile strength, yield strength, and elongation—are mainly considered, dividing them into high, medium, and low intervals, each corresponding to a different performance score. For TC4 titanium alloy, the high interval for tensile strength (above 960 MPa) scores 5 points, the medium interval (930-960 MPa) scores 3 points, and the low interval (900-930 MPa) scores 1 point; yield strength and elongation are divided similarly. Regarding defect type and confidence level, common defects are divided into three categories according to severity: severe defects (e.g., deep cracks), medium defects (e.g., surface scratches), and minor defects (e.g., minor oxidation), and defect scores are calculated based on the confidence level. A defect confidence level above 0.8 has a weight of 1, between 0.5 and 0.8 has a weight of 0.7, and below 0.5 has a weight of 0.4. The reliability assessment results are directly used as weighting factors, multiplied by the performance score and defect score to obtain the weighted total score. The performance level is ultimately determined by the total score range: 8-10 points is Grade A, 5-8 points is Grade B, 3-5 points is Grade C, and below 3 points is unqualified.

[0120] This method's multi-task output layer integrates three key functions: performance distribution prediction, defect type identification, and reliability assessment. By combining shared feature extraction and task-specific processing, it achieves comprehensive analysis and accurate rating of titanium plate performance. This method not only accurately predicts the performance distribution of titanium plates but also effectively identifies various defects and assesses the reliability of the analysis results. It provides a scientific basis for the quality control and performance assurance of titanium alloy plates, solving technical problems such as insufficient multi-task collaboration and single rating standards in traditional methods.

[0121] Optionally, the steps of using the performance distribution prediction results and the defect types and their confidence levels as evaluation criteria, calculating the credibility of each evaluation criterion, and weighting and combining the evaluation criteria according to the credibility to obtain the reliability evaluation results include:

[0122] The probability distribution deviation and the prediction error of the defect type confidence level of the performance distribution prediction results are calculated, and an uncertainty matrix is ​​constructed based on the probability distribution deviation and the prediction error. The temporal fluctuation characteristics of each evaluation criterion are calculated according to the uncertainty matrix to obtain a stability index for the evaluation criterion. Based on the stability index, the historical evaluation results of the evaluation criterion are tracked and analyzed to calculate the accuracy change trend of the evaluation results, and dynamic evaluation weights are generated based on the accuracy change trend. The stability index and the dynamic evaluation weights are combined to obtain the credibility of each evaluation criterion. The performance distribution prediction results and the defect type and its confidence level are weighted and fused based on the credibility to output a reliability evaluation result.

[0123] For example, calculating the probability distribution deviation of the performance distribution prediction results requires comparing the predicted distribution with a reference standard distribution. For TC4 titanium alloy plates, the reference standard distribution is typically established based on a large amount of historical test data. The probability distribution deviation is obtained by calculating the difference between the predicted distribution and the reference distribution, specifically by summing the absolute values ​​of the frequency differences between the two distributions in each probability interval. In practical applications, the tensile strength distribution range of 920-980 MPa is divided into 12 equally spaced intervals, each with a width of 5 MPa, and the frequency differences between the predicted and reference distributions in these 12 intervals are calculated. For TC4 titanium alloys with stable performance, the probability distribution deviation is typically less than 0.15; for samples with large performance fluctuations, the deviation can reach 0.3 or higher.

[0124] The prediction error of defect type confidence is calculated by comparing the predicted defect type and confidence with the actual verification result. The prediction error is calculated as the absolute value of the difference between the predicted confidence and the actual confidence (determined through expert evaluation or destructive testing). For example, for a defect predicted as a surface crack with a confidence of 0.85, if the actual verification confirms it as a surface crack with the same severity as predicted, the actual confidence is 0.9, and the prediction error is 0.05; if the actual verification confirms it as a surface scratch rather than a crack, the actual confidence is only 0.3, and the prediction error is 0.55. For accurate defect predictions, the prediction error is typically less than 0.1; for inaccurate predictions, the error exceeds 0.4.

[0125] An uncertainty matrix is ​​constructed based on probability distribution bias and prediction error. The uncertainty matrix is ​​a two-dimensional matrix where rows represent different evaluation criteria (such as the distribution prediction of tensile strength, yield strength, elongation, and various defects), and columns represent different sources of uncertainty (such as measurement error, model error, environmental interference, etc.). For the performance evaluation of TC4 titanium alloy plates, the uncertainty matrix is ​​typically a 5×4 matrix, containing 5 evaluation criteria and 4 sources of uncertainty. The value of each matrix element represents the error contribution of the corresponding evaluation criterion under a specific uncertainty source; a larger value indicates higher uncertainty. The uncertainty matrix is ​​constructed by decomposing probability distribution bias and prediction error into different sources of uncertainty, forming the matrix element values.

[0126] The temporal fluctuation characteristics of each assessment criterion are calculated based on the uncertainty matrix. These characteristics describe the changes in the assessment criterion across multiple consecutive assessments. The calculation method involves recording the change in uncertainty value for each assessment criterion over 10 consecutive assessments and calculating the ratio of its standard deviation to the mean. For stable assessment criterions, this ratio is typically less than 0.2; for unstable criterions, the ratio exceeds 0.5. The stability index is a normalized representation of the temporal fluctuation characteristics, calculated by subtracting the fluctuation ratio from 1. For the tensile strength prediction of TC4 titanium alloy, if the mean uncertainty value of 10 consecutive assessments is 0.12 and the standard deviation is 0.02, the fluctuation ratio is 0.167 and the stability index is 0.833. For surface defect prediction, if the mean uncertainty value is 0.25 and the standard deviation is 0.1, the fluctuation ratio is 0.4 and the stability index is 0.6.

[0127] The process of tracking and analyzing historical evaluation results based on stability indices requires establishing a historical evaluation database to record the results of each evaluation and its actual verification. For each evaluation criterion, its accuracy rate within different stability index ranges is calculated. Accuracy rate is the number of correct evaluations divided by the total number of evaluations. For example, for tensile strength prediction, when the stability index is between 0.8 and 0.9, the historical accuracy rate is 0.92; when the stability index is below 0.7, the historical accuracy rate drops to 0.75. The accuracy rate trend is an estimate of the derivative of accuracy rate with respect to the stability index, indicating the degree of influence of stability changes on accuracy rate. For the performance evaluation of TC4 titanium alloy, the accuracy rate trend is typically between 0.5 and 1.5, with higher values ​​indicating a greater impact of stability on accuracy rate.

[0128] The dynamic evaluation weight is a weight adjustment factor calculated based on the current stability index and the accuracy trend. The calculation method is to multiply the stability index by an adjustment function representing the accuracy trend. This adjustment function is a piecewise linear function, providing a larger weight adjustment when the accuracy trend is significant. For tensile strength prediction of TC4 titanium alloy, if the stability index is 0.85 and the accuracy trend is 1.2, the dynamic evaluation weight is 0.92; for defect prediction, if the stability index is 0.65 and the accuracy trend is 0.8, the dynamic evaluation weight is 0.75.

[0129] The reliability of each assessment basis is obtained by multiplying the stability index by 0.4 and adding the dynamic evaluation weight by 0.6. For tensile strength prediction, if the stability index is 0.85 and the dynamic evaluation weight is 0.92, the reliability is 0.892; for defect prediction, if the stability index is 0.65 and the dynamic evaluation weight is 0.75, the reliability is 0.71. Reliability reflects the reliability of the assessment basis and directly affects its weight in the final assessment.

[0130] The performance distribution prediction results and defect types and their confidence levels are weighted and fused based on their credibility to output a reliability assessment result. The weighted fusion process considers the importance and credibility of different assessment criteria. For TC4 titanium alloy plates, the basic weight of the performance distribution prediction results is typically set to 0.6, and the basic weight of defect types and confidence levels is set to 0.4. These basic weights are then multiplied by their corresponding credibility levels to obtain the actual weights. For example, if the credibility of the performance distribution prediction results is 0.892, and the credibility of defect types and confidence levels is 0.71, then the actual weight of the performance distribution prediction results is 0.6 × 0.892 = 0.5352, and the actual weight of defect types and confidence levels is 0.4 × 0.71 = 0.284, which, after normalization, are 0.653 and 0.347 respectively. The reliability assessment result is a weighted sum of the scores of each assessment criterion multiplied by their corresponding actual weights. For high-quality TC4 titanium alloy plates, the performance distribution prediction score is 0.95, the defect type prediction score is 0.9, and the calculated reliability assessment result is 0.95×0.653+0.9×0.347=0.933; for plates with certain problems, the scores are 0.8 and 0.6 respectively, and the reliability assessment result is 0.8×0.653+0.6×0.347=0.731.

[0131] The reliability assessment results include an overall reliability score and a credibility and contribution analysis of each assessment criterion. The overall reliability score is a value between 0 and 1, with scores closer to 1 indicating higher reliability of the titanium plate. For aerospace-grade TC4 titanium alloy, an overall reliability score of no less than 0.9 is required; for general industrial applications, a score of no less than 0.75 is required. The credibility analysis demonstrates the reliability of each assessment criterion, while the contribution analysis shows the degree of influence of each assessment criterion on the overall assessment. This information collectively constitutes the reliability assessment report for the titanium alloy plate performance, providing a scientific basis for quality control and application decisions.

[0132] This method, by introducing probability distribution bias and prediction error analysis, combined with time-series fluctuation characteristics and accuracy change trends, achieves accurate calculation and dynamic adjustment of the reliability of the assessment basis. This method not only considers static assessment results but also focuses on the temporal stability and historical accuracy of the assessment process, effectively solving the technical problems of traditional methods' singular, static, and highly subjective reliability assessments. It provides a more scientific and comprehensive reliability guarantee for the quality assessment of titanium alloy plates.

[0133] Secondly, a titanium plate performance analysis system is provided, including:

[0134] The first unit uses an optical scanner to scan the surface of the titanium plate under test to establish a three-dimensional structure and divide it into multiple test areas. Multiple test points are set in a matrix distribution within each test area. The second unit uses a multi-probe collaborative detection device to perform performance testing on each test point. This device includes stress probes, hardness probes, and conductivity probes uniformly arranged along the circumference, simultaneously collecting stress, hardness, and conductivity data from the test points. The third unit constructs a dual-branch feature extraction network, including a planar feature branch for extracting surface morphology features and a physical feature branch based on materials science knowledge. The physical feature branch includes a physical constraint layer and a physical correlation layer, used to establish the correlation between stress, hardness, and conductivity data. It calculates the distribution characteristics of the surface morphology and physical features in each test area, determines the corresponding feature weight coefficients based on these distribution characteristics, and performs weighted fusion of the features. The fourth unit generates titanium plate performance analysis results through a multi-task output layer based on the fused features, and determines the performance level of the titanium plate under test based on these performance analysis results.

[0135] Thirdly, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

Claims

1. A method for analyzing the performance of titanium plates, characterized in that, include: An optical scanner was used to scan the surface of the titanium plate under test to establish a three-dimensional structure and divide it into multiple test areas. Multiple test points were set in a matrix distribution within each test area. A multi-probe collaborative testing device is used to perform performance testing on each test point. The multi-probe collaborative testing device includes stress probes, hardness probes and conductivity probes that are uniformly arranged along the circumference, and simultaneously collects stress data, hardness data and conductivity data of the test points. A dual-branch feature extraction network is constructed, including a planar feature branch for extracting surface morphology features and a physical feature branch based on materials science knowledge. The physical feature branch includes a physical constraint layer and a physical correlation layer, used to establish the correlation between stress data, hardness data, and conductivity data. The distribution characteristics of the surface morphology features and physical features in each test area are calculated, and the corresponding feature weight coefficients are determined based on the distribution characteristics. The features are then weighted and fused, specifically including: calculating the local curvature value and height distribution probability based on the height data of the titanium plate surface to be tested, and generating morphology entropy features based on the local curvature value and height distribution probability; collecting the height deviation data of the titanium plate surface to be tested and calculating the surface roughness parameters; and combining the morphology entropy features with the physical features. The surface roughness parameters are constructed as surface morphology features; the hardness variation law corresponding to the stress data is analyzed to establish the correspondence between stress and hardness; the variation law of electrical conductivity data with dislocation density is analyzed to establish the correspondence between electrical conductivity and dislocation density; the correspondence between stress and hardness, and the correspondence between electrical conductivity and dislocation density are used as physical constraint features; an interaction matrix of stress, hardness, and electrical conductivity is constructed, and the interaction strength of the stress data, hardness data, and electrical conductivity data is calculated based on the interaction matrix; the stress data, hardness data, and electrical conductivity data are corrected based on the interaction strength, and the corrected data are constructed as physical correlation features; the morphology entropy spatial distribution of the surface morphology features in each test area is calculated. The morphology entropy spatial distribution is calculated by the negative logarithm of the surface morphology feature probability within the test area; the multi-physical quantity correlation strength of the physical feature in each test area is calculated, and the multi-physical quantity correlation strength is calculated based on the correlation coefficient matrix between physical quantities; the dispersion coefficient and anomaly detection index within the test area are calculated based on the morphology entropy spatial distribution and the multi-physical quantity correlation strength; the regional importance is obtained by weighting the dispersion coefficient and the multi-physical quantity correlation strength, and the feature reliability is calculated based on the anomaly detection index; the local weight coefficient of each test area is calculated according to the regional importance and the feature reliability, and the global weight coefficient is calculated according to the local weight coefficient; the surface weight coefficient within each test area is then calculated based on the local weight coefficient. Local fusion features are obtained by weighting morphological and physical features, and global fusion features are obtained by weighting the local fusion features based on the global weight coefficients. Specifically, the dispersion coefficient and anomaly detection index within the test area are calculated based on the spatial distribution of morphological entropy and the correlation strength of multiple physical quantities. This includes: calculating the scale dispersion coefficient within the test area based on the spatial distribution of morphological entropy, and calculating the directional dispersion coefficient based on the correlation strength of multiple physical quantities. The scale dispersion coefficient is calculated by weighting the ratio of the standard deviation to the mean at multiple feature scales, and the directional dispersion coefficient is obtained by normalizing the deviation of the feature gradient from the mean. The scale dispersion coefficient and the directional dispersion coefficient are then weighted and combined to obtain a comprehensive dispersion coefficient.Feature anomaly scores are calculated based on the spatial distribution of morphological entropy and the correlation strength of multiple physical quantities. Temporal weights are updated based on the temporal changes of the feature anomaly scores. Anomaly detection index is obtained by adjusting the feature anomaly scores based on the temporal weights. The regional importance of each test region is obtained by combining the comprehensive dispersion coefficient with the anomaly detection index. The feature weights of the test regions are updated based on the regional importance to obtain the local weight coefficients of the test regions. The local weight coefficients are weighted and combined with the feature similarity of adjacent test regions to obtain the global weight coefficients. The weight coefficients of the global weight coefficients are proportional to the feature similarity of adjacent test regions. Based on the fused features, the performance analysis results of the titanium plate are generated through a multi-task output layer, and the performance level of the titanium plate under test is determined according to the performance analysis results.

2. The method according to claim 1, characterized in that, The steps of using an optical scanner to scan the surface of the titanium plate under test to establish a three-dimensional structure and dividing it into multiple test areas, and setting multiple test points in a matrix distribution within each test area, include: Image data of the surface of the titanium plate under test is acquired using a dual-camera collaborative acquisition system; initial three-dimensional point cloud data is obtained by phase measurement based on the image data. The vibration error in the initial three-dimensional point cloud data is eliminated by the birefringence optical path difference method, and the temperature drift error in the initial three-dimensional point cloud data is compensated based on the pre-calibrated temperature-deformation curve to obtain the compensated three-dimensional point cloud data. Based on the compensated three-dimensional point cloud data, the local curvature distribution data of the surface of the titanium plate to be tested is calculated. According to the local curvature distribution data, the surface of the titanium plate to be tested is divided into multiple test areas, wherein the area of ​​each test area is inversely proportional to the local curvature value within the test area. Based on the local curvature distribution data, a test point matrix is ​​set in each test area, wherein the spacing between adjacent test points in each test area is inversely proportional to the local curvature value at the test point location.

3. The method according to claim 1, characterized in that, The steps for performing performance testing at each test point using a multi-probe collaborative testing device include: The stress release characteristic time of the titanium plate to be tested is obtained, and the measurement sequence of the stress probe, hardness probe and conductivity probe is determined based on the stress release characteristic time. During the stress probe measurement process, the load application rate is controlled to be proportional to the yield strength of the titanium plate under test, so as to obtain the stress data at the test point. During the conductivity probe measurement process, the initial conductivity data of the test point is acquired, the perturbation correction coefficient of the stress field on the conductivity is calculated based on the stress data, and the initial conductivity data is corrected based on the perturbation correction coefficient to obtain the conductivity data. During the hardness probe measurement process, the initial hardness data of the test point is acquired, the deformation compensation factor of the hardness measurement on the stress field is calculated based on the indentation depth, and the initial hardness data is corrected based on the deformation compensation factor to obtain the hardness data; the stress data and the conductivity data are corrected based on the deformation compensation factor to obtain the corrected stress data and conductivity data of the test point.

4. The method according to claim 1, characterized in that, Based on the fused features, the titanium plate performance analysis results are generated through a multi-task output layer. The steps for determining the performance level of the titanium plate under test based on the performance analysis results include: The fused features are input into a shared feature extraction layer and a task-specific layer. The shared feature extraction layer performs multi-scale hierarchical extraction on the fused features to generate multi-scale features. The task-specific layer includes a performance distribution prediction branch, a defect type identification branch, and a reliability assessment branch. In the performance distribution prediction branch, the spatial distribution data of the multi-scale features is calculated, the distribution interval of the performance index is determined based on the spatial distribution data, the distribution interval is dynamically updated, and a performance distribution prediction result is generated. In the defect type identification branch, the multi-scale features are matched layer by layer with preset defect features to determine the optimal matching feature, and the defect type and its corresponding confidence level are output based on the optimal matching feature. In the reliability assessment branch, the performance distribution prediction result and the defect type and its confidence level are used as assessment criteria, the credibility of each assessment criterion is calculated, and the assessment criteria are weighted and combined based on the credibility to obtain the reliability assessment result. Based on the pre-set performance level determination criteria, combined with the performance distribution prediction results, the defect types and their confidence levels, and the reliability assessment results, the performance level of the titanium plate to be tested is determined through interval mapping.

5. The method according to claim 4, characterized in that, The steps of using the performance distribution prediction results and the defect types and their confidence levels as evaluation criteria, calculating the confidence level of each evaluation criterion, and weighting and combining the evaluation criteria according to the confidence levels to obtain the reliability evaluation results include: The probability distribution deviation and the prediction error of the defect type confidence level of the performance distribution prediction results are calculated, and an uncertainty matrix is ​​constructed based on the probability distribution deviation and the prediction error. The temporal fluctuation characteristics of each evaluation criterion are calculated according to the uncertainty matrix to obtain a stability index for the evaluation criterion. Based on the stability index, the historical evaluation results of the evaluation criterion are tracked and analyzed to calculate the accuracy change trend of the evaluation results, and dynamic evaluation weights are generated based on the accuracy change trend. The stability index and the dynamic evaluation weights are combined to obtain the credibility of each evaluation criterion. The performance distribution prediction results and the defect type and its confidence level are weighted and fused based on the credibility to output a reliability evaluation result.

6. A titanium plate performance analysis system for implementing the method of any one of claims 1-5, characterized in that, include: The first unit is used to scan the surface of the titanium plate to be tested using an optical scanner to establish a three-dimensional structure and divide it into multiple test areas, and set multiple test points in a matrix distribution in each test area. The second unit is used to perform performance testing on each test point using a multi-probe collaborative testing device. The multi-probe collaborative testing device includes stress probes, hardness probes, and conductivity probes that are uniformly arranged along the circumference, and simultaneously collects stress data, hardness data, and conductivity data of the test points. The third unit is used to construct a dual-branch feature extraction network, including a planar feature branch for extracting surface morphology features and a physical feature branch constructed based on materials science knowledge. The physical feature branch includes a physical constraint layer and a physical correlation layer, which are used to establish the correlation between stress data, hardness data and electrical conductivity data. The distribution characteristics of the surface morphology features and physical features in each test area are calculated, and the corresponding feature weight coefficients are determined according to the distribution characteristics to perform weighted fusion of the features. The fourth unit is used to generate titanium plate performance analysis results through a multi-task output layer based on the fused features, and to determine the performance level of the titanium plate under test based on the performance analysis results.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images