A method for intelligent detection and defect analysis of gas turbine blades

Abstract

This invention relates to the field of artificial neural network technology, specifically to an intelligent detection and defect analysis method for gas turbine blades. The method includes the following steps: acquiring blade material property data, blade geometric structure data, and operating load data; calculating the stress transfer matrix; and analyzing multi-node stress gradient changes. In this invention, the accuracy of blade defect detection is improved based on stress analysis and multi-source data fusion. By calculating the stress transfer matrix, abrupt change regions are screened, and stress concentration points are accurately located, avoiding misjudgments caused by relying on single visual features. Combined with the stress disturbance pattern of the inner wall of the water flow channel, the material stress offset is calculated to identify hidden defects. The stress characteristics of the cooling channel surface are screened, and the stress direction offset angle is calculated, so that the defect classification criteria include mechanical features. Stress and image data are normalized, and neural network training is optimized, making the classification accuracy adaptable to complex operating conditions and improving the ability to identify multiple types of defects.

Description

Technical Field

[0001] This invention relates to the field of artificial neural network technology, and in particular to a method for intelligent detection and defect analysis of gas turbine blades. Background Technology

[0002] The field of artificial neural networks (ANNs) encompasses the construction and training of neural network models using artificial intelligence algorithms to simulate the computational and decision-making processes of the human brain. The core of this technology lies in forming multi-layered structures through the connection of neurons and weights, and iteratively training based on large amounts of data to extract features and recognize patterns. ANNs are widely used in computer vision, natural language processing, automatic control, intelligent detection, and many other fields. In the field of intelligent detection, ANNs can utilize convolutional neural networks, recurrent neural networks, or deep reinforcement learning to achieve image recognition, anomaly detection, and predictive analysis in complex environments. Research in this field mainly focuses on optimizing network structure, improving training efficiency, enhancing generalization ability, and adjusting models to meet the accuracy requirements of different detection tasks, taking into account specific industry needs.

[0003] The intelligent detection and defect analysis method for gas turbine blades refers to a technical solution that utilizes artificial neural network technology to intelligently detect gas turbine blades and analyze defect features. This method encompasses technical aspects such as image acquisition, data preprocessing, neural network training and optimization, defect feature extraction, and classification. Specifically, firstly, images of the blade surface are acquired using a high-resolution industrial camera or laser scanning equipment, and the image data is preprocessed using methods such as filtering enhancement and edge detection to improve data quality. Subsequently, a detection model based on a convolutional neural network is constructed to learn features from the blade images, and a hierarchical classifier is used to determine the defect category. Finally, a specific loss function is used to optimize the model to improve recognition accuracy, and the location, morphology, and expansion trend of blade damage are analyzed in conjunction with a defect feature extraction algorithm.

[0004] Current technologies rely on a single image recognition method, which is susceptible to the effects of lighting and surface contamination, making it difficult to identify minute defects. For complex curved surfaces of blades, changes in light and shadow can lead to misjudgments, and the lack of consideration for internal stress distribution makes it difficult to detect deep cracks and stress concentration damage. Classification mainly depends on surface morphology, without incorporating stress characteristics, making it difficult to distinguish defects with similar surfaces but different stress distributions. Model optimization does not take into account the stress evolution under high temperature and high pressure conditions on blades, limiting its generalization ability and resulting in a high misjudgment rate under different operating conditions. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent detection and defect analysis method for gas turbine blades.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for intelligent detection and defect analysis of gas turbine blades, comprising the following steps:

[0007] S1: Acquire blade material property data, blade geometric structure data and operating load data, calculate stress transfer matrix, analyze multi-node stress gradient changes, screen stress abrupt change areas, extract stress concentration points and abrupt change points, and generate stress anomaly characteristic values.

[0008] S2: Call the stress anomaly characteristic value to obtain the inner wall data of the water flow channel of the heavy gas turbine blade, calculate the stress distribution of the inner wall, analyze the stress disturbance mode of the defect area, identify stress concentration points, calculate the stress offset of the material, screen stress gradient abrupt change points, analyze the residual stress accumulation value, and establish the defect stress disturbance coefficient.

[0009] S3: Call the aforementioned defect stress disturbance coefficient, collect the stress distribution on the surface of the cooling channel of the water-cooled blade, screen the stress characteristics of three types of defects: cracks, spalling, and deformation, calculate the stress direction offset angle, stress mean and gradient abrupt change point, and generate a defect feature vector.

[0010] S4: Call the defect feature vector, combine it with the stress anomaly feature data, normalize the classification features, obtain the blade surface image features, construct the stress feature channel, the defect stress disturbance channel and the image feature channel, and establish a multi-channel input dataset.

[0011] As a further aspect of the present invention, the stress anomaly feature values ​​specifically include stress concentration points, abrupt change points, and stress abrupt change regions; the defect stress disturbance coefficient includes stress concentration points, material stress offset, stress gradient abrupt change points, and residual stress accumulation values; the defect feature vector specifically includes crack stress features, spalling stress features, deformation stress features, stress direction offset angle, stress mean, and gradient abrupt change points; and the multi-channel input dataset includes stress feature channels, defect stress disturbance channels, and image feature channels.

[0012] As a further aspect of the present invention, the step of obtaining the stress anomaly characteristic value specifically includes:

[0013] S101: Based on blade material property data, blade geometric structure data, and operating load data, calculate the stress transfer matrix of multiple nodes, call the stress transfer values ​​between nodes, and obtain the stress gradient values ​​in different directions according to the spatial coordinate relationship between nodes. Calculate the magnitude of the gradient change rate to obtain the stress gradient change coefficient.

[0014] S102: Call the stress gradient change coefficient, calculate the stress abrupt change interval for the multi-node gradient change situation, filter the stress anomaly area according to the gradient change situation in the multi-interval, obtain the extreme point in the area, calculate the stress change rate of the area adjacent to the extreme point, and generate a set of stress abrupt change points.

[0015] S103: Based on the set of stress abrupt change points, calculate the stress concentration degree corresponding to multiple abrupt change points using the following formula:

[0016]

[0017] The stress concentration coefficients of multiple mutation points are obtained through calculation. The spatial coordinate information in the mutation point set is called to establish the morphological distribution of the stress anomaly region and obtain the stress anomaly characteristic value.

[0018] Where S1 represents the characteristic value of stress anomaly, Δσ 1,i d represents the stress change at the abrupt change point i. 1,i n represents the Euclidean distance between mutation point i and its neighboring points, n1 represents the total number of mutation points, and C represents the stability parameter, used to avoid the influence of the denominator approaching zero.

[0019] As a further aspect of the present invention, the step of obtaining the defect stress disturbance coefficient specifically includes:

[0020] S201: Call the stress anomaly feature value, calculate the stress distribution value of the inner wall based on the data of the inner wall of the water flow channel of the heavy gas turbine blade, call the stress anomaly feature value, compare and analyze the stress change amplitude at multiple locations on the inner wall, screen the stress change abrupt point, determine the stress anomaly distribution area of ​​the inner wall of the blade, and obtain the inner wall stress distribution data.

[0021] S202: Based on the inner wall stress distribution data, analyze the stress disturbance mode in the defect area, calculate the gradient change of local stress, screen stress gradient abrupt change points, and use the formula:

[0022]

[0023] The calculation obtains the local stress gradient change and calculates the stress offset value of the stress concentration point to generate the stress offset value of the stress concentration point.

[0024] Where S2 represents the change in local stress gradient, Δσ 2,i Δx represents the stress change at point i. i ρ represents the spacing between stress measurement points, E represents the elastic modulus of the material, ρ represents the density of the material, and n2 represents the total number of stress measurement points.

[0025] S203: Based on the stress offset of the stress concentration point, calculate the cumulative value of residual stress on the inner wall of the blade water flow channel, extract the residual stress distribution in the defect area, calculate the stress disturbance degree, and establish the defect stress disturbance coefficient.

[0026] As a further aspect of the present invention, the step of obtaining the defect feature vector specifically includes:

[0027] S301: Call the aforementioned defect stress disturbance coefficient, collect the stress distribution on the surface of the cooling channel of the water-cooled blade, calculate the corresponding stress gradient change based on the stress values ​​at multiple locations in the cooling channel, detect stress mutation points in multiple areas of the cooling channel, filter the stress intervals where the stress mutation points are located, and obtain the distribution of stress gradient mutation points.

[0028] S302: Based on the distribution of stress gradient abrupt change points, the stress characteristics of three types of defects—cracks, spalling, and deformation—are screened, the average stress value within the corresponding region is calculated, and the local stress direction shift is analyzed using the following formula:

[0029]

[0030] The stress direction offset angle of multiple defect regions is calculated, and the stress direction offset angle distribution is established.

[0031] Where V represents the variance of the mean stress, S 3,i S represents the stress value at the i-th position. 3,i,prev n represents the stress value of the previous measurement point at the i-th location, and n3 represents the total number of measurement points within the defect area.

[0032] S303: Call the stress direction offset angle distribution and combine it with the stress gradient mutation point distribution to construct the feature vector of each defect region. For defect regions of different types, extract the mean, direction offset angle and normalized values ​​of gradient mutation points of the feature vector to obtain the defect feature vector.

[0033] As a further aspect of the present invention, the step of obtaining the multi-channel input dataset specifically includes:

[0034] S401: Based on the defect feature vector and combined with the stress anomaly feature data, calculate the root mean square error and standard deviation of the feature values, determine the fluctuation range of the feature data based on the root mean square error, select the feature data with larger weights based on the standard deviation, perform normalization processing, and generate a feature normalization weight matrix.

[0035] S402: Call the feature normalization weight matrix to obtain the blade surface image features, calculate the principal component contribution rate of multiple image feature dimensions, filter features with high contribution rates, calculate the correlation coefficient between multiple channels, and construct the stress feature channel, defect stress disturbance channel, and image feature channel using the following formula:

[0036]

[0037] Calculate the overall feature channel influence degree to obtain the multi-channel feature influence matrix;

[0038] Where M represents the multi-channel feature influence matrix, W 4,k F represents the normalized feature weights. 4,k m represents the normalized single feature value, m4 represents the total number of features, and C 4,kp n represents the correlation coefficient between channels, and n4 represents the total number of normalized weighted features.

[0039] S403: Call the multi-channel feature influence matrix, calculate the multi-channel feature contribution rate, filter the channel data with high contribution rate, allocate the data input weight according to the normalized distribution weight of the feature influence matrix, and integrate to establish a multi-channel input dataset.

[0040] As a further aspect of the present invention, the method further includes:

[0041] S5: Call the multi-channel input dataset, construct the neural network input layer, set the weight constraints of the stress anomaly feature channel and the defect stress disturbance channel, optimize the hidden layer loss function, train the classification model, and establish the defect classification prediction result;

[0042] The defect classification prediction results specifically refer to the classification labels for three types of defects: cracks, spalling, and deformation.

[0043] As a further aspect of the present invention, the steps for obtaining the defect classification prediction results are specifically as follows:

[0044] S501: Call the multi-channel input dataset, preprocess the multi-channel stress data and defect stress disturbance data, calculate the stress feature vector of the multi-channel input data, adjust the normalization parameter according to the data range, calculate and store the initial weight distribution of the multi-feature channels, and obtain the input layer weight constraint matrix.

[0045] S502: Based on the input layer weight constraint matrix, adjust the hidden layer connection parameters, and perform a weighted summation of the multi-channel stress features using the following formula:

[0046]

[0047] The stress response value of the hidden layer was calculated;

[0048] Among them, R h W represents the stress response value of the hidden layer. 5,p S represents the weight of the input channel p. 5,p P represents the stress eigenvalue of the input channel p.5,q The weight D represents the stress disturbance q caused by the defect. 5,q This represents the defect stress disturbance value, where n5 is the total number of input channels and m5 is the total number of defect stress disturbance channels.

[0049] S503: Based on the stress response value of the hidden layer, train the classification model, calculate the classification boundary weights, select the classification thresholds that meet the classification criteria of the stress anomaly characteristic channels, calculate the classification boundary parameters and store the classification information, and establish the defect classification prediction results.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0051] This invention improves the accuracy of blade defect detection by combining stress analysis with multi-source data fusion. By calculating the stress transfer matrix, abrupt change regions are screened, and stress concentration points are precisely located, avoiding misjudgments caused by relying on single visual features. Combining the stress disturbance pattern of the inner wall of the water flow channel, material stress offset is calculated to identify hidden defects. The stress characteristics of the cooling channel surface are screened, and the stress direction offset angle is calculated, so that the defect classification criteria include mechanical characteristics, improving stability. Normalization of stress and image data optimizes neural network training, making the classification accuracy adaptable to complex working conditions and improving the ability to identify multiple types of defects. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the main steps of the present invention;

[0053] Figure 2 This is a flowchart of the steps for obtaining stress anomaly characteristic values ​​according to the present invention;

[0054] Figure 3 This is a flowchart illustrating the steps for obtaining the defect stress disturbance coefficient in this invention.

[0055] Figure 4 This is a flowchart illustrating the steps for obtaining the defect feature vector of the present invention.

[0056] Figure 5 This is a flowchart illustrating the steps for acquiring the multi-channel input dataset according to the present invention.

[0057] Figure 6 This is a flowchart illustrating the steps for obtaining the defect classification prediction results of this invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0060] Example 1

[0061] Please see Figure 1 This invention provides a technical solution: a method for intelligent detection and defect analysis of gas turbine blades, comprising the following steps:

[0062] S1: Acquire blade material property data, blade geometric structure data and operating load data, calculate stress transfer matrix, analyze multi-node stress gradient changes, screen stress abrupt change areas, extract stress concentration points and abrupt change points, and generate stress anomaly characteristic values.

[0063] S2: Call the stress anomaly characteristic value to obtain the inner wall data of the water flow channel of the heavy gas turbine blade, calculate the stress distribution of the inner wall, analyze the stress disturbance mode of the defect area, identify stress concentration points, calculate the stress offset of the material, screen stress gradient abrupt change points, analyze the residual stress accumulation value, and establish the defect stress disturbance coefficient.

[0064] S3: Call the defect stress disturbance coefficient, collect the stress distribution on the surface of the cooling channel of the water-cooled blade, filter the stress characteristics of three types of defects: cracks, spalling and deformation, calculate the stress direction offset angle, stress mean and gradient change point, and generate defect feature vector.

[0065] S4: Call the defect feature vector, combine it with the stress anomaly feature data, normalize the classification features, obtain the blade surface image features, construct the stress feature channel, the defect stress disturbance channel and the image feature channel, and establish a multi-channel input dataset.

[0066] S5: Call the multi-channel input dataset, construct the neural network input layer, set the weight constraints of the stress anomaly feature channel and the defect stress disturbance channel, optimize the hidden layer loss function, train the classification model, and establish the defect classification prediction results.

[0067] The stress anomaly feature values ​​specifically include stress concentration points, abrupt change points, and stress abrupt change regions. The defect stress perturbation coefficient includes stress concentration points, material stress offset, stress gradient abrupt change points, and residual stress accumulation. The defect feature vector specifically includes crack stress features, spalling stress features, deformation stress features, stress direction offset angle, stress mean, and gradient abrupt change points. The multi-channel input dataset includes stress feature channels, defect stress perturbation channels, and image feature channels. The defect classification prediction results specifically refer to the classification labels of three types of defects: cracks, spalling, and deformation.

[0068] Please see Figure 2 The specific steps for obtaining stress anomaly characteristic values ​​are as follows:

[0069] S101: Based on blade material property data, blade geometric structure data, and operating load data, calculate the stress transfer matrix of multiple nodes, call the stress transfer values ​​between nodes, and obtain the stress gradient values ​​in different directions according to the spatial coordinate relationship between nodes. Calculate the magnitude of the gradient change rate to obtain the stress gradient change coefficient.

[0070] The blade material properties data include parameters such as density, elastic modulus, and Poisson's ratio. The blade geometry data includes information such as blade length, thickness, chord length, and blade curvature. The operational load data involves factors such as aerodynamic loads, inertial loads, and centrifugal forces experienced by the blade during operation. After inputting this data into the calculation program, the blade model is discretized using a mesh generation method. The stress distribution values ​​at each mesh node are calculated using the finite element method. The stress values ​​between adjacent mesh nodes are correlated according to spatial coordinates, and the stress gradient value at each node is calculated using a difference method. The rate of change of the stress gradient is then analyzed. The stress gradient change at each node in the three principal directions is obtained through point-by-point calculation. Normalization is used to ensure consistent calculation across blade models of different scales. The stress gradient change coefficient is then calculated, defined as the ratio of the stress gradient change amplitude between adjacent grid nodes. Regions with drastic changes are filtered out to further improve calculation accuracy. Based on actual working conditions, assuming the blade material's elastic modulus is 210 GPa, Poisson's ratio is 0.3, the stress value at a certain node is 150 MPa, the stress value at an adjacent node is 140 MPa, and the distance between them is 5 mm, the stress gradient change in that direction is calculated as follows: After calculations in all directions, a complete set of stress gradient variation coefficients can be obtained.

[0071] S102: Call the stress gradient change coefficient, calculate the stress abrupt change interval for the gradient change of multiple nodes, filter the stress anomaly area based on the gradient change in the multiple intervals, obtain the extreme points in the area, calculate the stress change rate of the area adjacent to the extreme points, and generate a set of stress abrupt change points.

[0072] By using the stress gradient variation coefficient, a sliding window method is employed to determine stress abrupt change intervals for multi-node gradient changes. This involves analyzing the stress variation amplitude within a certain range. When the stress change within a certain interval exceeds a set rate of change threshold, it is identified as a stress abrupt change region. To ensure rationality, this threshold can be set as the 90th quantile of blade stress variation. For example, assuming the stress change rate corresponding to the 90th quantile after statistical data analysis is 8 MPa / mm, then when the stress change rate within a certain interval exceeds 8 MPa / mm, that region is marked as a stress abrupt change interval. Within this region, extreme points are calculated, i.e., the maximum and minimum local stress values ​​are selected. For instance, if the nodal stresses within a certain region are [120, 130, 125, 140, 135, 145] MPa, then the extreme points are 130 MPa and 145 MPa, respectively. Subsequently, the stress change rate of these extreme points and their adjacent regions is calculated to determine the degree of local abrupt change. If the stress change rate of adjacent regions is greater than 10 MPa / mm, this point is considered a abrupt change point, thus generating a set of stress abrupt change points.

[0073] S103: Based on the set of stress abrupt change points, calculate the stress concentration corresponding to multiple abrupt change points using the following formula:

[0074]

[0075] The stress concentration coefficients of multiple mutation points are obtained through calculation. The spatial coordinate information in the mutation point set is called to establish the morphological distribution of the stress anomaly region and obtain the stress anomaly characteristic value.

[0076] Where S1 represents the characteristic value of stress anomaly, Δσ 1,i d represents the stress change at the abrupt change point i. 1,i n represents the Euclidean distance between mutation point i and its neighboring points, n1 represents the total number of mutation points, and C represents the stability parameter, used to avoid the influence of the denominator approaching zero.

[0077] The formula is as follows:

[0078]

[0079] Where S1 represents the characteristic value of stress anomaly, Δσ 1,i d represents the stress change at the abrupt change point i. 1,i Let represent the Euclidean distance between the abrupt change point i and its neighboring points, n1 represent the total number of abrupt change points, and C be the stability parameter. Assume that within a certain blade region, five stress abrupt change points have been identified, with stress changes of the following values:

[0080] Given stress concentrations of [12, 15, 10, 18, 14] MPa and Euclidean distances between adjacent points of [3, 4, 3.5, 4.2, 3.8] mm, and taking a stability parameter C = 1 to avoid the influence of the denominator approaching zero, the stress concentration factor is calculated as follows:

[0081]

[0082]

[0083]

[0084] The results indicate that there is a relatively concentrated stress abrupt change region in this blade area. This value can be further compared with historical stress anomaly distribution data to determine the degree of anomaly in this region. At the same time, by calling the spatial coordinate information in the set of abrupt change points and using a three-dimensional fitting method, the morphological distribution of the stress anomaly region can be established, and stress anomaly characteristic values ​​can be obtained.

[0085] Table 1. Stress abrupt change points and calculation parameters

[0086] Mutation point numbering Stress change (MPa) Euclidean distance between adjacent points (mm) 1 12 3 2 15 4 3 10 3.5 4 18 4.2 5 14 3.8

[0087] Table 1 shows the stress changes at five abrupt change points and the Euclidean distances between adjacent points. These data are used to calculate the stress concentration factor and ultimately determine the characteristic values ​​of the stress anomaly.

[0088] Please see Figure 3 The specific steps for obtaining the defect stress disturbance coefficient are as follows:

[0089] S201: Call the stress anomaly feature value, calculate the stress distribution value of the inner wall based on the data of the inner wall of the water flow channel of the heavy gas turbine blade, call the stress anomaly feature value, compare and analyze the stress change amplitude at multiple locations on the inner wall, screen the stress change abrupt point, determine the stress anomaly distribution area of ​​the inner wall of the blade, and obtain the inner wall stress distribution data.

[0090] First, a discretized model of the inner wall surface needs to be established. This model must include the mesh division information of the blade surface and ensure that stress values ​​can be obtained at different locations of fluid flow. In terms of the arrangement of discrete points, an equal-spacing distribution method is adopted, with the interval between each measuring point set to 2 mm to ensure the continuity of data. Subsequently, stress measurement instruments are used to perform stress tests on each measuring point, and the stress value of each point is obtained and recorded in the database to obtain complete inner wall stress distribution data.

[0091] Next, the stress distribution on the inner wall needs to be calculated. This requires traversing the stress values ​​at each measuring point and performing statistical analysis. When calculating the characteristic values ​​of stress anomalies, the mean stress σ is calculated first. avg and standard deviation σ std The calculation formula is as follows:

[0092]

[0093]

[0094] Where, σ i Let n represent the stress value at the i-th measuring point and n represent the total number of measuring points. After calculation, the method for filtering stress anomalies is as follows: set an anomaly judgment threshold σ. thr =σ avg +2σ std Anything greater than σ thr All measuring points were identified as stress anomaly points.

[0095] When comparing and analyzing the stress variation amplitude at multiple locations on the inner wall, the stress variation Δσ between adjacent measuring points is selected. i =|σ i -σ i+1 After calculating the stress change between all measuring points, select those that satisfy Δσ i >1.5σ std The measuring points are used as abrupt stress change points, and the location of the abrupt stress points is used to determine the stress anomaly distribution area, and finally the stress distribution data of the inner wall is obtained.

[0096] Table 2 shows the stress distribution data of some measuring points on the inner wall of the water flow channel of a gas turbine blade, including the location of the measuring points, the measured stress values, and the calculated stress changes:

[0097] Table 2. Stress Distribution Data at Measuring Points on the Inner Wall of the Blade Flow Channel

[0098]

[0099]

[0100] As shown in Table 2, the stress changes at measuring points 4 and 6 significantly exceeded the average level, indicating stress abrupt change points. Therefore, it can be determined that there is an abnormal stress distribution in this area on the inner wall of the blade's water flow channel.

[0101] S202: Based on the stress distribution data of the inner wall, analyze the stress disturbance mode in the defect area, calculate the gradient change of local stress, screen the stress gradient abrupt change points, and use the following formula:

[0102]

[0103] The calculation obtains the local stress gradient change and calculates the stress offset value of the stress concentration point to generate the stress offset value of the stress concentration point.

[0104] Where S2 represents the change in local stress gradient, Δσ 2,iΔx represents the stress change at point i. i ρ represents the spacing between stress measurement points, E represents the elastic modulus of the material, ρ represents the density of the material, and n2 represents the total number of stress measurement points.

[0105] formula:

[0106]

[0107] Where, Δσ 2,i =σ i+1 -σ i Δx represents the stress change at the i-th measuring point. i The distance between measuring points is represented by E, the elastic modulus of the material is E, the density of the material is ρ, and the total number of measuring points is n2.

[0108] Selecting the blade material with an elastic modulus E = 210 GPa and a density ρ = 7800 kg / m³, calculate the local stress gradient change at each measuring point:

[0109] Calculate the gradient change at the first measuring point:

[0110]

[0111]

[0112] Calculate the gradient change at the 4th measurement point:

[0113]

[0114]

[0115] After calculating for all measuring points, select those that satisfy S. 2,i Points with a stress concentration value greater than 50 are considered stress concentration points. For example, if measuring point number 4 meets the condition, it is determined to be a stress concentration point, and its stress offset value σ is calculated. 偏移 :

[0116] σ 偏移 =σ max -σ avg =350-280=70MPa;

[0117] The result indicates that the stress at measuring point 4 is 70 MPa higher than the average stress, which can be used as the basis for calculating the subsequent defect stress disturbance coefficient.

[0118] S203: Based on the stress offset at the stress concentration point, calculate the cumulative residual stress on the inner wall of the blade's water flow channel, extract the residual stress distribution in the defect area, calculate the degree of stress disturbance, and establish the defect stress disturbance coefficient.

[0119] The residual stress is calculated using a point-by-point integration method, and the formula is as follows:

[0120]

[0121] Input data and calculate:

[0122] σ 累积 = (250-280)×2+(260-280)×2+(280-280)×2+(350-

[0123] 280)×2+(370-280)×2+(400-280)×2;

[0124] =(-60)+(-40)+0+140+180+240=460MPa·mm;

[0125] This value represents the accumulated residual stress on the inner wall of the blade's water flow channel due to stress concentration. Finally, the degree of stress disturbance is calculated, and the defect stress disturbance coefficient is established.

[0126]

[0127] The results indicate that the defect stress disturbance on the inner wall of the blade is relatively high, with a value of 2.55. This coefficient can be used to assess the stability of the stress distribution on the blade.

[0128] Please see Figure 4 The specific steps for obtaining the defect feature vector are as follows:

[0129] S301: Call the defect stress disturbance coefficient, collect the stress distribution on the surface of the cooling channel of the water-cooled blade, calculate the corresponding stress gradient change based on the stress values ​​at multiple locations in the cooling channel, detect stress mutation points in multiple areas of the cooling channel, filter the stress intervals where the stress mutation points are located, and obtain the distribution of stress gradient mutation points.

[0130] First, stress sensors are placed in different areas of the cooling channel to collect instantaneous stress values ​​at each measurement point. Assuming eight measurement points are placed in a certain area, the obtained instantaneous stress values ​​are σ1, σ2, σ3, ..., σ8. Then, based on the stress data of these measurement points, the stress gradient between each point and its adjacent points is calculated, and the change in stress gradient is defined as ΔS = |σ i -σ i-1 The stress gradient change sequence {ΔS2, ΔS3, ..., ΔS8} is calculated for all measurement points. Then, based on the set stress gradient abrupt change threshold θ, abrupt change points are selected, i.e., ΔS8 is determined. iFor measurement points >θ, assuming θ is set to 5MPa, when a measurement point satisfies conditions such as ΔS5 = 6.2MPa and ΔS7 = 5.6MPa, the regions corresponding to measurement points 5 and 7 can be identified as the intervals where stress abrupt change points are located. Then, based on the determination of stress abrupt change points, the stress intervals where the abrupt change points are located are screened. That is, regional boundaries are set between multiple adjacent stress abrupt change points. By comparing the stress change rate before and after the abrupt change point, it is determined whether they belong to the same region. If the stress gradient abrupt change amount of adjacent points is greater than the set abrupt change threshold θ and the change trend is consistent, they are classified into the same region. Otherwise, they are divided independently. Finally, the distribution of stress gradient abrupt change points in the cooling channel is obtained.

[0131] S302: Based on the distribution of stress gradient abrupt change points, the stress characteristics of three types of defects—cracks, spalling, and deformation—are screened. The average stress value within the corresponding region is calculated, and the local stress direction shift is analyzed using the following formula:

[0132]

[0133] The stress direction offset angle of multiple defect regions is calculated, and the stress direction offset angle distribution is established.

[0134] Where V represents the variance of the mean stress, S 3,i S represents the stress value at the i-th position. 3,i,prev n represents the stress value of the previous measurement point at the i-th location, and n3 represents the total number of measurement points within the defect area.

[0135] First, stress data are screened within each defect region, and the average stress value for each region is calculated. Assume there are n³ measurement points in the crack region, with stress values ​​of {S}. 3,1 ,S 3,2 ,...,S 3,n3 When calculating its mean and variance V, it is necessary to consider the stress value S of the previous measurement point. 3,i,prev Through the formula:

[0136]

[0137] Calculations are performed, assuming the crack region contains 5 measurement points with stress values ​​{S}. 3,1 =120,S 3,2 =118,S 3,3 =110,S 3,4 =105,S 3,5 =100}MPa, the stress values ​​at the previous measurement point are {S 3,0 =122,S 3,1 =120,S 3,2 =118,S 3,3 =110,S 3,4=105}MPa, then its mean stress variance can be calculated:

[0138]

[0139]

[0140]

[0141] Simultaneously, the local stress direction shift in each defect area is analyzed. That is, by comparing the stress distribution at each measurement point, the stress change direction between the measurement points is calculated, and the stress direction shift angle of each area is calculated by the directional deviation of the stress gradient change. Assuming that the directional change angle of the stress gradient in a certain area is θ1 = 10°, θ2 = 15°, and θ3 = 5°, the stress direction shift angle distribution of that area can be constructed.

[0142] S303: Call the stress direction offset angle distribution and combine it with the stress gradient mutation point distribution to construct the feature vector of each defect region. For defect regions with different types, extract the mean, direction offset angle and normalized values ​​of gradient mutation points of the feature vector to obtain the defect feature vector.

[0143] For all measured stress values ​​{S1,S2,...,S} n Set the maximum value S. max and minimum value S min Calculate the normalized value:

[0144]

[0145] Assume the stress values ​​at the measurement points in the crack region are {120, 118, 110, 105, 100} MPa, where the maximum value is S. max =120MPa, minimum value S min =100MPa, then the normalized value is calculated as follows:

[0146]

[0147]

[0148]

[0149]

[0150]

[0151] Finally, the normalized stress feature vector [1,0.9,0.5,0.25,0] is obtained. Combined with the stress direction offset angle {10°,15°,5°} and gradient abrupt change point data, the defect feature vector is fully described.

[0152] Table 3. Stress data and normalized feature vectors of the defect area

[0153] Measurement point number Stress value (MPa) Normalized stress value Directional offset angle (°) 1 120 1.00 10 2 118 0.90 15 3 110 0.50 5 4 105 0.25 - 5 100 0.00 -

[0154] As shown in Table 3, the stress values ​​in the crack defect region are normalized to form a feature vector. Combined with the direction offset angle parameter, the defect feature vector in this region is made more specific, which is convenient for subsequent analysis and judgment.

[0155] Please see Figure 5 The specific steps for obtaining the multi-channel input dataset are as follows:

[0156] S401: Based on the defect feature vector and stress anomaly feature data, calculate the root mean square error and standard deviation of the feature values, determine the fluctuation range of the feature data based on the root mean square error, select the feature data with larger weights based on the standard deviation, perform normalization processing, and generate the feature normalization weight matrix.

[0157] First, defect feature data needs to be collected, including but not limited to defect morphology (e.g., length 10mm, width 2mm), depth (e.g., 0.5mm), and material properties (e.g., elastic modulus 210GPa). Then, stress anomaly feature data is collected, such as maximum principal stress 150MPa, minimum principal stress 50MPa, and shear stress 20MPa. For these data, the root mean square error (RMSE) is calculated to measure the stability of the feature data. The RMSE calculation formula is as follows:

[0158]

[0159] Where, x i Representing the feature values ​​of each sample, The representative characteristic is the mean. Taking the maximum principal stress as an example, assuming the sample data are 148MPa, 152MPa, and 149MPa, and their mean is 149.67MPa, then the RMSE is calculated as follows:

[0160]

[0161] The RMSE threshold is set to 2 MPa. When the RMSE is below 2 MPa, the data fluctuation range is considered stable. Next, the standard deviation (σ) is calculated using the following formula:

[0162]

[0163] After calculating the standard deviation of each feature, the weighting formula is used. After normalization, the following normalized feature weight matrix is ​​obtained:

[0164] W=[0.35 0.30 0.20 0.10 0.05];

[0165] The results indicate that features with higher weights have a greater impact on the overall fluctuation range of the data. Therefore, these features with higher weights should be given priority in subsequent steps and given greater influence during the calculation process to ensure the representativeness and stability of the data.

[0166] S402: Call the feature normalization weight matrix to obtain blade surface image features, calculate the principal component contribution rate of multiple image feature dimensions, filter features with high contribution rates, calculate the correlation coefficient between multiple channels, and construct stress feature channels, defect stress disturbance channels, and image feature channels using the following formula:

[0167]

[0168] Calculate the overall feature channel influence degree to obtain the multi-channel feature influence matrix;

[0169] Where M represents the multi-channel feature influence matrix, W 4,k F represents the normalized feature weights. 4,k m represents the normalized single feature value, m4 represents the total number of features, and C 4,kp n represents the correlation coefficient between channels, and n4 represents the total number of normalized weighted features.

[0170] For example, if an image has a GLCM contrast of 0.45 and a correlation of 0.80, the contribution rate is calculated as follows:

[0171]

[0172] Assuming the maximum contrast is 0.5 and the maximum correlation is 1, then the calculation is as follows:

[0173]

[0174] If the contribution rate threshold is set to 1.5, this feature is retained. Then, the correlation coefficient C between channels is calculated. 4,kp For example, if the correlation between a certain feature and the stress channel is set to 0.85, and the correlation with the defect disturbance channel is set to 0.75, then a multi-channel feature matrix is ​​constructed, and the comprehensive feature channel influence is calculated according to the following formula:

[0175]

[0176] For example, let W 4,1 =0.35, F 4,1 =0.45, C 4,1p =0.85, then calculate:

[0177]

[0178] The results indicate that the correlation between different channels has a significant effect on the calculation of the multi-channel feature influence matrix. Among them, high-weight features have a larger influence weight in the stress and defect disturbance channels, which will play a decisive role in the subsequent dataset construction process.

[0179] S403: Call the multi-channel feature influence matrix, calculate the multi-channel feature contribution rate, filter the channel data with high contribution rate, allocate the data input weights according to the normalized distribution weights of the feature influence matrix, and integrate them to establish a multi-channel input dataset.

[0180] The method is as follows:

[0181]

[0182] Assuming the feature influence matrix is ​​M = [0.116, 0.203, 0.178], and the sum is 0.497, then the normalized input weights are:

[0183] W′=[0.233 0.409 0.358];

[0184] The final multi-channel input dataset is shown in Table 4 below:

[0185] Table 4 Multichannel Input Dataset

[0186]

[0187] The results show that when constructing a multi-channel input dataset, using a feature influence matrix for normalized distribution weight allocation can ensure that the contribution rate of each channel data is reasonably measured, while ensuring that the input dataset can reflect the key correlations between each channel to the greatest extent and be used for subsequent analysis and processing.

[0188] Please see Figure 6 The specific steps for obtaining the defect classification prediction results are as follows:

[0189] S501: Call the multi-channel input dataset, preprocess the multi-channel stress data and defect stress disturbance data, calculate the stress feature vector of the multi-channel input data, adjust the normalization parameters according to the data range, calculate and store the initial weight distribution of the multi-feature channels, and obtain the input layer weight constraint matrix.

[0190] First, the collected multi-channel stress data and defect stress disturbance data are preprocessed. During this process, the stress data undergoes noise reduction, and a moving average method is used to eliminate random interference terms. For example, for the stress value S(t) at a certain time t, the data can be smoothed by calculating a weighted average of five adjacent data points. To obtain a more stable stress signal, the data is then normalized. First, the maximum and minimum values ​​of the stress data are calculated. Assuming the stress data range is between -500 and 500 MPa, the normalization formula is used. Normalization is performed to map the data range to [0,1], ensuring consistent numerical scale across all channels. Next, the stress eigenvectors for each channel are calculated. Principal component analysis (PCA) is used for dimensionality reduction to extract the principal stress eigenvalues ​​for each channel. For example, the covariance matrix Σ is calculated for the three-channel data {S1,S2,S3}. After eigenvalue decomposition, the principal eigenvector v1 is taken as the principal component direction, yielding the dimensionality-reduced stress eigenvalues. Then, adjust the normalization parameters according to the data range. If the input data distribution varies significantly, the mean μ needs to be recalculated. S and standard deviation σ S Then normalization can be adopted. Zero-mean normalization is performed. Furthermore, the initial weight distribution for multiple feature channels is calculated and stored. Assuming there are five input channels, the weights are initialized using a uniform distribution W0 = [0.2, 0.2, 0.2, 0.2, 0.2], and the input layer weight constraint matrix W is obtained. in Its initialization can be set using a variance balancing method to ensure that the initial contributions of all channels are balanced, such as setting... This ensures the balance of influence from different stress channels, thereby completing the preprocessing of input data and weight initialization.

[0191] S502: Based on the input layer weight constraint matrix, the hidden layer connection parameters are adjusted, and the multi-channel stress features are weighted and summed using the following formula:

[0192]

[0193] The stress response value of the hidden layer was calculated;

[0194] Among them, R h W represents the stress response value of the hidden layer. 5,p S represents the weight of the input channel p. 5,p P represents the stress eigenvalue of the input channel p. 5,q The weight D represents the stress disturbance q caused by the defect. 5,q This represents the defect stress disturbance value, where n5 is the total number of input channels and m5 is the total number of defect stress disturbance channels.

[0195] First, determine the weight distribution W5 for each input channel and the weight P5 for the defect stress disturbance channel. Assuming the number of input channels is n5 = 4 and the number of defect stress disturbance channels is m5 = 3, then for each channel, the stress eigenvalue S... 5,pTo avoid the negative stress affecting the calculation, absolute value calculations are performed, followed by the calculation of the weighted summation part. For example, if W5 = [0.25, 0.25, 0.3, 0.2] and

[0196] S5 = [50, -30, 20, -10] MPa, then calculate R. input =0.25×50+0.25×30+0.3×20+0.2×10=27.5MPa. Next, calculate the root mean square value of the defect stress disturbance component. Assuming the defect stress disturbance value D5=[9,16,25]MPa and the corresponding weight P5=[0.3,0.5,0.2], calculate the square root value. Then calculate the weighted average. That is, R defect = (0.3×3+0.5×4+0.2×5) / 3 = 1.93MPa, and finally calculate the stress response value R of the hidden layer. h =R input +R defect =27.5+1.93=29.43MPa, thus obtaining the stress response value of the hidden layer.

[0197] S503: Based on the stress response value of the hidden layer, a classification model is trained, the classification boundary weights are calculated, the classification thresholds that meet the classification criteria of the stress anomaly characteristic channels are selected, the classification boundary parameters are calculated and the classification information is stored, and the defect classification prediction results are established.

[0198] First, define the classification boundary weight W. cls The gradient descent method is used for optimization, and the algorithm is updated in each iteration. To minimize the classification error L, where the learning rate α = 0.01, a classification threshold T that meets the classification criteria for stress anomaly feature channels is then selected. cls Assuming that the anomaly classification threshold is set to T based on historical data statistics. cls =28MPa, then determine the stress response value R of the hidden layer. h Does 29.43 MPa exceed the threshold? The result satisfies R. h >T cls Therefore, the data is classified as a defect category, and then the classification boundary parameters, including the weight W, are calculated. cls and bias term b cls Assuming the initial parameter is W cls =0.8,b cls =2, then the classification decision function is f(R) h ) = W cls R h +b clsCalculate f(29.43) = 0.8 × 29.43 + 2 = 25.54, store classification information, including classification result labels and calculation parameters, and establish defect classification prediction results.

[0199] Table 5. Example of calculating stress response values ​​of hidden layers

[0200] Enter channel number Stress characteristic value (MPa) Input weights Weighted stress (MPa) 1 50 0.25 12.5 2 -30 0.25 7.5 3 20 0.30 6.0 4 -10 0.20 2.0 total — — 27.5

[0201] Table 5 lists an example of stress characteristic value calculation. The weighted sum of each input channel is calculated to be 27.5 MPa. Then, the defect stress perturbation part is added to finally obtain the hidden layer stress response value of 29.43 MPa. This value is used for classification model training and defect classification prediction.

[0202] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligent detection and defect analysis of gas turbine blades, characterized in that, Includes the following steps: S1: Acquire blade material property data, blade geometric structure data and operating load data, calculate stress transfer matrix, analyze multi-node stress gradient changes, screen stress abrupt change areas, extract stress concentration points and abrupt change points, and generate stress anomaly characteristic values. S2: Call the stress anomaly characteristic value to obtain the inner wall data of the water flow channel of the heavy gas turbine blade, calculate the stress distribution of the inner wall, analyze the stress disturbance mode of the defect area, identify stress concentration points, calculate the stress offset of the material, screen stress gradient abrupt change points, analyze the residual stress accumulation value, and establish the defect stress disturbance coefficient. S3: Call the aforementioned defect stress disturbance coefficient, collect the stress distribution on the surface of the cooling channel of the water-cooled blade, screen the stress characteristics of three types of defects: cracks, spalling, and deformation, calculate the stress direction offset angle, stress mean and gradient abrupt change point, and generate a defect feature vector. S4: Call the defect feature vector, combine it with the stress anomaly feature data, normalize the classification features, obtain the blade surface image features, construct the stress feature channel, the defect stress disturbance channel and the image feature channel, and establish a multi-channel input dataset.

2. The intelligent detection and defect analysis method for gas turbine blades according to claim 1, characterized in that, The stress anomaly feature values ​​specifically include stress concentration points, abrupt change points, and stress abrupt change regions. The defect stress disturbance coefficient includes stress concentration points, material stress offset, stress gradient abrupt change points, and residual stress accumulation values. The defect feature vector specifically includes crack stress features, spalling stress features, deformation stress features, stress direction offset angle, stress mean, and gradient abrupt change points. The multi-channel input dataset includes stress feature channels, defect stress disturbance channels, and image feature channels.

3. The intelligent detection and defect analysis method for gas turbine blades according to claim 2, characterized in that, The specific steps for obtaining the stress anomaly characteristic values ​​are as follows: S101: Based on blade material property data, blade geometric structure data, and operating load data, calculate the stress transfer matrix of multiple nodes, call the stress transfer values ​​between nodes, and obtain the stress gradient values ​​in different directions according to the spatial coordinate relationship between nodes. Calculate the magnitude of the gradient change rate to obtain the stress gradient change coefficient. S102: Call the stress gradient change coefficient, calculate the stress abrupt change interval for the multi-node gradient change situation, filter the stress anomaly area according to the gradient change situation in the multi-interval, obtain the extreme point in the area, calculate the stress change rate of the area adjacent to the extreme point, and generate a set of stress abrupt change points. S103: Based on the set of stress abrupt change points, calculate the stress concentration degree corresponding to multiple abrupt change points using the following formula: The stress concentration coefficients of multiple mutation points are obtained through calculation. The spatial coordinate information in the mutation point set is called to establish the morphological distribution of the stress anomaly region and obtain the stress anomaly characteristic value. Where S1 represents the characteristic value of stress anomaly, Δσ 1,i d represents the stress change at the abrupt change point i. 1,i n represents the Euclidean distance between mutation point i and its neighboring points, n1 represents the total number of mutation points, and C represents the stability parameter, used to avoid the influence of the denominator approaching zero.

4. The intelligent detection and defect analysis method for gas turbine blades according to claim 3, characterized in that, The specific steps for obtaining the defect stress disturbance coefficient are as follows: S201: Call the stress anomaly feature value, calculate the stress distribution value of the inner wall based on the data of the inner wall of the water flow channel of the heavy gas turbine blade, call the stress anomaly feature value, compare and analyze the stress change amplitude at multiple locations on the inner wall, screen the stress change abrupt point, determine the stress anomaly distribution area of ​​the inner wall of the blade, and obtain the inner wall stress distribution data. S202: Based on the inner wall stress distribution data, analyze the stress disturbance mode in the defect area, calculate the gradient change of local stress, screen stress gradient abrupt change points, and use the formula: The calculation obtains the local stress gradient change and calculates the stress offset value of the stress concentration point to generate the stress offset value of the stress concentration point. Where S2 represents the change in local stress gradient, Δσ 2,i Δx represents the stress change at point i. i ρ represents the spacing between stress measurement points, E represents the elastic modulus of the material, ρ represents the density of the material, and n2 represents the total number of stress measurement points. S203: Based on the stress offset of the stress concentration point, calculate the cumulative value of residual stress on the inner wall of the blade water flow channel, extract the residual stress distribution in the defect area, calculate the stress disturbance degree, and establish the defect stress disturbance coefficient.

5. The intelligent detection and defect analysis method for gas turbine blades according to claim 4, characterized in that, The specific steps for obtaining the defect feature vector are as follows: S301: Call the aforementioned defect stress disturbance coefficient, collect the stress distribution on the surface of the cooling channel of the water-cooled blade, calculate the corresponding stress gradient change based on the stress values ​​at multiple locations in the cooling channel, detect stress mutation points in multiple areas of the cooling channel, filter the stress intervals where the stress mutation points are located, and obtain the distribution of stress gradient mutation points. S302: Based on the distribution of stress gradient abrupt change points, the stress characteristics of three types of defects—cracks, spalling, and deformation—are screened, the average stress value within the corresponding region is calculated, and the local stress direction shift is analyzed using the following formula: The stress direction offset angle of multiple defect regions is calculated, and the stress direction offset angle distribution is established. Where V represents the variance of the mean stress, S 3,i S represents the stress value at the i-th position. 3,i,prev n represents the stress value of the previous measurement point at the i-th location, and n3 represents the total number of measurement points within the defect area. S303: Call the stress direction offset angle distribution and combine it with the stress gradient mutation point distribution to construct the feature vector of each defect region. For defect regions of different types, extract the mean, direction offset angle and normalized values ​​of gradient mutation points of the feature vector to obtain the defect feature vector.

6. The intelligent detection and defect analysis method for gas turbine blades according to claim 5, characterized in that, The specific steps for obtaining the multi-channel input dataset are as follows: S401: Based on the defect feature vector and combined with the stress anomaly feature data, calculate the root mean square error and standard deviation of the feature values, determine the fluctuation range of the feature data based on the root mean square error, select the feature data with larger weights based on the standard deviation, perform normalization processing, and generate a feature normalization weight matrix. S402: Call the feature normalization weight matrix to obtain the blade surface image features, calculate the principal component contribution rate of multiple image feature dimensions, filter features with high contribution rates, calculate the correlation coefficient between multiple channels, and construct the stress feature channel, defect stress disturbance channel, and image feature channel using the following formula: Calculate the overall feature channel influence degree to obtain the multi-channel feature influence matrix; Where M represents the multi-channel feature influence matrix, W 4,k F represents the normalized feature weights. 4,k m represents the normalized single feature value, m4 represents the total number of features, and C 4,kp n represents the correlation coefficient between channels, and n4 represents the total number of normalized weighted features. S403: Call the multi-channel feature influence matrix, calculate the multi-channel feature contribution rate, filter the channel data with high contribution rate, allocate the data input weight according to the normalized distribution weight of the feature influence matrix, and integrate to establish a multi-channel input dataset.

7. The intelligent detection and defect analysis method for gas turbine blades according to claim 6, characterized in that, The method further includes: S5: Call the multi-channel input dataset, construct the neural network input layer, set the weight constraints of the stress anomaly feature channel and the defect stress disturbance channel, optimize the hidden layer loss function, train the classification model, and establish the defect classification prediction result; The defect classification prediction results specifically refer to the classification labels for three types of defects: cracks, spalling, and deformation.

8. The intelligent detection and defect analysis method for gas turbine blades according to claim 7, characterized in that, The specific steps for obtaining the defect classification prediction results are as follows: S501: Call the multi-channel input dataset, preprocess the multi-channel stress data and defect stress disturbance data, calculate the stress feature vector of the multi-channel input data, adjust the normalization parameter according to the data range, calculate and store the initial weight distribution of the multi-feature channels, and obtain the input layer weight constraint matrix. S502: Based on the input layer weight constraint matrix, adjust the hidden layer connection parameters, and perform a weighted summation of the multi-channel stress features using the following formula: The stress response value of the hidden layer was calculated; Among them, R h W represents the stress response value of the hidden layer. 5,p S represents the weight of the input channel p. 5,p P represents the stress eigenvalue of the input channel p. 5,q The weight D represents the stress disturbance q caused by the defect. 5,q This represents the defect stress disturbance value, where n5 is the total number of input channels and m5 is the total number of defect stress disturbance channels. S503: Based on the stress response value of the hidden layer, train the classification model, calculate the classification boundary weights, select the classification thresholds that meet the classification criteria of the stress anomaly characteristic channels, calculate the classification boundary parameters and store the classification information, and establish the defect classification prediction results.

Images