A crack slip type recognition method based on neural network and pre-training principle

Abstract

The application discloses a crack slip type recognition method based on a neural network and a pre-training principle, and comprises the following steps: establishing a numerical calculation model according to the geometric features and material parameters of a real object structure; adding a crack source and setting the size, position and type of the crack; adding a calculation result output position, and the output position corresponds to the arrangement position of a sensor in the real object structure; using the numerical model, calculating an elastic wave field excited by crack slip / extension, outputting an elastic wave time domain signal curve at the set position, picking up the amplitude of the signal curve at each position; randomly changing the type of the crack, keeping other parameters unchanged, calculating the signal amplitude corresponding to different crack types, and establishing a data set of crack type characteristic parameters-signal amplitude; based on a feedforward neural network model, designing the neuron number of an input layer, an intermediate layer and an output layer; training the neural network model; evaluating the training result, and obtaining a crack slip type recognition model.

Description

TECHNICAL FIELD

[0001] The application relates to a crack slip type identification method, in particular to a crack slip type identification method based on a neural network and a pre-training principle. BACKGROUND

[0002] Crack identification technology is a key technology for evaluating structural integrity and predicting the remaining service life of a structure, and can provide important data support for structural safety analysis. At present, the traditional crack identification method mainly identifies the parameters such as the expansion stage, position and length of a crack. In order to further improve the crack identification accuracy and identify the slip type and direction of a crack, scholars have developed a moment tensor theory. The crack inversion method based on the moment tensor theory is an excellent crack dynamic expansion identification method. The method can accurately calculate and identify the detailed characteristics of a crack through the elastic wave signals excited by crack slip, determine which type (shear or tensile) the crack belongs to, and then provide detailed data support for the analysis of crack expansion speed and stress / strain distribution characteristics in the crack area, so as to realize the judgment of structural damage condition and accurate evaluation of remaining life.

[0003] The existing crack identification technology based on the moment tensor theory solves the moment tensor of a crack through the spatial distribution characteristics of acoustic emission signals, and then calculates the crack surface slip direction and normal direction of the corresponding crack through the eigenvalues and eigenvectors of the moment tensor. Specifically, the existing method arranges a sensor array in the structure, records the elastic wave signals excited by the same crack surface slip at different positions in the structure, and extracts the first arrival wave amplitude of each signal. Then, the moment tensor solving formula containing the unknown moment tensor elements and the known displacement amplitude is listed according to the simplified Green function. In addition, the material parameters (wave velocity) and crack source-sensor connection direction cosine contained in the formula can be obtained through the pre-test or crack source auxiliary positioning technology. Therefore, the solving formula only contains unknown moment tensor elements. According to the number of unknown elements, a proper number of sensor array signals are selected to solve the moment tensor corresponding to the crack slip source, and then the crack size, slip direction and direction can be obtained through the tensor eigenvalue and eigenvector analysis.

[0004] The disadvantage of the existing crack identification technology based on the moment tensor theory is poor applicability of the method. The reason for this disadvantage is that the crack identification method based on the moment tensor theory depends on the accurate Green function, that is, the explicit expression of the elastic wave field excited by the unit load. At present, only simple models (infinite elastic body and thin plate structure) have the explicit expression of the Green function. For structures with complex shape and material, it is very difficult to solve the explicit expression of the Green function, which leads to the fact that the existing crack identification method based on the moment tensor theory can only be applied to simple structure forms. SUMMARY

[0005] The application aims to provide a crack slip type recognition method based on a neural network and a pre-training principle, directly fitting a mapping relationship between an elastic wave field feature and a crack slip equivalent moment tensor through a data-driven principle, and solving the problems of crack surface slip equivalent moment tensor solving and type recognition in a complex structure.

[0006] The technical scheme comprises the following steps:

[0007] S1: A numerical calculation model is established according to the geometric features and material parameters of a real object structure;

[0008] S2: A crack source is added in the numerical model, and the size, position and type of the crack are set;

[0009] S3: A calculation result output position is added in the numerical model, and the output position corresponds to the arrangement position of a sensor in the real object structure;

[0010] S4: The numerical model is used to calculate an elastic wave field excited by crack slip / extension, output the elastic wave time-domain signal curve at the position set in S3, and pick up the amplitude of the signal curve at each position;

[0011] S5: Steps S2-S4 are repeated, the type of the crack is randomly changed, other parameters are kept unchanged, the signal amplitudes corresponding to different crack types are calculated, and a crack type feature parameter-signal amplitude data set is established;

[0012] S6: The number of neurons in the input layer, the intermediate layer and the output layer is designed;

[0013] S7: The crack type-signal amplitude data set obtained in S5 is divided into training, testing and verification data sets, and a neural network model is trained;

[0014] S8: The training result is evaluated, and a crack slip type recognition model is obtained.

[0015] The numerical calculation model is established in detail as follows:

[0016] The geometric shape and size of the real object structure are measured, and the numerical calculation model is established;

[0017] The model is optimized, and the secondary structure on the structure that does not affect the propagation of the elastic wave is deleted;

[0018] The decoration and shape-retaining structure on the structure are deleted;

[0019] The material parameters of the object structure are obtained, and the parameters are assigned to the corresponding numerical calculation model.

[0020] The crack source is added in detail as follows:

[0021] According to the stress-strain distribution characteristics of the real object structure, the possible position of crack occurrence is determined;

[0022] The slip type of the crack is determined by randomly selecting a slip vector and a normal vector of the crack surface.

[0023] The crack size is selected as a constant according to the initial crack characteristics of the test object.

[0024] The calculation result output position is added, specifically:

[0025] According to the positions of the sensors on the real object structure, signal output positions are set for the numerical model, and the output positions are completely the same as the sensor positions.

[0026] According to the sensitive direction of the sensors on the real object structure, the output displacement components are set.

[0027] The amplitude of the signal curve at each position is picked up, specifically including:

[0028] The displacement curve along the sensitive direction of the sensor at the signal output position is calculated using a numerical method.

[0029] The displacement amplitude of the signal is extracted according to the tangent slope of the displacement curve data.

[0030] The displacement amplitudes at all signal output displacements are arranged in order into a column vector.

[0031] The equivalent moment tensor of the crack is calculated according to the set crack slip vector and normal vector, the proportions of the isotropic component, the double force couple component and the supplementary linear vector couple dipole component are obtained by decomposing the moment tensor, and the three proportions form a column vector, which is the crack type characteristic parameter.

[0032] The displacement amplitude column vector and the proportion of the three basic components of the moment tensor are combined together to form a training data pair, wherein the displacement amplitude column vector is the model input parameter, and the proportion of the three basic components of the moment tensor is the model output parameter.

[0033] The crack type characteristic parameter-signal amplitude data set is established, specifically including: randomly setting the crack surface slip vector and normal vector; storing each displacement amplitude column vector-proportion column vector data pair to form a data set.

[0034] The number of neurons in the input layer, the middle layer and the output layer is designed based on the feedforward neural network model.

[0035] The number of neurons in the input layer, the middle layer and the output layer is designed, specifically including:

[0036] The number of input layer neurons is designed according to the number of sensors, and the number of output layer neurons is equal to the number of sensors / signal output positions using a double-layer feedforward neural network;

[0037] For the double-layer feedforward neural network, the number of neurons contained in the hidden layer is not less than 3.

[0038] For the double-layer feedforward neural network, the proportion of each of the isotropic, double couple and supplementary linear vector dipole components of the crack moment tensor decomposition is output respectively, and sequentially corresponds to three elements of the proportion column vector.

[0039] The training neural network model; specifically:

[0040] The displacement amplitude column vector-moment tensor decomposition basic component proportion column vector data set is divided into training data set, test data set and validation data set according to the proportion;

[0041] The training data set is used to train the feedforward neural network;

[0042] The test data set is used to test the performance of the trained feedforward neural network, and the validation data set is used to verify the performance of the neural network.

[0043] The training result is evaluated, specifically:

[0044] The data correlation coefficient R value in the training, testing and validation process is analyzed;

[0045] If the model prediction accuracy meets the requirements, the model parameters are output and a usable prediction model is obtained; if the model prediction accuracy does not meet the requirements, step S6 is returned, the neural network model is trained, tested and evaluated by modifying the number of neural units in the intermediate layer, and the prediction accuracy is analyzed.

[0046] Beneficial effects: the numerical model is used to simulate the elastic wave field excited by crack surface slip, the elastic wave field-moment tensor training data set is established, the neural network model is trained using the data set, and a crack identification method applicable to a specific structure is obtained; the method does not depend on the explicit expression of the Green function, directly fits the mapping relationship between the elastic wave field characteristics and the crack slip equivalent moment tensor through the data-driven principle, and solves the problems of solving and type identification of the crack surface slip equivalent moment tensor in a complex structure. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 It is a model shape, crack source position and sensor arrangement form schematic diagram in the embodiment of the application;

[0048] Figure 2 It is a crack random slip type selection schematic diagram in the embodiment of the application;

[0049] Figure 3A schematic diagram of a feedforward neural network structure in an embodiment of the present application;

[0050] Figure 4 Neural network training, testing and verification results in an embodiment of the present application; wherein (a) is the training result; (b) is the verification result; (c) is the test result; (d) is the overall result. DETAILED DESCRIPTION

[0051] The present application will be further described below with reference to the accompanying drawings.

[0052] As Figure 1 shown, the crack slip type recognition method based on neural network and pre-training principle of the present application comprises the following steps:

[0053] S1: According to the geometric characteristics and material parameters of the real object structure, a numerical calculation model is established; specifically:

[0054] S11: Measure the geometric shape and size of the real object structure, and establish a numerical calculation model;

[0055] S12: Optimize the model as necessary, delete the secondary structures that do not affect the propagation of elastic waves on the structure, and delete or equivalently to the main structure of the structure surface skin, etc.

[0056] S13: Delete the decoration and shape-retaining structure on the structure;

[0057] S14: Obtain the material parameters of the object structure, and assign the parameters to the corresponding numerical calculation model.

[0058] S2: In the numerical model, add a crack source, and set the size, position and type of the crack; wherein adding a crack source specifically comprises the following steps:

[0059] S21: According to the stress and strain distribution characteristics of the real object structure, determine the possible position of the crack;

[0060] S22: The slip type of the crack is determined by randomly selecting the slip vector and normal vector of the crack surface;

[0061] S23: The crack size is selected as a constant according to the initial crack characteristics of the test object.

[0062] S3: In the numerical model, add a calculation result output position, and the output position corresponds to the arrangement position of the sensor in the real object structure; wherein adding the calculation result output position specifically comprises:

[0063] S31: According to the position of the sensor on the real object structure, set the signal output position for the numerical model. The output position should be exactly the same as the sensor position;

[0064] S32: According to the sensitive direction of the real object structure sensor, the displacement component of the output is set.

[0065] S4: Using the numerical model, the elastic wave field excited by crack slip / extension is calculated, the elastic wave time domain signal curve at the position set by S3 is output, and the amplitude of the signal curve at each position is picked up; Specifically, it includes:

[0066] S41: Using numerical method to calculate the displacement curve along the sensitive direction of the sensor at the signal output position;

[0067] S42: According to the tangent slope of the displacement curve data, the displacement amplitude of the signal is extracted;

[0068] S43: The displacement amplitudes at all signal output positions are arranged in order to form a column vector;

[0069] S44: According to the set crack slip vector and normal vector, the equivalent moment tensor of the crack is calculated, and the proportion of the three basic components of isotropic component (ISO), double couple component (DC) and complementary linear vector dipole component (CLVD) is obtained by decomposing the moment tensor. The three proportion column vectors are combined to form a column vector. The proportion column vector is the crack type characteristic parameter, and the crack slip type can be judged according to the column vector;

[0070] S45: The displacement amplitude column vector and the proportion of the three basic components of the moment tensor are combined together to form a training data pair. The displacement amplitude column vector is the model input parameter, and the proportion of the three basic components of the moment tensor is the model output parameter.

[0071] S5: Repeat steps S2-S4, randomly change the type of crack, keep other parameters unchanged, calculate the signal amplitude corresponding to different crack types, and establish the data set of crack type characteristic parameter-signal amplitude; Specifically, it includes:

[0072] S51: Randomly set the crack surface slip vector and normal vector, and repeat steps S2-S4.

[0073] S52: Store each set of displacement amplitude column vector-proportion column vector data pair to form a data set.

[0074] S6: Based on the feedforward neural network model, the number of neurons in the input layer, the middle layer and the output layer is designed; Specifically, it includes:

[0075] S61: Using double-layer feedforward neural network, the number of input layer neurons is designed according to the number of sensors. The number of output layer neurons is equal to the number of sensors / signal output positions;

[0076] S62: for the double-layer feedforward neural network, the number of hidden layers is set to 1, and the number of neurons contained in the hidden layer is not less than 3;

[0077] S63: for the double-layer feedforward neural network, the number of neurons contained in the output layer is 3, which respectively outputs the proportion of isotropic (ISO), double couple (DC) and complementary linear vector doublet (CLVD) components in the crack moment tensor decomposition, that is, the three elements of the proportion column vector in turn.

[0078] S7: the crack type-signal amplitude data set obtained in S5 is divided into training, test and validation data sets, and the neural network model is trained; specifically:

[0079] S71: the displacement amplitude column vector-moment tensor decomposition basic component proportion column vector data set obtained in S5 is divided into training data set, test data set and validation data set according to a certain proportion;

[0080] S72: the training data set is used to train the feedforward neural network, and the Levenberg-Marquardt algorithm is used as the training method;

[0081] S73: the test data set is used to test the performance of the trained feedforward neural network, and the validation data set is used to verify the performance of the neural network.

[0082] S8: evaluate the training result to obtain a crack slip type recognition model for the target structure in S1, specifically:

[0083] S81: analyze the data correlation coefficient R value in the training, test and validation process, the closer the R value is to 1, the stronger the correlation between the predicted value and the true value, and the higher the model prediction accuracy;

[0084] S82: if the model prediction accuracy meets the requirements, the model parameters are output and a usable prediction model is obtained; if the model prediction accuracy does not meet the requirements, step S6 can be returned, the number of neural units in the intermediate layer is modified, and steps S6-S8 are repeated to train, test and evaluate the neural network model, and the prediction accuracy is analyzed.

[0085] The application can improve the recognition accuracy of crack slip type in the structure, overcome the problem that the existing crack moment tensor inversion method cannot be applied to complex structure, and simplify the solution process of crack recognition. The crack moment tensor solving method based on neural network and pre-training principle can be applied to any structure, and can quantitatively and accurately analyze the crack slip type in the structure, and provide accurate data support for structure life assessment based on fracture mechanics theory.

[0086] Embodiment

[0087] This embodiment is aimed at Figure 1The cubic-shaped object structure establishes a crack slip type identification method. The object structure is a three-dimensional solid structure. According to the foregoing S1-S8 steps, the embodiment establishes a crack slip type identification method for the object structure based on a neural network model and a pre-training principle. Specifically, it includes the following steps:

[0088] S1: According to the geometric characteristics and material parameters of the real object structure, a numerical calculation model is established. The modeling method includes the following steps:

[0089] S11: Measure the geometric dimensions of the three-dimensional solid structure. The size of the three-dimensional solid structure is 1 m, and a numerical calculation model is established;

[0090] S12: Simplify the model as necessary. The model used in the embodiment is simple in structure and does not include connecting or skinning structures, so no further simplification is required. This step is omitted;

[0091] S13: Delete the decoration and shape retention structure on the structure. The model used in the embodiment is the simplest model structure, so there is no need to delete related structures. This step is omitted;

[0092] S14: Obtain the material parameters of the object structure. In the embodiment, the material of the three-dimensional solid structure is rock, and the elastic modulus, Poisson's ratio and density are 54 Gpa, 0.23 and 2300 kg / m3, respectively. Assign the material parameters to the numerical model and proceed to the subsequent steps.

[0093] S2: In the numerical model, add a crack source and set the size, position and type of the crack. Adding a crack source includes the following steps:

[0094] S21: According to the stress condition of the three-dimensional solid structure, it can be determined that the position of the crack that may occur in the three-dimensional solid structure is located inside the solid structure, with a depth of 0.5 meters from the upper surface;

[0095] S22: The crack slip type is determined by randomly selecting the crack surface slip vector and normal vector. As shown in Figure 2 The crack slip type is determined by the crack surface slip vector u and normal vector v. The crack slip type is set by randomly selecting the angle between vectors u and v (between 0 and 90 degrees);

[0096] S23: The crack size is selected according to the geometric characteristics of the possible cracks in the structure. The relevant data can be obtained by consulting relevant literature. For this example, the initial crack size of the rock is selected as 0.01.

[0097] S3: In the numerical model, add the output position of the calculation results, and the output position corresponds to the arrangement position of the sensor in the real object structure. The sensor position selection includes the following steps:

[0098] S31: According to the arrangement form of the sensor position on the real object structure, in this example, the sensor is installed on the upper surface of the structure, and the number of sensors is 6, of which 5 sensors are arranged at equal angles on the surrounding ring, and the remaining 1 sensor is arranged at the center point, as shown in Figure 1 ;

[0099] S32: According to the propagation direction of the elastic wave excited by the crack, the sensitive direction of the sensor is set along the vertical direction of the upper surface.

[0100] S4: Using a numerical model, calculate the elastic wave field excited by crack slip / expansion, output the elastic wave time domain signal curve at the position set in S3, pick up the amplitude of the signal curve at each position; Specifically, the following steps are included:

[0101] S41: Calculate the elastic wave field caused by crack slip using numerical calculation method, and output the elastic wave waveform at the sensor position;

[0102] S42: Extract the amplitude of the signal first arrival wave according to the tangent slope;

[0103] S43: Arrange the displacement amplitudes at all signal output positions in order as a column vector;

[0104] S44: According to the crack surface slip vector and normal vector set in step S2, calculate the moment tensor corresponding to the crack. Solve the eigenvalues of the moment tensor, and arrange the eigenvalues along the diagonal to form the principal moment tensor. Decompose the principal moment tensor into the proportions of three basic components (ISO, DC and CLVD components), and arrange the three proportion values in order as a column vector;

[0105] S45: Combine the displacement amplitude column vector and the proportion of three basic components of the moment tensor together to form a training data set. The displacement amplitude column vector is the input parameter, and the proportion of three basic components of the moment tensor is the output parameter.

[0106] S5: Repeat steps S2-S4, randomly change the type of crack, keep other parameters unchanged, calculate the signal amplitude corresponding to different crack types, and establish a data set of crack type characteristic parameters-signal amplitude; Specifically, the following steps are included:

[0107] S51: Randomly change the angle between the crack surface slip vector and the normal vector, and repeat steps S2-S4;

[0108] S52: For each calculation, store each set of displacement amplitude column vector-basic component proportion column vector data pair, repeat the calculation 3000 times, and obtain the data set of amplitude column vector-three basic component proportions.

[0109] S6: Based on as Figure 3The feedforward neural network model shown, the number of neurons in the input layer, the middle layer and the output layer is designed; including the following steps:

[0110] S61: using a double-layer feedforward neural network model, the neural network structure is established. The number of output layer neurons is equal to the number of sensors, and the number of sensors in this example is 6, so the input layer has 6 neurons, which correspond to the 6 elements of the displacement amplitude column vector respectively;

[0111] S62: the number of hidden layers of the neural network in this example is 1, and this layer has 3 neurons;

[0112] S63: the model output parameters are 3, which respectively output the proportion of isotropic (ISO), double couple (DC) and complementary linear vector doublet (CLVD) components in the crack tensor decomposition.

[0113] S7: the crack type-signal amplitude data set obtained in S5 is divided into training, test and validation data sets, and the neural network model is trained; including the following steps:

[0114] S71: the data set containing 3000 groups of data obtained in step S5 is divided into training data set, test and validation data set, and the proportion of each data set is 70%, 15% and 15% respectively;

[0115] S72: the neural network is trained using the training data set, and the training method adopts Levenberg-Marquardt algorithm;

[0116] S73: the feedforward neural network is trained using the training data set, and then the test data set is used to test the classification accuracy of the model for cracks, and the validation data set is used to verify the classification accuracy, and the corresponding fitting results are shown in Figure 4 (a)~(d).

[0117] S8: evaluate the training results

[0118] From Figure 4 The fitting results show that in the neural network obtained by training, the correlation coefficient between the true value and the predicted value is close to 1, and the model can better fit the mapping relationship between the displacement amplitude column vector and the three basic component column vectors. Therefore, the neural network meets the prediction accuracy requirement and can be used for crack slip type identification of real structure.

Claims

1. A crack slip type identification method based on neural networks and pre-training principles, characterized in that, Includes the following steps: S1: Establish a numerical calculation model based on the geometric features and material parameters of the real object structure; S2: In the numerical calculation model, add crack sources and set the size, location, and type of the cracks; S3: In the numerical calculation model, add the output positions of the calculation results, and the output positions correspond to the sensor placement positions in the real object structure. S4: Using a numerical calculation model, calculate the elastic wave field excited by crack slip / propagation, output the elastic wave time-domain signal curve at the position set in S3, and pick up the amplitude of the signal curve at each position. Specifically, picking up the amplitude of the signal curve at each position includes: The displacement curve along the sensor's sensitive direction at the signal output position is calculated using numerical methods. The displacement amplitude of the signal is extracted based on the slope of the tangent line in the displacement curve data; Arrange the displacement amplitudes at all signal output displacements into a column vector in order; The equivalent moment tensor of the crack is calculated based on the set crack slip vector and normal vector. The moment tensor is decomposed to obtain the proportions of isotropic components, bicouple components and supplementary linear vector dipole components. The three proportions are combined into a column vector, which is the crack type characteristic parameter. The displacement magnitude column vector and the proportions of the three basic components of the moment tensor are combined to form training data pairs, where the displacement magnitude column vector is the model input parameter and the proportions of the three basic components of the moment tensor are the model output parameters. S5: Repeat steps S2-S4, randomly change the type of crack, keep other parameters unchanged, calculate the signal amplitude corresponding to different crack types, and establish a dataset of crack type feature parameters and signal amplitude. S6: Design the number of neurons in the input layer, intermediate layer, and output layer; S7: Divide the crack type-signal amplitude dataset obtained in S5 into training, testing, and validation datasets, and train the neural network model; S8: Evaluate the training results and obtain a crack slip type recognition model.

2. The crack slip type identification method based on neural network and pre-training principle according to claim 1, characterized in that, The establishment of the numerical calculation model specifically involves: Measure the geometry and dimensions of the real object structure and establish a numerical calculation model; The model was optimized by removing secondary structures that did not affect the propagation of elastic waves. Remove structural decorations and shape-preserving structures; Obtain the material parameters of the object structure and assign the parameters to the corresponding numerical calculation model.

3. The crack slip type identification method based on neural network and pre-training principle according to claim 2, characterized in that, The addition of crack sources specifically includes the following steps: Based on the stress-strain distribution characteristics of the actual object structure, the possible locations where cracks may occur can be determined. The slip type of the crack is determined by randomly selecting the slip vector and normal vector of the crack surface; The crack size is selected as a constant value based on the initial crack characteristics of the test object.

4. The crack slip type identification method based on neural network and pre-training principle according to claim 3, characterized in that, The specific steps for adding the output location of the calculation results are as follows: Based on the position of the sensor on the actual object structure, the signal output position is set for the numerical model, and the output position is exactly the same as the sensor position. The output displacement component is set according to the sensor's sensitive direction on the actual object structure.

5. The crack slip type identification method based on neural network and pre-training principle according to claim 1, characterized in that, The establishment of the crack type feature parameter-signal amplitude dataset specifically includes: randomly selecting and setting the crack surface slip vector and normal vector; storing each set of displacement amplitude column vector-proportion column vector data pairs to form the dataset.

6. The crack slip type identification method based on neural network and pre-training principle according to claim 5, characterized in that, The number of neurons in the input layer, intermediate layer, and output layer is designed based on a feedforward neural network model.

7. The crack slip type identification method based on neural network and pre-training principle according to claim 6, characterized in that, The number of neurons in the input layer, intermediate layer, and output layer is specifically included as follows: Using a two-layer feedforward neural network, the number of neurons in the input layer is designed according to the number of sensors, and the number of neurons in the output layer is equal to the number of sensors / the number of signal output positions; For a two-layer feedforward neural network, the hidden layer contains no fewer than three neurons; For a two-layer feedforward neural network, the proportions of isotropic, bicouple, and supplementary linear vector dipole components of the crack moment tensor decomposition are output respectively, corresponding to the three elements of the proportion column vector.

8. The crack slip type identification method based on neural network and pre-training principle according to claim 7, characterized in that, The trained neural network model is specifically: The displacement magnitude column vector-moment tensor decomposition basic component proportion column vector dataset is divided into training dataset, test dataset and validation dataset according to the proportion; Train the feedforward neural network using the training dataset; Test the performance of the trained feedforward neural network using the test dataset, and validate the performance of the neural network using the validation dataset.

9. A crack slip type identification method based on neural networks and pre-training principles according to claim 8, characterized in that, The evaluation of the training results specifically includes: Analyze the correlation coefficients of data during the training, testing, and validation processes. R value; If the model's prediction accuracy meets the requirements, the model parameters are output and a usable prediction model is obtained; if the model's prediction accuracy does not meet the requirements, the process returns to step S6, where the neural network model is trained, tested, and evaluated by modifying the number of neural units in the intermediate layers, and the prediction accuracy is analyzed.

Images