An eddy current material detection method and system based on linear discriminant analysis

Abstract

The present application relates to the technical field of nondestructive testing, in particular to an eddy current material detection method and system based on linear discriminant analysis. The method comprises: obtaining resistance and inductance characteristic data of a standard sample set under an eddy current sensor; calculating mean vector and intra-class scatter matrix of each sample; using linear discriminant analysis (LDA) algorithm to solve an optimal projection direction vector and determine a classification judgment threshold; in an online detection stage, collecting impedance data of a sample to be detected in real time, projecting the impedance data to the optimal projection direction to obtain a one-dimensional characteristic value, and realizing material classification by comparing the one-dimensional characteristic value with the judgment threshold. The present application maximizes inter-class dispersion and minimizes intra-class dispersion, effectively overcomes problems such as low efficiency, sensitivity to probe lift-off and temperature drift of a traditional geometric partition method, compresses two-dimensional characteristics into a one-dimensional threshold judgment, greatly reduces calculation complexity, and realizes high-speed and high-robustness material detection at a microsecond level.

Description

Technical Field

[0001] This invention relates to the fields of nondestructive testing and signal processing technology, and in particular to an eddy current material detection method and system that combines the Linear Discriminant Analysis (LDA) algorithm, which is mainly used for high-speed discrimination of metal material classification, alloy grade identification and heat treatment state. Background Technology

[0002] Eddy current testing (ECT) is a widely used non-destructive testing method. Its basic principle is to use a probe coil to generate an alternating magnetic field, inducing eddy currents within a conductive material. The intensity and distribution of these eddy currents are determined by the material's electromagnetic properties (such as resistivity). magnetic permeability The complex impedance of the coil (including its equivalent resistance) is determined by the interaction between the reverse magnetic field generated by the eddy current and the original magnetic field of the coil. and equivalent inductance A characteristic change occurs. Theoretically, the properties of a material can be determined by detecting this change in impedance.

[0003] However, existing traditional eddy current material testing methods have significant limitations in practical industrial applications:

[0004] 1) Low efficiency and accuracy of the discrimination model: Traditional methods often use rule-based geometric partitioning (e.g., manually defining rectangular or sector-shaped regions on the RL complex plane). For materials with similar electromagnetic properties (such as different grades of aluminum alloys), the points where their impedance signals fall on the complex plane are extremely close or even overlap, making it difficult for geometric partitioning to distinguish them accurately, and the computational logic is inefficient under complex shapes.

[0005] 2) Insufficient robustness against interference: The detection signal is highly susceptible to non-material factors, such as slight changes in probe lift-off, ambient temperature drift, and material surface roughness. These interferences can cause the signal point to shift on the complex plane, thus sliding out of the preset area and resulting in misjudgment.

[0006] 3) Stringent sampling conditions: In automated production lines, materials are usually in a state of variable speed movement. Traditional methods often rely on capturing the "peak" of the signal waveform or the characteristic point of a specific phase. If the fluctuation of the material speed causes the sampling point to deviate from the peak, it is very easy to cause waveform distortion and detection failure.

[0007] Linear Discriminant Analysis (LDA), a classic supervised learning algorithm, projects high-dimensional data into a low-dimensional space through linear transformation, achieving both maximum inter-class scatter and minimum intra-class scatter. While LDA is maturely applied in fields such as face recognition, its use in eddy current detection, particularly in addressing feature compression and anti-interference issues on the impedance plane, has not yet been fully developed and applied. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing an eddy current material detection method and system based on linear discriminant analysis, which aims to solve the problems of difficulty in distinguishing similar materials, weak anti-interference energy, and sensitivity to motion state in the prior art.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] An eddy current material testing method based on linear discriminant analysis is applied to an eddy current material testing system. The system acquires impedance characteristic data of standard samples using eddy current sensors. The standard samples include at least two categories. The method includes:

[0011] Impedance characteristic data of standard samples of different material types under the action of an eddy current sensor are collected. The impedance characteristic data includes equivalent resistance and equivalent inductance.

[0012] Based on the impedance characteristic data, the mean vector, global mean vector, and total intraclass scatter matrix of various standard samples are calculated, and the optimal projection direction vector for distinguishing the different material types is solved by linear discriminant analysis algorithm.

[0013] Based on the optimal projection direction vector, calculate the projection center value of each type of standard sample in the optimal projection direction, and determine at least one classification threshold based on the projection center value;

[0014] Obtain the real-time impedance characteristic data of the sample to be tested, and calculate the projection value of the real-time impedance characteristic data based on the optimal projection direction vector;

[0015] The projection value is compared with the classification threshold to determine the material category of the sample to be tested.

[0016] The various standard samples include a first type of standard sample and a second type of standard sample. Based on impedance characteristic data, the mean vector and total within-class scatter matrix of each type of standard sample are calculated, and the optimal projection direction vector for distinguishing the different material types is solved using a linear discriminant analysis algorithm, including:

[0017] The impedance feature data is constructed into a two-dimensional feature vector, where the first dimension is the equivalent resistance feature and the second dimension is the equivalent inductance feature.

[0018] Calculate the mean vector of the first type of standard samples and the mean vector of the second type of standard samples, respectively;

[0019] Calculate the covariance matrix of the first type of standard samples and the covariance matrix of the second type of standard samples, respectively;

[0020] Add the covariance matrix of the first class of standard samples to the covariance matrix of the second class of standard samples to generate the total within-class scatter matrix;

[0021] The optimal projection direction vector is solved based on the difference between the total in-class scatter matrix and the mean vectors of the first and second class standard samples.

[0022] The step of solving for the optimal projection direction vector based on the difference between the total in-class scatter matrix and the mean vectors of the first and second class standard samples includes:

[0023] Calculate the difference vector between the mean vector of the first type of standard samples and the mean vector of the second type of standard samples;

[0024] Perform matrix inversion on the total intra-class scatter matrix to generate the intra-class scatter inverse matrix;

[0025] The inverse of the intraclass divergence matrix is ​​multiplied by the difference vector to generate the optimal projection direction vector, wherein the optimal projection direction vector includes a first component and a second component, the first component corresponds to the projection weight of the equivalent resistance feature, and the second component corresponds to the projection weight of the equivalent inductance feature.

[0026] The step of calculating the projection center value of each type of standard sample in the optimal projection direction based on the optimal projection direction vector, and determining at least one classification threshold based on the projection center value, includes:

[0027] The mean vector of the first type of standard samples is multiplied by the optimal projection direction vector to obtain the projection center value of the first type.

[0028] The mean vector of the second type of standard samples is multiplied by the optimal projection direction vector to obtain the projection center value of the second type.

[0029] Calculate the arithmetic mean of the projection center values ​​of the first class and the projection center values ​​of the second class to generate a classification threshold.

[0030] The optimal projection direction vector includes a first component and a second component. The step of acquiring the real-time impedance characteristic data of the sample under test and calculating the projection value of the real-time impedance characteristic data based on the optimal projection direction vector includes:

[0031] Real-time equivalent resistance and real-time equivalent inductance data of the sample under test are collected using an eddy current sensor.

[0032] The real-time equivalent resistance data is multiplied with the first component of the optimal projection direction vector to generate the first projection component;

[0033] The real-time equivalent inductance data is multiplied with the second component of the optimal projection direction vector to generate the second projection component;

[0034] The first projection component is added to the second projection component to generate a one-dimensional projection value of the sample to be tested.

[0035] The various standard samples include a first type of standard sample and a second type of standard sample. The step of comparing the projection value with the classification threshold to determine the material category of the sample to be tested includes:

[0036] Determine whether the projection value of the sample to be tested is greater than or equal to the classification threshold, wherein the projection value is a one-dimensional projection value;

[0037] When the projection value is greater than or equal to the classification threshold, the sample to be tested is determined to belong to the first type of material, and the first type of material is the same material type as the first type of standard sample.

[0038] When the projection value is less than the classification threshold, the sample to be tested is determined to belong to the second type of material, which is the same material type as the second type of standard sample.

[0039] The impedance characteristic data of standard samples of different material types under the action of an eddy current sensor are collected, including:

[0040] Under multiple preset probe lift-off distance conditions, the equivalent resistance data and equivalent inductance data of various standard samples were collected respectively.

[0041] Under multiple preset temperature conditions, the equivalent resistance and equivalent inductance data of various standard samples were collected respectively.

[0042] Impedance characteristic data collected under different probe lift-off distances and temperature conditions are included in the training sample set. Through the calculation of the total class scatter matrix, the impedance characteristic fluctuations caused by changes in probe lift-off distance and temperature are physically modeled.

[0043] An eddy current material detection system based on linear discriminant analysis is provided for executing the aforementioned eddy current material detection method based on linear discriminant analysis. The system includes:

[0044] A sensor probe unit, the sensor probe unit including a detection coil, configured to generate an alternating magnetic field and induce eddy currents on the surface of the sample to be tested;

[0045] An impedance measurement unit is connected to the sensing probe unit and is configured to inject a high-frequency AC signal into the detection coil of the sensing probe unit and calculate the equivalent resistance and equivalent inductance of the detection coil in real time.

[0046] The processing and control unit is communicatively connected to the impedance measurement unit. It is configured to store the optimal projection direction vector and classification threshold obtained through pre-training, receive the equivalent resistance and equivalent inductance transmitted by the impedance measurement unit in real time, calculate the projection value, classify the material of the sample to be tested, and output the detection result.

[0047] The processing and control unit is configured to perform vector inner product calculation and threshold comparison within a single impedance sampling period.

[0048] The system also includes a motion transmission device configured to drive the sample to be tested through the detection area of ​​the sensor probe unit. The processing and control unit is configured to perform classification and discrimination based solely on single-point sampling data when the speed of the sample to be tested changes.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] 1) Improve the accuracy of distinguishing similar materials: By using LDA to find the best projection angle, the differences between classes are maximized, effectively overcoming the confusion caused by the overlap of impedance points of similar materials.

[0051] 2) Enhanced robustness against interference: By calculating the intraclass scatter matrix, the algorithm learns the noise distribution caused by factors such as lift-off and temperature during the training phase, and automatically suppresses the components in these directions during the projection process, so that the final one-dimensional feature is more focused on the differences in the material itself.

[0052] 3) Reduced computational complexity: The complex two-dimensional planar geometric judgment is transformed into a simple one-dimensional linear projection (vector dot product) and threshold comparison. The operation is extremely simple, and the time taken for a single detection can be reduced to the microsecond level, which is very suitable for implementation in low-cost embedded systems.

[0053] 4) Reduced sampling requirements: In binary classification applications, the system does not need to capture the complete peak value of the signal waveform. Even if the sampling position is random due to the variable speed movement of the material, as long as the sensing signal falls within the effective range, the projection discrimination can still guarantee a very high accuracy. Attached Figure Description

[0054] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0055] Figure 1 A schematic diagram of the structure of the eddy current material detection system based on linear discriminant analysis provided in an embodiment of this application;

[0056] Figure 2 A flowchart illustrating the eddy current material detection method based on linear discriminant analysis provided in this application embodiment;

[0057] Figure 3 This is a schematic diagram of the sampling modeling of the linear discriminant analysis algorithm provided in the embodiments of this application on the RL plane;

[0058] Figure 4 This is a schematic diagram illustrating the operation of online real-time detection of eddy current materials based on linear discriminant analysis, as provided in an embodiment of this application. Detailed Implementation

[0059] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0060] Example 1

[0061] Please see Figure 1 The present invention provides an embodiment of an eddy current material detection system based on linear discriminant analysis, the system comprising:

[0062] Sample (1): The metal workpiece to be tested, such as a composite metal sheet with different materials on the front and back, moves at a speed V on a conveyor belt.

[0063] Sensor probe (3): contains a detection coil (2); the detection coil (2) is placed above the sample to generate a high-frequency alternating magnetic field.

[0064] Impedance measurement circuit (4): connected to the detection coil (2), using the bridge balance principle or IQ demodulation technology, injects a high-frequency AC signal (e.g. 10kHz-1MHz) into the coil, and separates and detects the equivalent resistance R and equivalent inductance L of the coil in real time.

[0065] Controller (5): As the core processing unit, it can be a microcontroller such as STM32 or an FPGA. Its input terminal receives R and L data, runs the LDA discrimination algorithm internally, and outputs the classification result signal P (e.g., GPIO level flipping or communication message).

[0066] Example 2:

[0067] Please see Figure 2 The present invention provides an embodiment: an eddy current material detection method based on linear discriminant analysis, wherein the execution of the method can be divided into two stages: offline training and online detection.

[0068] The specific steps in the offline training phase include:

[0069] S1: Collect impedance characteristic data of standard samples of different material types under the action of eddy current sensor. The impedance characteristic data includes equivalent resistance and equivalent inductance.

[0070] For example, such as Figure 3 As shown, in this embodiment, two metal sheets of different materials were selected and labeled as Category 1 and Category 2, respectively. The horizontal axis X represents the equivalent resistance R; the vertical axis Y represents the equivalent inductance L; blue scatter dots represent Category 1 samples; orange scatter dots represent Category 2 samples; light blue polygons represent the distribution envelope (intra-class scattering) of Category 2 samples in two-dimensional space; and red diagonal lines represent the discrimination direction obtained from linear discriminant analysis (i.e., the direction corresponding to the optimal projection direction vector W).

[0071] Furthermore, to ensure the sufficiency and diversity of experimental data, a total of 200 samples were collected, with 100 samples from each category of metal sheets. To further improve the robustness and adaptability of the model, small variations were artificially introduced during data acquisition. For example, the probe height could be adjusted to fluctuate within ±0.5 mm during acquisition to simulate the impact of lift-off variations on the data. Additionally, the ambient temperature could be changed, or the samples could be subjected to a certain temperature range to increase the complexity and realism of the data. These operations help the model better cope with various uncertainties in real-world applications, thereby improving its performance and reliability.

[0072] Sample data can be represented as:

[0073]

[0074] in, The equivalent resistance R in the corresponding impedance characteristic data, The equivalent inductance L in the corresponding impedance characteristic data.

[0075] The characteristics of samples in category 1 are:

[0076]

[0077] The characteristics of samples in category 2 are:

[0078]

[0079] S2: Based on the impedance characteristic data, calculate the mean vector, global mean vector, and total intraclass scatter matrix of various standard samples, and solve for the optimal projection direction vector used to distinguish the different material types through linear discriminant analysis algorithm.

[0080] For example, based on the impedance feature data, the mean vector of each class is calculated: the mean vector reflects the central position of all samples of that class in the impedance feature space, and is the core basis for calculating inter-class differences in linear discriminant analysis.

[0081] The mean vector for category 1 is: (i.e., the arithmetic mean of the impedance characteristics of all samples in category 1)

[0082]

[0083]

[0084]

[0085] in, This represents the total number of samples in category 1. For category 1, the first The impedance characteristic vector of each sample (containing two dimensions: equivalent resistance R and equivalent inductance L). Indicates from 1 to Sum the feature vectors of all samples.

[0086] The mean vector for category 2 is: (The calculation logic is the same as for category 1)

[0087]

[0088]

[0089]

[0090] in, This represents the total number of samples in category 2. Let j be the impedance feature vector of the j-th sample in category 2. Indicates from 1 to Sum the feature vectors of all samples.

[0091] Furthermore, based on the mean vector of category 1 and the mean vector of category 2, the difference between the mean vectors of category 1 and category 2 is calculated: , The mean vector difference reflects the degree of offset between the center positions of the two classes of samples in the impedance feature space, and is one of the core indicators for measuring inter-class differences in linear discriminant analysis. The specific calculation steps are as follows: Each dimension component of the mean vector of class 1 (equivalent resistance)... Equivalent inductance The corresponding dimension components (equivalent resistance) of the mean vector of category 2. Equivalent inductance Subtracting each term one by one, the resulting difference vector is the difference between the two class mean vectors. The larger the magnitude of this vector, the more significant the difference in the feature distribution centers of the two classes of samples, and the easier it is to achieve classification through linear projection.

[0092] Furthermore, based on the impedance characteristic data, the covariance matrix of category 1 is calculated. The covariance matrix describes the dispersion of each dimension (equivalent resistance R, equivalent inductance L) of Class 1 samples in the impedance feature space and the correlation between dimensions. It is a core indicator for measuring the compactness of the intra-class feature distribution. The mathematical expression is:

[0093]

[0094] Calculate the covariance matrix S2 for category 2: (The calculation logic is the same as for category 1, and it is used to describe the dispersion of the feature distribution and the dimensionality correlation of the category 2 samples)

[0095]

[0096] Furthermore, calculate the total within-class scatter matrix for both category 1 and category 2. :

[0097]

[0098] In a physical sense, a matrix quantifies the degree of "ambiguity" in data caused by interference factors.

[0099] Furthermore, based on the core theory of Linear Discriminant Analysis (LDA), the optimal projection direction vector is solved, with the goal of finding a projection direction. The goal is to maximize the separation of projections of samples from different classes along this direction (maximizing inter-class distance) while minimizing the compactness of projections of samples from the same class (minimizing intra-class distance). The core theoretical basis of LDA is the "Fischer criterion," which determines the optimal projection direction by maximizing the ratio of inter-class divergence to intra-class divergence (objective function J(W) = inter-class divergence / intra-class divergence). For binary classification problems, the optimal projection direction... The analytical solution can be derived through matrix operations: it is the product of the inverse of the within-class scatter matrix and the vector of the difference between the means of the two classes, expressed by the formula:

[0100]

[0101] Furthermore, substituting into the inverse formula of a two-dimensional matrix (for the two characteristic dimensions of resistance R and inductance L), we can obtain... The specific analytical form (the proportionality constant that does not affect the direction can be ignored during the derivation):

[0102]

[0103] in, Corresponding equivalent resistance The projection weighting coefficient, Corresponding equivalence The projection weight coefficients. Vector Essentially, it finds an "optimal perspective" in a two-dimensional feature space. By projecting the original RL features onto this direction, it maximizes the distance between samples of different materials, providing the most discriminative one-dimensional feature for threshold classification. Its mathematical derivation is based on the associative law of matrix inversion and vector multiplication. By substituting into the formula for inverting a two-dimensional matrix (for the two feature dimensions of resistance R and inductance L), a specific analytical form can be obtained (the proportional constant that does not affect the direction can be ignored during the derivation), ultimately ensuring that the inter-class differences are maximized and the intra-class differences are minimized after projection.

[0104] S3: Based on the optimal projection direction vector, calculate the projection center value of each type of standard sample in the optimal projection direction, and determine at least one classification threshold based on the projection center value.

[0105] For example, projecting the mean vector onto Direction: This projection calculation is the core step in compressing two-dimensional impedance characteristics (equivalent resistance R, equivalent inductance L) into one-dimensional discriminant characteristics. The missing variable in the formula is the mean vector component of the two classes of standard samples—wherein... The equivalent resistance component of the mean vector of category 1. The equivalent inductance component of the category 1 mean vector; The equivalent resistance component of the category 2 mean vector. The equivalent inductance component of the category 2 mean vector; The projection weighting coefficient corresponding to the equivalent resistance R, The projection weighting coefficient corresponding to the equivalent inductance L.

[0106]

[0107]

[0108] Geometrically, this projection process involves projecting the mean vector in the two-dimensional feature space along the optimal projection direction. Perform "dimensionality reduction mapping", the projected values ​​after mapping , These represent the center positions of the two types of samples in the discrimination direction, and their difference directly reflects the inter-class separation degree.

[0109] Furthermore, the midpoint is taken as the classification threshold: the projection center value. , The midpoint is the most commonly used discrimination threshold in binary classification scenarios. Essentially, it divides the decision boundary by the midpoint between the projection centers of the two classes. When the projection value of the sample to be tested falls on one side of the midpoint, it can be determined as the material of the corresponding category. This threshold setting method can maximize the classification accuracy and simplify the calculation logic.

[0110]

[0111] Ultimately, the controller (5) only needs to save three parameters: .

[0112] The specific steps in the online detection phase include:

[0113] S4: Obtain the real-time impedance characteristic data of the sample to be tested, and calculate the projection value of the real-time impedance characteristic data based on the optimal projection direction vector.

[0114] The sample to be tested is transported to the sensor detection area along with the production line, triggering the photoelectric sensor (or proximity switch) to send a detection start signal. The sample surface maintains a preset lifting distance of 0.5-2mm from the sensor probe, and the alternating magnetic field of the coil (frequency 100kHz-2MHz) penetrates the sample surface to excite eddy currents.

[0115] The impedance measurement circuit (4) acquires the complex impedance signal of the coil in real time based on the resonance method, converts it into a digital signal through a 12-bit ADC at a sampling rate of 1MHz, and calculates the current equivalent resistance value through the DSP module. and equivalent inductance value The data update cycle is ≤10μs.

[0116] The controller performs fast linear projection calculations: The projection coefficient matrix is ​​preloaded by the LDA model during the training phase, and the computation time is ≤2μs.

[0117] S5: Compare the projection value with the classification threshold to determine the material category of the sample to be tested.

[0118] The controller performs fast linear projection calculations: The projection coefficient matrix is ​​preloaded by the LDA model during the training phase, and the computation time is ≤2μs.

[0119] Comparison and discrimination under signal peak sampling detection: A sliding window (window size 3-5 sampling points) is used to extract the peak value of the projected value. If the peak value... If the peak value is determined to be category 1; It is classified as category 2, with a determination delay of ≤5μs.

[0120] For further details, please refer to Figure 4 The graph shows the time-series waveform of impedance characteristics during eddy current material detection, as the sampling sequence number changes. The horizontal axis represents the sampling sequence number; the vertical axis represents a specific impedance characteristic (such as equivalent resistance R); the blue curve represents the real-time impedance characteristics obtained from continuous sampling; the red horizontal line represents the classification threshold; and the F0 mark indicates the signal segment corresponding to a specific material or a specific judgment result. Overall, it reflects that as the workpiece continuously passes through the sensor, different materials exhibit stable and repeatable response patterns in the time-series signal.

[0121] The continuous sampling detection comparison and discrimination process is as follows: Continuous sampling (number of sampling points ≥ 100) is performed throughout the entire process of the sample passing through the detection area (duration ≥ 1ms). If the projected values ​​of more than 90% of the sampling points are... If the projected value fluctuates beyond a preset threshold (e.g., ±5%) or does not intersect with a known category threshold, it is classified as an unknown material with a false positive rate of ≤0.1%.

[0122] The method used in this embodiment has significant engineering advantages:

[0123] 1. Calculation simplicity: Online inspection only requires two multiplications, one addition, and one comparison operation, with extremely short calculation time, which can fully meet the needs of high-speed production lines (such as inspecting thousands of workpieces per minute).

[0124] 2. Anti-interference characteristics: If the training data includes data on probe jitter, the anti-interference characteristics calculated through linear discriminant analysis (LDA) are as follows: The vector automatically selects the direction perpendicular to the jitter direction as the projection direction, thereby eliminating lift-off interference at the mathematical level, which is difficult to achieve with traditional geometric methods.

[0125] 3. Advantages of non-fixed-point sampling: Given the linear characteristics of projection transformation, as long as the sample's response under the sensor lies within the feature cluster of the RL plane, regardless of whether it is at the rising edge or the flat top region of the waveform, its projected value... It can usually be maintained on the same side of the threshold, so there is no need for strict synchronous triggering or peak hold circuitry.

[0126] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting eddy current materials based on linear discriminant analysis, applied to an eddy current material detection system, wherein the system acquires impedance characteristic data of standard samples through an eddy current sensor, the standard samples comprising at least two categories, characterized in that... The method includes: Impedance characteristic data of standard samples of different material types under the action of an eddy current sensor are collected. The impedance characteristic data includes equivalent resistance and equivalent inductance. Based on the impedance characteristic data, the mean vector, global mean vector, and total intraclass scatter matrix of various standard samples are calculated, and the optimal projection direction vector for distinguishing the different material types is solved by linear discriminant analysis algorithm. Based on the optimal projection direction vector, calculate the projection center value of each type of standard sample in the optimal projection direction, and determine at least one classification threshold based on the projection center value; Obtain the real-time impedance characteristic data of the sample to be tested, and calculate the projection value of the real-time impedance characteristic data based on the optimal projection direction vector; The projection value is compared with the classification threshold to determine the material category of the sample to be tested.

2. The method for detecting eddy current materials based on linear discriminant analysis according to claim 1, characterized in that, The various standard samples include a first type of standard sample and a second type of standard sample. Based on impedance characteristic data, the mean vector and total within-class scatter matrix of each type of standard sample are calculated, and the optimal projection direction vector for distinguishing the different material types is solved using a linear discriminant analysis algorithm, including: The impedance feature data is constructed into a two-dimensional feature vector, where the first dimension is the equivalent resistance feature and the second dimension is the equivalent inductance feature. Calculate the mean vector of the first type of standard samples and the mean vector of the second type of standard samples, respectively; Calculate the covariance matrix of the first type of standard samples and the covariance matrix of the second type of standard samples, respectively; Add the covariance matrix of the first class of standard samples to the covariance matrix of the second class of standard samples to generate the total within-class scatter matrix; The optimal projection direction vector is solved based on the difference between the total in-class scatter matrix and the mean vectors of the first and second class standard samples.

3. The eddy current material detection method based on linear discriminant analysis according to claim 2, characterized in that, The step of solving for the optimal projection direction vector based on the difference between the total in-class scatter matrix and the mean vectors of the first and second class standard samples includes: Calculate the difference vector between the mean vector of the first type of standard samples and the mean vector of the second type of standard samples; Perform matrix inversion on the total intra-class scatter matrix to generate the intra-class scatter inverse matrix; The inverse of the intraclass divergence matrix is ​​multiplied by the difference vector to generate the optimal projection direction vector, wherein the optimal projection direction vector includes a first component and a second component, the first component corresponds to the projection weight of the equivalent resistance feature, and the second component corresponds to the projection weight of the equivalent inductance feature.

4. The method for detecting eddy current materials based on linear discriminant analysis according to claim 2, characterized in that, The step of calculating the projection center value of each type of standard sample in the optimal projection direction based on the optimal projection direction vector, and determining at least one classification threshold based on the projection center value, includes: The mean vector of the first type of standard samples is multiplied by the optimal projection direction vector to obtain the projection center value of the first type. The mean vector of the second type of standard samples is multiplied by the optimal projection direction vector to obtain the projection center value of the second type. Calculate the arithmetic mean of the projection center values ​​of the first class and the projection center values ​​of the second class to generate a classification threshold.

5. The method for detecting eddy current materials based on linear discriminant analysis according to claim 1, characterized in that, The optimal projection direction vector includes a first component and a second component. The step of acquiring the real-time impedance characteristic data of the sample under test and calculating the projection value of the real-time impedance characteristic data based on the optimal projection direction vector includes: Real-time equivalent resistance and real-time equivalent inductance data of the sample under test are collected using an eddy current sensor. The real-time equivalent resistance data is multiplied with the first component of the optimal projection direction vector to generate the first projection component; The real-time equivalent inductance data is multiplied with the second component of the optimal projection direction vector to generate the second projection component; The first projection component is added to the second projection component to generate a one-dimensional projection value of the sample to be tested.

6. The method for detecting eddy current materials based on linear discriminant analysis according to claim 1, characterized in that, The various standard samples include a first type of standard sample and a second type of standard sample. The step of comparing the projection value with the classification threshold to determine the material category of the sample to be tested includes: Determine whether the projection value of the sample to be tested is greater than or equal to the classification threshold, wherein the projection value is a one-dimensional projection value; When the projection value is greater than or equal to the classification threshold, the sample to be tested is determined to belong to the first type of material, and the first type of material is the same material type as the first type of standard sample. When the projection value is less than the classification threshold, the sample to be tested is determined to belong to the second type of material, which is the same material type as the second type of standard sample.

7. The eddy current material detection method based on linear discriminant analysis according to claim 1, characterized in that, The impedance characteristic data of standard samples of different material types under the action of an eddy current sensor are collected, including: Under multiple preset probe lift-off distance conditions, the equivalent resistance data and equivalent inductance data of various standard samples were collected respectively. Under multiple preset temperature conditions, the equivalent resistance and equivalent inductance data of various standard samples were collected respectively. Impedance characteristic data collected under different probe lift-off distances and temperature conditions are included in the training sample set. Through the calculation of the total class scatter matrix, the impedance characteristic fluctuations caused by changes in probe lift-off distance and temperature are physically modeled.

8. An eddy current material detection system based on linear discriminant analysis, used to execute the eddy current material detection method based on linear discriminant analysis as described in any one of claims 1-7, characterized in that, The system includes: A sensor probe unit, the sensor probe unit including a detection coil, configured to generate an alternating magnetic field and induce eddy currents on the surface of the sample to be tested; An impedance measurement unit is connected to the sensing probe unit and is configured to inject a high-frequency AC signal into the detection coil of the sensing probe unit and calculate the equivalent resistance and equivalent inductance of the detection coil in real time. The processing and control unit is communicatively connected to the impedance measurement unit. It is configured to store the optimal projection direction vector and classification threshold obtained through pre-training, receive the equivalent resistance and equivalent inductance transmitted by the impedance measurement unit in real time, calculate the projection value, classify the material of the sample to be tested, and output the detection result.

9. The eddy current material detection system based on linear discriminant analysis according to claim 8, characterized in that, The processing and control unit includes one of an embedded microcontroller, a DSP, and an FPGA chip, configured to perform vector inner product operations and threshold comparisons within a single impedance sampling period.

10. The eddy current material detection system based on linear discriminant analysis according to claim 8, characterized in that, The system also includes a motion transmission device configured to drive the sample to be tested through the detection area of ​​the sensor probe unit. The processing and control unit is configured to perform classification and discrimination based solely on single-point sampling data when the speed of the sample to be tested changes.

Images