Nonlinear ultrasonic multi-harmonic multi-parameter decoupling identification method and system for micro-crack group

Abstract

This invention discloses a nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method and system for microcrack clusters, belonging to the field of ultrasonic nondestructive testing technology. The method includes: S1, acquiring the time-domain nonlinear ultrasonic response signal of the defect caused by multi-parameter feature coupling of the microcrack cluster; S2, data preprocessing, dividing the preprocessed data into training and testing sets; S3, constructing a deep belief network; and S4, importing the acquired microcrack cluster response signal into the deep belief network to establish a multi-harmonic nonlinear ultrasonic decoupling identification model. This invention leverages the powerful feature extraction and signal processing advantages of deep learning, utilizing four nonlinear ultrasonic effects generated by multi-parameter coupling of microcrack clusters to construct a deep belief network, classifying and identifying different distribution center deviations and different average discrete sizes of the microcrack clusters. This addresses the problems of traditional crack identification, which mainly relies on manual labor, resulting in high costs, long processing times, and low reliability.

Description

Technical Field

[0001] This invention belongs to the field of ultrasonic nondestructive testing technology, specifically relating to a nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method and system for microcrack clusters. Background Technology

[0002] Currently, additive manufacturing technology is considered one of the key technologies for technological development and industrial transformation. Its advantages over traditional manufacturing technologies lie mainly in one-piece molding, a wide variety of materials, low production costs, and short manufacturing cycles, making it widely used in the production of small-batch, complex, and time-sensitive parts. However, due to the layered manufacturing principle of additive manufacturing, the rapid heating and cooling process during the layer-by-layer forming process easily leads to early defects such as microcrack clusters, voids, and residual stress within the metal workpiece. Furthermore, research has shown that microcrack clusters with a concentration effect have a greater impact on the mechanical properties of the structure and the service life of the workpiece. The principle of nonlinear ultrasonic testing for microcracks mainly involves studying the strong nonlinear response caused by the interaction between microcracks and ultrasonic waves. Utilizing the high-order harmonics, subharmonics, DC components, and mixed-frequency signals generated by the interaction of ultrasonic signals with microcracks, microcracks smaller than half a wavelength can be accurately detected, offering significant advantages over other detection technologies in the characteristic evaluation of microcracks.

[0003] Research has found that internal defects in additively manufactured structural components often exist in the form of microcrack clusters. The complex coupling effect of the multi-parameter characteristics of microcrack clusters can cause problems such as difficulty in feature extraction, low recognition efficiency, and coarse crack information recognition in nonlinear ultrasonic testing. Current research on microcrack clusters only focuses on the identification of single cracks or provides a general qualitative description of microcrack clusters based on density, lacking specific quantitative evaluation of the number of deviations from the distribution center of different microcrack clusters and different average discrete sizes. Among these, the size of the microcrack cluster affects the crack propagation rate, while the number affects the crack propagation behavior and the service life of the workpiece.

[0004] Therefore, designing and developing a nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method and system for microcrack clusters is of great significance. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method and system for microcrack clusters. Relying on the powerful feature extraction and signal processing advantages of deep learning, a deep belief network model is constructed using four nonlinear ultrasonic effects generated by the interaction between the incident longitudinal wave and the microcrack cluster. This model classifies and identifies microcrack clusters based on the number of deviations from their distribution centers and different average discrete sizes. This solves the problems of high cost, long time consumption, and low reliability associated with traditional crack identification, which mainly relies on manual labor.

[0006] The specific technical solution adopted in this invention is as follows:

[0007] The primary objective of this patent is to provide a nonlinear ultrasonic multi-harmonic multi-feature decoupling and identification method for microcrack clusters, including:

[0008] S1. Acquire the time-domain nonlinear ultrasonic response signal of the defect with multi-parameter characteristic coupling of microcrack groups, specifically:

[0009] First, a microcrack cluster model is constructed, which includes microcrack clusters with different areas, different number ranges and their deviations, different average sizes and different deviations. Then, a 1MHz longitudinal wave is loaded onto the microcrack cluster model. Finally, m transmission wave time-domain nonlinear ultrasonic response signals of the defects are collected at m receiving points after the longitudinal wave passes through the microcrack cluster region; M is a natural number greater than 0.

[0010] S2. Data preprocessing, specifically:

[0011] First, the time-domain nonlinear ultrasonic response signals of m defects are filtered to obtain the multi-harmonic nonlinear ultrasonic signals generated by the interaction of ultrasonic signals and microcracks. The multi-harmonic nonlinear ultrasonic signals include the zero-frequency ultrasonic signal, the frequency-division harmonic ultrasonic signal, the second harmonic ultrasonic signal, and the third harmonic ultrasonic signal of each defect time-domain nonlinear ultrasonic response signal.

[0012] Then, the ultrasonic signals from different receiving points are combined, and white noise is added to enhance the data to obtain preprocessed data;

[0013] Finally, the preprocessed data is divided into a training set and a test set;

[0014] S3. Construct a deep belief network model, specifically as follows:

[0015] First, an initial network model is constructed; then, the training set is used to train the initial network model, and the test set is used for the identification of various features of microcrack clusters to obtain the identification results of each harmonic nonlinear signal for different feature parameters; finally, based on the feature information of microcrack clusters in the modeling, the initial network model is adjusted to obtain a deep belief network model.

[0016] S4. The acquired time-domain nonlinear ultrasonic response signal of the defect is imported into a deep belief network to establish a multi-harmonic nonlinear ultrasonic decoupling identification model. The multi-harmonic nonlinear ultrasonic decoupling identification model is used to identify multi-harmonic nonlinear ultrasound.

[0017] Preferably, S1 specifically includes:

[0018] A microcrack swarm model with different areas, different number ranges and their deviations, different average sizes and different deviations was constructed using the finite element software ABAQUS.

[0019] m receiving points are set at 5 mm from the right boundary of the crack area to receive the transmitted time-domain nonlinear ultrasonic response signal of the defect.

[0020] Preferably, when filtering the time-domain nonlinear ultrasonic response signals of m defects, the filtering process for each defect's time-domain nonlinear ultrasonic response signal specifically includes:

[0021] The zero-frequency ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack is obtained by low-pass filtering the time-domain nonlinear ultrasonic response signal of the defect with a 0.2MHz filter.

[0022] The time-domain nonlinear ultrasonic response signal of the defect is bandpass filtered by 0.25MHz to 0.75MHz to obtain the frequency-division harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack;

[0023] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered at 1.75MHz to 2.25MHz to obtain the second harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0024] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered by 2.75MHz to 3.25MHz to obtain the third harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0025] Preferably, the process of combining signals from different receiving points is as follows:

[0026] Zero-frequency ultrasonic signals were extracted from the time-domain nonlinear ultrasonic response signals of each defect to form a zero-frequency dataset;

[0027] The frequency-division harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a frequency-division harmonic dataset;

[0028] The second harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a second harmonic dataset;

[0029] The third harmonic ultrasonic signal of the time-domain nonlinear ultrasonic response signal of each defect is extracted to form a third harmonic dataset.

[0030] Preferably, M=3, and the deep belief network model includes a visible layer, a hidden layer, and an output layer; the initial network model includes three stacked restricted Boltzmann machines; each restricted Boltzmann machine consists of a visible layer and a hidden layer; the first hidden layer contains 512 neurons, the second hidden layer contains 256 neurons, and the third hidden layer contains 128 neurons, and all three hidden layers use ReLU as the activation function to introduce nonlinearity; the output layer uses softmax as the activation function and contains n neurons, the value of n being determined according to the microcrack cluster characteristics, specifically: n=3 when identifying the microcrack number range and its deviation, and n=5 when identifying the statistical average size of microcracks and its deviation.

[0031] A second objective of this invention is to provide a nonlinear ultrasonic multi-harmonic multi-feature decoupling identification system for microcrack clusters, comprising:

[0032] Defect time-domain nonlinear ultrasonic response signal acquisition module: First, a microcrack group model is constructed, which includes microcrack groups with different numbers, different statistical average sizes, and different deviations; then, a 1MHz longitudinal wave is loaded onto the microcrack group model; finally, m transmitted wave defect time-domain nonlinear ultrasonic response signals are acquired at m receiving points after the longitudinal wave passes through the microcrack group region; M is a natural number greater than 0.

[0033] The data preprocessing module first filters the m time-domain nonlinear ultrasonic response signals of defects to obtain multi-harmonic nonlinear ultrasonic signals generated by the interaction of ultrasonic signals and microcracks. These multi-harmonic nonlinear ultrasonic signals include the zero-frequency ultrasonic signal, frequency-division harmonic ultrasonic signal, second harmonic ultrasonic signal, and third harmonic ultrasonic signal of each defect's time-domain nonlinear ultrasonic response signal. Then, the ultrasonic signals from different receiving points are combined, and white noise is added to enhance the data and obtain preprocessed data. Finally, the preprocessed data is divided into a training set and a test set.

[0034] The deep belief network model construction module first constructs an initial network model; then, the training set is used to train the initial network model, and the test set is used for the identification of various features of microcrack clusters to obtain the identification results of each harmonic nonlinear signal for different feature parameters; finally, the initial network model is adjusted according to the feature information of the microcrack clusters in the modeling to obtain the deep belief network model.

[0035] Decoupling identification module: The acquired time-domain nonlinear ultrasonic response signal of the defect is imported into a deep belief network to establish a multi-harmonic nonlinear ultrasonic decoupling identification model, and the multi-harmonic nonlinear ultrasonic identification is performed using the multi-harmonic nonlinear ultrasonic decoupling identification model.

[0036] Preferably, the defect time-domain nonlinear ultrasonic response signal acquisition process includes:

[0037] A microcrack group model was constructed using the finite element software ABAQUS;

[0038] m receiving points are set at 5 mm from the right boundary of the crack area to receive the transmitted time-domain nonlinear ultrasonic response signal of the defect.

[0039] Preferably, when filtering the time-domain nonlinear ultrasonic response signals of m defects, the filtering process for each defect's time-domain nonlinear ultrasonic response signal specifically includes:

[0040] The zero-frequency ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack is obtained by low-pass filtering the time-domain nonlinear ultrasonic response signal of the defect with a 0.2MHz filter.

[0041] The time-domain nonlinear ultrasonic response signal of the defect is bandpass filtered by 0.25MHz to 0.75MHz to obtain the frequency-division harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack;

[0042] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered at 1.75MHz to 2.25MHz to obtain the second harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0043] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered by 2.75MHz to 3.25MHz to obtain the third harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0044] Preferably, the process of combining signals from different receiving points is as follows:

[0045] Zero-frequency ultrasonic signals were extracted from the time-domain nonlinear ultrasonic response signals of each defect to form a zero-frequency dataset;

[0046] The frequency-division harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a frequency-division harmonic dataset;

[0047] The second harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a second harmonic dataset;

[0048] The third harmonic ultrasonic signal of the time-domain nonlinear ultrasonic response signal of each defect is extracted to form a third harmonic dataset.

[0049] Preferably, M=3, and the deep belief network model includes a visible layer, a hidden layer, and an output layer; the initial network model includes three stacked restricted Boltzmann machines; each restricted Boltzmann machine consists of a visible layer and a hidden layer; the first hidden layer contains 512 neurons, the second hidden layer contains 256 neurons, and the third hidden layer contains 128 neurons, and all three hidden layers use ReLU as the activation function to introduce nonlinearity; the output layer uses softmax as the activation function and contains n neurons, the value of n being determined according to the microcrack cluster characteristics, specifically: n=3 when identifying the microcrack number range and its deviation, and n=5 when identifying the statistical average size of microcracks and its deviation.

[0050] By adopting the above technical solution, the present invention has the following technical effects:

[0051] 1. Current research on single microcrack parameters fails to accurately reflect actual loss conditions, severely impacting the accuracy of workpiece material mechanical property assessment. This invention addresses the varying impacts of different distribution center deviations and average discrete size parameters of microcrack clusters on material lifespan. Using Abaqus finite element software, an ultrasonic testing simulation model of microcrack clusters is constructed, incorporating different distribution center deviations and average discrete size parameters. The quantity range and its deviation are set with a specific deviation based on the distribution center, and the size feature is similarly set with a specific dispersion based on the average size, making it more consistent with actual loss conditions compared to single cracks. Under the coupled conditions of different distribution center deviations and average discrete sizes, a deep belief network is used for decoupling, eliminating mutual interference between the two features. This allows for the extraction and classification of feature information from the nonlinear signal for both different distribution center deviations and average discrete sizes.

[0052] 2. By receiving defect signals at the receiving point and processing them, nonlinear signals of different harmonic orders (zero-frequency, sub-harmonics, second harmonic, and third harmonic) can be obtained. A deep belief network model is established to decouple and identify two characteristics of microcrack clusters: the number of deviations from the distribution center and the average discrete size. Based on the statistical quantitative identification results of the coupling parameters of microcrack clusters using the deep belief network model, both the number of deviations from the distribution center and the average discrete size exhibit high identification accuracy, demonstrating the effectiveness of the constructed deep belief network model in decoupling these two characteristics.

[0053] 3. In this invention, different nonlinear signals exhibit varying recognition performance for different deviations of the distribution center of microcrack clusters and different average discrete sizes. The third harmonic recognition model converges fastest and achieves the highest final recognition accuracy. The second harmonic recognition model has lower convergence speed and recognition accuracy on the test set than the third harmonic model. The zero-frequency and sub-frequency harmonic models show the lowest recognition performance and the lowest average recognition accuracy, and are difficult to converge. Both exhibit consistency: different harmonics show significant differences in sensitivity to different deviations of the distribution center of microcrack clusters and different average discrete sizes. The network performs best when the third harmonic signal is input, which is expected to provide a theoretical basis for practical detection. Attached Figure Description

[0054] Figure 1 A flowchart of a preferred embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of a defect signal acquired in a preferred embodiment of the present invention. The left side is a time domain diagram, and the right side is a frequency domain diagram.

[0056] Figure 3 In a preferred embodiment of the present invention, Figure 2 A schematic diagram of the zero-frequency signal after processing the defect signal in the middle;

[0057] Figure 4 In a preferred embodiment of the present invention, Figure 2 A schematic diagram of the time domain of the frequency-divided harmonic signal after processing the defect signal;

[0058] Figure 5 In a preferred embodiment of the present invention, Figure 2 A schematic diagram of the second harmonic signal after processing the defect signal in the middle;

[0059] Figure 6 In a preferred embodiment of the present invention, Figure 2 A schematic diagram of the time domain of the third harmonic signal after processing the defect signal;

[0060] Figure 7 In a preferred embodiment of the present invention, Figure 3 Schematic diagram of data enhancement by adding white noise with different signal-to-noise ratios to the mid-zero frequency signal;

[0061] Figure 8 In a preferred embodiment of the present invention, Figure 4 Schematic diagram of data enhancement for mid-frequency harmonic signals by adding white noise with different signal-to-noise ratios;

[0062] Figure 9 In a preferred embodiment of the present invention, Figure 5 Schematic diagram of data enhancement for middle second harmonic signal by adding white noise with different signal-to-noise ratios;

[0063] Figure 10 In a preferred embodiment of the present invention, Figure 6 Schematic diagram of data enhancement for the middle third harmonic signal by adding white noise with different signal-to-noise ratios;

[0064] Figure 11 This is a time-domain diagram of the zero-frequency signal from multiple receiving points with the same microcrack group characteristics in a preferred embodiment of the present invention.

[0065] Figure 12 This is a time-domain diagram of the frequency-division harmonic signals from multiple receiving points with the same microcrack group characteristics in a preferred embodiment of the present invention;

[0066] Figure 13 This is a time-domain diagram of the second harmonic signal from multiple receiving points of the same microcrack group in a preferred embodiment of the present invention.

[0067] Figure 14 This is a time-domain diagram of the third harmonic signal from multiple receiving points with the same microcrack group characteristics in a preferred embodiment of the present invention.

[0068] Figure 15 This is a structural diagram of the deep belief network model in a preferred embodiment of the present invention;

[0069] Figure 16 This is a curve showing the accuracy and loss value of zero-frequency signal for identifying quantity ranges and their deviation characteristics in a preferred embodiment of the present invention.

[0070] Figure 17 The curves showing the accuracy and loss values ​​of the frequency-division harmonic signal for identifying the quantity range and its deviation characteristics in a preferred embodiment of the present invention are shown.

[0071] Figure 18 This is a curve showing the accuracy and loss value of the second harmonic signal for identifying the quantity range and its deviation characteristics in a preferred embodiment of the present invention.

[0072] Figure 19 The curves showing the accuracy and loss value of the third harmonic signal for identifying the quantity range and its deviation characteristics in a preferred embodiment of the present invention are shown.

[0073] Figure 20 The curves showing the accuracy and loss values ​​of the zero-frequency signal for identifying size ranges and their discreteness features in a preferred embodiment of the present invention are shown.

[0074] Figure 21 The curves showing the accuracy and loss values ​​of the frequency-division harmonic signal for identifying size ranges and their discreteness characteristics in a preferred embodiment of the present invention are shown.

[0075] Figure 22 This is a curve showing the accuracy and loss value of the second harmonic signal for identifying size ranges and their discreteness characteristics in a preferred embodiment of the present invention.

[0076] Figure 23 The curves showing the accuracy and loss values ​​of the third harmonic signal for identifying size ranges and their discreteness characteristics in a preferred embodiment of the present invention are shown.

[0077] Figure 24 This is the confusion matrix for identifying the quantity range and its deviation characteristics of the zero-frequency signal in a preferred embodiment of the present invention;

[0078] Figure 25 This is the confusion matrix for identifying the frequency division harmonic signal for its quantity range and deviation characteristics in a preferred embodiment of the present invention;

[0079] Figure 26 This is the confusion matrix for identifying the quantity range and deviation characteristics of the second harmonic signal in a preferred embodiment of the present invention;

[0080] Figure 27 This is a confusion matrix for identifying the quantity range and deviation characteristics of the third harmonic signal in a preferred embodiment of the present invention.

[0081] Figure 28 This is the confusion matrix for identifying the size range and its discreteness characteristics of the zero-frequency signal in a preferred embodiment of the present invention;

[0082] Figure 29 This is the confusion matrix for identifying the size range and its discreteness characteristics of the frequency division harmonic signal in a preferred embodiment of the present invention;

[0083] Figure 30 This is the confusion matrix for identifying the size range and its discreteness characteristics of the second harmonic signal in a preferred embodiment of the present invention;

[0084] Figure 31 This is a confusion matrix for identifying the size range and its discreteness characteristics of the third harmonic signal in a preferred embodiment of the present invention. Detailed Implementation

[0085] To further understand the invention's content, features, and effects, the following embodiments are provided, and detailed descriptions are given in conjunction with the accompanying drawings.

[0086] Please see Figures 1 to 31 As shown:

[0087] The English translations in the illustration are as follows:

[0088] Amplitude: range;

[0089] Frequency: frequency;

[0090] Time: time period;

[0091] Curves for Each Row: Curves for each row;

[0092] Value: Amplitude;

[0093] InputLayer: Input layer;

[0094] Flatten: a layer that has been flattened out;

[0095] Dense: Fully connected layer;

[0096] Epoch: Number of iterations;

[0097] Training and validation accuracy: The accuracy rate of training and validation.

[0098] Training and Validation loss: The loss value between training and validation.

[0099] True Label: The actual label;

[0100] Predicted Label: The predicted label;

[0101] Normalized Confusion Matrix: The matrix used to normalize confusion.

[0102] Model accuracy: Model accuracy;

[0103] model-loss: The model loss value.

[0104] A nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method for microcrack clusters, specifically including:

[0105] S1. Acquire the time-domain nonlinear ultrasonic response signal of the defect, specifically:

[0106] First, a microcrack cluster model is constructed, comprising microcrack clusters with different areas, different number ranges and their deviations, different average sizes and different deviations. Then, a 1MHz longitudinal wave is applied to the microcrack cluster model. Finally, m transmission wave defect time-domain nonlinear ultrasonic response signals are acquired at m receiving points after the longitudinal wave passes through the microcrack cluster region; M is a natural number greater than 0. This step S1 can be implemented using the following specific technical means:

[0107] Microcrack swarm models with different areas, number ranges and their deviations, different average sizes and different deviations and dispersions were established using the finite element software ABAQUS. An ultrasonic wave propagation model within a damaged elastic body was also established. An ultrasonic excitation signal of 10 periods and 1MHz Hanning window modulated sinusoidal wave with an amplitude of 1×10⁻⁶ was applied to the left side of the microcrack swarm model.-3 mm; the longitudinal wave propagates along the x-direction in the model. After the longitudinal wave passes through the microcrack cluster region, three different receiving points are set at 5 mm on the right boundary of the crack region to receive the transmitted wave signal; in order to ensure that the complete defect signal can be collected, the sampling frequency of the receiving points is set to 200 MHz.

[0108] S2. Data preprocessing, specifically:

[0109] First, the time-domain nonlinear ultrasonic response signals of m defects are filtered to obtain the multi-harmonic nonlinear ultrasonic signals generated by the interaction of ultrasonic signals and microcracks. The multi-harmonic nonlinear ultrasonic signals include the zero-frequency ultrasonic signal, the frequency-division harmonic ultrasonic signal, the second harmonic ultrasonic signal, and the third harmonic ultrasonic signal of each defect time-domain nonlinear ultrasonic response signal.

[0110] Then, the ultrasonic signals from different receiving points are combined, and white noise is added to enhance the data to obtain preprocessed data;

[0111] Finally, the preprocessed data is divided into a training set and a test set; wherein, step S2 can be implemented using the following specific technical means:

[0112] The defect time-domain nonlinear ultrasonic response signal received at each receiving point is subjected to a 0.25MHz low-pass filter and band-pass filters of 0.25MHz-0.75MHz, 1.75-2.25MHz, and 2.75-3.25MHz, respectively, to obtain four nonlinear ultrasonic signals: zero-frequency signal, frequency-division harmonic signal, second harmonic signal, and third harmonic signal generated by the interaction of the ultrasonic longitudinal wave signal with the microcrack. Specifically, the filtering process for each defect time-domain nonlinear ultrasonic response signal includes:

[0113] The zero-frequency ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack is obtained by low-pass filtering the time-domain nonlinear ultrasonic response signal of the defect with a 0.2MHz filter.

[0114] The time-domain nonlinear ultrasonic response signal of the defect is bandpass filtered by 0.25MHz to 0.75MHz to obtain the frequency-division harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack;

[0115] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered at 1.75MHz to 2.25MHz to obtain the second harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0116] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered by 2.75MHz to 3.25MHz to obtain the third harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0117] The process of combining signals from different receiving points is as follows:

[0118] Zero-frequency ultrasonic signals were extracted from the time-domain nonlinear ultrasonic response signals of each defect to form a zero-frequency dataset;

[0119] The frequency-division harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a frequency-division harmonic dataset.

[0120] The second harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a second harmonic dataset;

[0121] The third harmonic ultrasonic signal of the time-domain nonlinear ultrasonic response signal of each defect is extracted to form a third harmonic dataset.

[0122] After obtaining each nonlinear signal, the first step is to normalize the nonlinear signal to 0-1, mapping the data to the interval [0, 1].

[0123]

[0124] Where: X scaled The data is normalized, and X is the original data. min X is the minimum value in the original data. max This represents the maximum value in the original data.

[0125] By using signals from multiple receivers, microcrack defect information can be extracted more comprehensively. One-dimensional time-domain signals from n data points collected by m receivers with the same microcrack group parameters are stacked to form an m×n two-dimensional matrix. White noise with different signal-to-noise ratios is added to the sample signals for data augmentation to expand the sample size, avoid over-reliance on the original data format, and improve the model's generalization ability.

[0126] SNR=10*log10(Ps / Pn) (2);

[0127] Where: SNR is the signal-to-noise ratio, Ps is the power of the signal, and Pn is the power of the noise.

[0128] After the above data preprocessing, the one-dimensional defect signals received by each receiving point of the microcrack group with the same number and size parameters are processed to obtain m×n two-dimensional data with different nonlinear signals.

[0129] S3. Construct a deep belief network model, specifically as follows:

[0130] First, an initial network model is constructed; then, the training set is used to train the initial network model, and the test set is used for the identification of various features of microcrack clusters to obtain the identification results of each harmonic nonlinear signal for different feature parameters; finally, based on the feature information of the microcrack clusters in the modeling, the initial network model is adjusted to obtain a deep belief network model; wherein, step S3 can be implemented using the following specific technical means:

[0131] A deep belief network (DBN) was constructed for classification and identification, comprising three stacked Restricted Boltzmann Machines (RBMs). The initial weight matrix of the network was set. The network model was trained using a pre-defined training set, and the test set was used to identify various features of microcrack clusters, obtaining the identification results of each harmonic nonlinear signal for different feature parameters. Finally, the sensitivity of different nonlinear signals to each microcrack cluster feature was investigated. Specifically, a single-signal DBN was constructed using different nonlinear signal inputs. The DBN consisted of three stacked RBMs. Each RBM consisted of a visible layer and a hidden layer: the first hidden layer contained 512 neurons, the second contained 256 neurons, and the third contained 128 neurons. ReLU was used as the activation function in all three hidden layers to introduce nonlinearity. Finally, a softmax layer was used for classification, the specific number of neurons determined according to the different microcrack cluster parameters. The optimizer was an ADMA optimizer, and the cross-entropy loss function was used. Furthermore, L2 regularization was used to prevent overfitting. For each single nonlinear signal, 9720 samples were obtained, with the training and test sets divided in a 4:1 ratio (7776 samples for training and 1944 samples for testing). The training set of each sample was fed into its respective network model for training, and finally, the test set was used for testing to obtain the identification results of the number and size of microcrack clusters for each nonlinear signal.

[0132] S4. The acquired time-domain nonlinear ultrasonic response signal of the defect is imported into a deep belief network to establish a multi-harmonic nonlinear ultrasonic decoupling identification model. The multi-harmonic nonlinear ultrasonic decoupling identification model is used to identify multi-harmonic nonlinear ultrasound.

[0133] To verify the effectiveness of the proposed multi-harmonic nonlinear ultrasound-based method for multi-feature quantification identification of microcrack clusters, microcrack cluster models with different center deviations and average discrete size parameters, as well as an ultrasonic wave propagation model in a damaged elastic body, were established using Abaqus. The transmitted wave signals after the excitation longitudinal wave passed through the microcrack cluster were received at multiple receiving points. Figure 2 As shown, the quantity range parameter in the microcrack swarm model is set to 80 to 240 with a fixed step size of 10. Based on the distribution center, no offset, 15% offset, and 30% offset are set. The distribution center is calculated based on the Manhattan distance of the cracks, and the cluster sum of squares is calculated and normalized according to the formula for the sum of squares within the crack cluster. The quantity range and its deviation characteristics are divided into three categories. The average size parameter of the microcrack swarm is set to 110μm to 300μm with a fixed step size of 10μm. Simultaneously, based on the size of a portion of the microcracks relative to the overall average size, dispersions of 10%, 20%, and 30% are set. Based on the standard normal distribution, the size of each microcrack falls within the set dispersion range, and the size characteristics are divided into five categories.

[0134] The arrival wave packet is calculated based on the wave velocity and then processed. The received defect time-domain nonlinear ultrasonic response signal is subjected to a 0.2MHz low-pass filter to obtain the zero-frequency signal, a 0.25-0.75MHz band-pass filter to obtain the frequency-division harmonic signal with a center frequency of 0.5MHz, a 1.75-2.25MHz band-pass filter to obtain the second harmonic signal with a center frequency of 2MHz, and a 2.75-3.25MHz band-pass filter to obtain the third harmonic signal with a center frequency of 3MHz—four types of nonlinear ultrasonic signals. Figures 3 to 6 As shown; simultaneously, to avoid overfitting during training, data augmentation is employed by adding white noise to the sample signals to expand the sample size, prevent the model from over-relying on the original form of the data, and improve the model's generalization ability, such as... Figures 7 to 10 As shown; since the wavelength of the excitation longitudinal wave is much larger than the size of the microcrack, the signal from a single receiving point cannot fully represent the defect information. Therefore, the one-dimensional nonlinear signals collected from three receiving points with the same microcrack group parameters are combined to form a 3×2800 two-dimensional matrix. Input samples of nonlinear signals of different harmonic orders are shown in the figure. Figures 11 to 14 As shown.

[0135] Based on the data preprocessing, a total of 54×60×3=9720 samples were obtained. To ensure that the number of data sets is the same as the number of labels, the data sets were divided into training sets and test sets in a 4:1 ratio. The training set contains 7776 samples and the test set contains 1944 samples.

[0136] Deep Belief Network (DBN) is a deep learning network model composed of multiple stacked Restricted Boltzmann Machines (RBMs). It possesses powerful feature extraction and processing capabilities. A DBN mainly consists of a visible layer, hidden layers, and an output layer. The connections between each layer are bidirectional, meaning that nodes in the previous layer are connected to all nodes in the next layer. The DBN designed in this invention consists of three stacked Restricted Boltzmann Machines; each Restricted Boltzmann Machine comprises one visible layer and one hidden layer: the first hidden layer contains 512 neurons, the second contains 256 neurons, and the third contains 128 neurons. All three hidden layers use ReLU as the activation function to introduce nonlinearity. The output layer uses softmax as the activation function and contains n neurons. The value of n is mainly determined based on the microcrack cluster features; that is, the number of different feature categories, n, varies. Based on the label processing in the code, it corresponds to the unique value of the label. When identifying the range of microcrack numbers and their deviations, n=3, such as... Figure 15 As shown, when identifying the statistical mean size and deviation of microcrack clusters, n=5.

[0137] For other parameter selection, the ADMA optimizer is used, employing a unique encoding method when constructing labels for different samples. Therefore, the cross-entropy loss function (categorical cross-entropy) is used, the batch size is set to 128, and the number of iterations is set to 500. Parameter initialization is a crucial step in training deep belief networks. Good parameter initialization can help the model converge faster and avoid problems such as vanishing or exploding gradients. This invention uses information on the deviation of different distribution centers and different average discrete sizes in the modeling of microcrack clusters to calculate the stress values ​​for predicting crack propagation, forming the initial weight matrix of the network. This allows the model to converge to a suitable solution more quickly, improving the stability and speed of training.

[0138] Data grouping is performed using a five-fold verification method to avoid highly similar samples. Taking the identification of quantity ranges and their deviation characteristics as an example... Figures 16 to 19 The curves show the accuracy and loss values ​​for identifying the number of deviations from the distribution center of different microcrack clusters using different nonlinear signals of the network. Figures 20 to 23 The curves show the accuracy and loss values ​​of identifying different size ranges and discreteness characteristics of microcrack clusters using different nonlinear signals from this network.

[0139] The confusion matrix provides the final identification results for different deviations from the distribution center of the microcrack and different average discrete size characteristic parameters under different nonlinear signal inputs, as shown in Table 1. Figure 24-31 As shown, the highest accuracy rates are all above 92%. This demonstrates the effectiveness of the present invention in decoupling and classifying microcrack clusters based on the number of deviations from their distribution centers and different average discrete size characteristics.

[0140] Table 1 shows the accuracy results of different nonlinear signals for identifying different features of microcrack clusters.

[0141]

[0142] A nonlinear ultrasonic multi-harmonic multi-feature decoupling identification system for microcrack clusters includes a defect time-domain nonlinear ultrasonic response signal acquisition module, a data preprocessing module, a deep belief network model construction module, and a decoupling identification module; wherein:

[0143] The specific operation process of the defect time-domain nonlinear ultrasonic response signal acquisition module is as follows: First, a microcrack group model is constructed, which includes microcrack groups with different areas, different number ranges and their deviations, different average sizes and different deviations; then, a 1MHz longitudinal wave is loaded onto the microcrack group model; finally, m transmitted wave defect time-domain nonlinear ultrasonic response signals are acquired at m receiving points after the longitudinal wave passes through the microcrack group region; M is a natural number greater than 0; This specific operation process can be implemented using the following specific technical means: The defect time-domain nonlinear ultrasonic response signal acquisition module can be implemented using the following specific technical means:

[0144] A microcrack swarm model with different areas, number ranges and their deviations, different average sizes and different deviations and dispersions was established using the finite element software ABAQUS. An ultrasonic wave propagation model within a damaged elastic body was also established. The ultrasonic excitation signal applied to the left side of the microcrack swarm model was a 10-period, 1MHz Hanning window modulated sine wave with an excitation amplitude of 1×10⁻⁶. -3 mm; the longitudinal wave propagates along the x-direction in the model. After the longitudinal wave passes through the microcrack cluster region, three different receiving points are set at 5 mm on the right boundary of the crack region to receive the transmitted wave signal; in order to ensure that the complete defect signal can be collected, the sampling frequency of the receiving points is set to 200 MHz.

[0145] The data preprocessing module performs the following steps: First, the time-domain nonlinear ultrasonic response signals of m defects are filtered to obtain the multi-harmonic nonlinear ultrasonic signals generated by the interaction between the ultrasonic signals and the microcracks. The multi-harmonic nonlinear ultrasonic signals include the zero-frequency ultrasonic signal, the frequency-division harmonic ultrasonic signal, the second harmonic ultrasonic signal, and the third harmonic ultrasonic signal of each defect time-domain nonlinear ultrasonic response signal.

[0146] Then, the ultrasonic signals from different receiving points are combined, and white noise is added to enhance the data to obtain preprocessed data;

[0147] Finally, the preprocessed data is divided into a training set and a test set; the data preprocessing module can be implemented using the following specific technical means:

[0148] The defect time-domain nonlinear ultrasonic response signal received at each receiving point is subjected to a 0.25MHz low-pass filter and band-pass filters of 0.25MHz-0.75MHz, 1.75-2.25MHz, and 2.75-3.25MHz, respectively, to obtain four nonlinear ultrasonic signals: zero-frequency signal, frequency-division harmonic signal, second harmonic signal, and third harmonic signal generated by the interaction of the ultrasonic longitudinal wave signal with the microcrack. Specifically, the filtering process for each defect time-domain nonlinear ultrasonic response signal includes:

[0149] The zero-frequency ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack is obtained by low-pass filtering the time-domain nonlinear ultrasonic response signal of the defect with a 0.2MHz filter.

[0150] The time-domain nonlinear ultrasonic response signal of the defect is bandpass filtered by 0.25MHz to 0.75MHz to obtain the frequency-division harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack;

[0151] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered at 1.75MHz to 2.25MHz to obtain the second harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0152] The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered by 2.75MHz to 3.25MHz to obtain the third harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

[0153] The process of combining signals from different receiving points is as follows:

[0154] Zero-frequency ultrasonic signals were extracted from the time-domain nonlinear ultrasonic response signals of each defect to form a zero-frequency dataset;

[0155] The frequency-division harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a frequency-division harmonic dataset.

[0156] The second harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a second harmonic dataset;

[0157] The third harmonic ultrasonic signal of the time-domain nonlinear ultrasonic response signal of each defect is extracted to form a third harmonic dataset.

[0158] After obtaining each nonlinear signal, the first step is to normalize the nonlinear signal to 0-1, mapping the data to the interval [0, 1].

[0159]

[0160] Where: X scaled The data is normalized, and X is the original data. min X is the minimum value in the original data. max This represents the maximum value in the original data.

[0161] By using signals from multiple receivers, microcrack defect information can be extracted more comprehensively. One-dimensional time-domain signals from n data points collected by m receivers with the same microcrack group parameters are stacked to form an m×n two-dimensional matrix. White noise with different signal-to-noise ratios is added to the sample signals for data augmentation to expand the sample size, avoid over-reliance on the original data format, and improve the model's generalization ability.

[0162] SNR=10*log10(Ps / Pn) (2);

[0163] Where: SNR is the signal-to-noise ratio, Ps is the power of the signal, and Pn is the power of the noise.

[0164] After the above data preprocessing, the one-dimensional defect signals received by each receiving point of the microcrack group with the same number and size parameters are processed to obtain m×n two-dimensional data with different nonlinear signals.

[0165] The construction process of the deep belief network model building module is as follows: First, an initial network model is constructed; then, the training set is used to train the initial network model, and the test set is used for the identification of various features of microcrack clusters to obtain the identification results of each harmonic nonlinear signal for different feature parameters; finally, based on the feature information of the microcrack clusters in the modeling, the initial network model is adjusted to obtain the deep belief network model; wherein, the construction process of this deep belief network model building module can be implemented using the following specific technical means:

[0166] A deep belief network (DBN) was constructed for classification and identification, comprising three stacked Restricted Boltzmann Machines (RBMs). The initial weight matrix of the network was set. The network model was trained using a pre-defined training set, and the test set was used to identify various features of microcrack clusters, obtaining the identification results of each harmonic nonlinear signal for different feature parameters. Finally, the sensitivity of different nonlinear signals to each microcrack cluster feature was investigated. Specifically, a single-signal DBN was constructed using different nonlinear signal inputs. The DBN consisted of three stacked RBMs. Each RBM consisted of a visible layer and a hidden layer: the first hidden layer contained 512 neurons, the second contained 256 neurons, and the third contained 128 neurons. ReLU was used as the activation function in all three hidden layers to introduce nonlinearity. Finally, a softmax layer was used for classification, the specific number of neurons determined according to the different microcrack cluster parameters. The optimizer was an ADMA optimizer, and the cross-entropy loss function was used. Furthermore, L2 regularization was used to prevent overfitting. For each single nonlinear signal, 9720 samples were obtained, with the training and test sets divided in a 4:1 ratio (7776 samples for training and 1944 samples for testing). The training set of each sample was fed into its respective network model for training, and finally, the test set was used for testing to obtain the identification results of the number and size of microcrack clusters for each nonlinear signal.

[0167] Decoupling identification module: The acquired time-domain nonlinear ultrasonic response signal of the defect is imported into a deep belief network to establish a multi-harmonic nonlinear ultrasonic decoupling identification model, and the multi-harmonic nonlinear ultrasonic identification is performed using the multi-harmonic nonlinear ultrasonic decoupling identification model.

[0168] A computer program for implementing the nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method for microcrack groups in the preferred embodiment described above.

[0169] An information data processing terminal for implementing the nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method for microcrack clusters in the above preferred embodiment.

[0170] A computer-readable storage medium includes instructions that, when executed on a computer, cause the computer to perform the nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method for microcrack clusters described in the preferred embodiment above.

[0171] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0172] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall fall within the scope of the technical solution of the present invention.

Claims

1. A nonlinear ultrasonic multi-harmonic multi-feature decoupling and identification method for microcrack clusters, characterized in that, include: S1. Acquire the time-domain nonlinear ultrasonic response signal of the defect with multi-parameter characteristic coupling of microcrack groups, specifically: First, a microcrack cluster model is constructed, which includes microcrack clusters with different areas, different number ranges and their deviations, different average sizes and different deviations. Then, a 1MHz longitudinal wave is loaded onto the microcrack cluster model. Finally, m transmission wave time-domain nonlinear ultrasonic response signals of the defects are collected at m receiving points after the longitudinal wave passes through the microcrack cluster region; M is a natural number greater than 0. S2. Data preprocessing, specifically: First, the time-domain nonlinear ultrasonic response signals of m defects are filtered to obtain the multi-harmonic nonlinear ultrasonic signals generated by the interaction of ultrasonic signals and microcracks. The multi-harmonic nonlinear ultrasonic signals include the zero-frequency ultrasonic signal, the frequency-division harmonic ultrasonic signal, the second harmonic ultrasonic signal, and the third harmonic ultrasonic signal of each defect time-domain nonlinear ultrasonic response signal. Then, the ultrasonic signals from different receiving points are combined, and white noise is added to enhance the data to obtain preprocessed data; Finally, the preprocessed data is divided into a training set and a test set; S3. Construct a deep belief network model, specifically as follows: First, an initial network model is constructed; then, the training set is used to train the initial network model, and the test set is used for the identification of various features of microcrack clusters to obtain the identification results of each harmonic nonlinear signal for different feature parameters; finally, based on the feature information of microcrack clusters in the modeling, the initial network model is adjusted to obtain a deep belief network model. S4. The acquired time-domain nonlinear ultrasonic response signal of the defect is imported into a deep belief network to establish a multi-harmonic nonlinear ultrasonic decoupling identification model. The multi-harmonic nonlinear ultrasonic decoupling identification model is used to identify multi-harmonic nonlinear ultrasound.

2. The nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method for microcrack clusters according to claim 1, characterized in that, S1 specifically includes: A microcrack swarm model with different areas, different number ranges and their deviations, different average sizes and different deviations was constructed using the finite element software ABAQUS. m receiving points are set at 5 mm from the right boundary of the crack area to receive the transmitted time-domain nonlinear ultrasonic response signal of the defect.

3. The nonlinear ultrasonic multi-harmonic multi-feature decoupling and identification method for microcrack clusters according to claim 1, characterized in that, When filtering m time-domain nonlinear ultrasonic response signals of defects, the filtering process for each defect's time-domain nonlinear ultrasonic response signal specifically includes: The zero-frequency ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack is obtained by low-pass filtering the time-domain nonlinear ultrasonic response signal of the defect with a 0.2MHz filter. The time-domain nonlinear ultrasonic response signal of the defect is bandpass filtered by 0.25MHz to 0.75MHz to obtain the frequency-division harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack; The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered at 1.75MHz to 2.25MHz to obtain the second harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack. The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered by 2.75MHz to 3.25MHz to obtain the third harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

4. The nonlinear ultrasonic multi-harmonic multi-feature decoupling and identification method for microcrack clusters according to claim 1, characterized in that, The process of combining signals from different receiving points is as follows: Zero-frequency ultrasonic signals were extracted from the time-domain nonlinear ultrasonic response signals of each defect to form a zero-frequency dataset; The frequency-division harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a frequency-division harmonic dataset; The second harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a second harmonic dataset; The third harmonic ultrasonic signal of the time-domain nonlinear ultrasonic response signal of each defect is extracted to form a third harmonic dataset.

5. The nonlinear ultrasonic multi-harmonic multi-feature decoupling identification method for microcrack clusters according to claim 1, characterized in that: M=3, the deep belief network model includes a visible layer, a hidden layer, and an output layer; the initial network model includes three stacked restricted Boltzmann machines; each restricted Boltzmann machine consists of a visible layer and a hidden layer; the first hidden layer contains 512 neurons, the second hidden layer contains 256 neurons, and the third hidden layer contains 128 neurons, all three hidden layers use ReLU as the activation function to introduce nonlinearity; the output layer uses softmax as the activation function and contains n neurons, the value of n is determined according to the microcrack cluster characteristics, specifically: n=3 when identifying the microcrack number range and its deviation, and n=5 when identifying the statistical average size of microcracks and its deviation.

6. A nonlinear ultrasonic multi-harmonic multi-feature decoupling identification system for microcrack clusters, characterized in that, include: Defect time-domain nonlinear ultrasonic response signal acquisition module: First, a microcrack group model is constructed, which includes microcrack groups with different numbers, different statistical average sizes, and different deviations; then, a 1MHz longitudinal wave is loaded onto the microcrack group model; finally, m transmitted wave defect time-domain nonlinear ultrasonic response signals are acquired at m receiving points after the longitudinal wave passes through the microcrack group region; M is a natural number greater than 0. The data preprocessing module first filters the m time-domain nonlinear ultrasonic response signals of defects to obtain multi-harmonic nonlinear ultrasonic signals generated by the interaction of ultrasonic signals and microcracks. These multi-harmonic nonlinear ultrasonic signals include the zero-frequency ultrasonic signal, frequency-division harmonic ultrasonic signal, second harmonic ultrasonic signal, and third harmonic ultrasonic signal of each defect's time-domain nonlinear ultrasonic response signal. Then, the ultrasonic signals from different receiving points are combined, and white noise is added to enhance the data and obtain preprocessed data. Finally, the preprocessed data is divided into a training set and a test set. The deep belief network model construction module first constructs an initial network model; then, the training set is used to train the initial network model, and the test set is used for the identification of various features of microcrack clusters to obtain the identification results of each harmonic nonlinear signal for different feature parameters; finally, the initial network model is adjusted according to the feature information of the microcrack clusters in the modeling to obtain the deep belief network model. Decoupling identification module: The acquired time-domain nonlinear ultrasonic response signal of the defect is imported into a deep belief network to establish a multi-harmonic nonlinear ultrasonic decoupling identification model, and the multi-harmonic nonlinear ultrasonic identification is performed using the multi-harmonic nonlinear ultrasonic decoupling identification model.

7. The nonlinear ultrasonic multi-harmonic multi-feature decoupling identification system for microcrack clusters according to claim 6, characterized in that, The acquisition process of the time-domain nonlinear ultrasonic response signal of the defect includes: A microcrack group model was constructed using the finite element software ABAQUS; m receiving points are set at 5 mm from the right boundary of the crack area to receive the transmitted time-domain nonlinear ultrasonic response signal of the defect.

8. The nonlinear ultrasonic multi-harmonic multi-feature decoupling identification system for microcrack clusters according to claim 6, characterized in that, When filtering m time-domain nonlinear ultrasonic response signals of defects, the filtering process for each defect's time-domain nonlinear ultrasonic response signal specifically includes: The zero-frequency ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack is obtained by low-pass filtering the time-domain nonlinear ultrasonic response signal of the defect with a 0.2MHz filter. The time-domain nonlinear ultrasonic response signal of the defect is bandpass filtered by 0.25MHz to 0.75MHz to obtain the frequency-division harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack; The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered at 1.75MHz to 2.25MHz to obtain the second harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack. The time-domain nonlinear ultrasonic response signal of the defect was bandpass filtered by 2.75MHz to 3.25MHz to obtain the third harmonic ultrasonic signal generated by the interaction between the ultrasonic signal and the microcrack.

9. The nonlinear ultrasonic multi-harmonic multi-feature decoupling identification system for microcrack clusters according to claim 6, characterized in that, The process of combining signals from different receiving points is as follows: Zero-frequency ultrasonic signals were extracted from the time-domain nonlinear ultrasonic response signals of each defect to form a zero-frequency dataset; The frequency-division harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a frequency-division harmonic dataset; The second harmonic ultrasonic signals of the time-domain nonlinear ultrasonic response signal of each defect are extracted to form a second harmonic dataset; The third harmonic ultrasonic signal of the time-domain nonlinear ultrasonic response signal of each defect is extracted to form a third harmonic dataset.

10. The nonlinear ultrasonic multi-harmonic multi-feature decoupling identification system for microcrack clusters according to claim 6, characterized in that: M=3, the deep belief network model includes a visible layer, a hidden layer, and an output layer; the initial network model includes three stacked restricted Boltzmann machines; each restricted Boltzmann machine consists of a visible layer and a hidden layer; the first hidden layer contains 512 neurons, the second hidden layer contains 256 neurons, and the third hidden layer contains 128 neurons, all three hidden layers use ReLU as the activation function to introduce nonlinearity; the output layer uses softmax as the activation function and contains n neurons, the value of n is determined according to the microcrack cluster characteristics, specifically: n=3 when identifying the microcrack number range and its deviation, and n=5 when identifying the statistical average size of microcracks and its deviation.

Images