Concrete damage real-time positioning method based on laser induction and dynamic wave velocity correction

Abstract

A concrete damage real-time positioning method based on laser induction and dynamic wave velocity correction comprises the following steps: constructing a damage induction platform integrated with a high-energy pulse laser, combining with a three-dimensional mobile platform to realize accurate positioning of a laser focus, and arranging a phase demodulation type optical fiber acoustic emission sensor array to construct a full-coverage monitoring network; synchronous acquisition of high-frequency acoustic emission signals is realized by adopting a multi-channel coupling phase demodulation system, and noise suppression is completed in combination with a self-adaptive wavelet threshold algorithm; a dynamic wave velocity attenuation model is established based on space-time reference information, a propagation path is optimized through a Basic Theta * path planning algorithm, a wave velocity field is reconstructed by applying an ART algebraic iteration algorithm, and a thermodynamic diagram is generated; and finally, constructing a convolution-graph neural joint network model, and realizing accurate monitoring and positioning analysis of the concrete structure damage through a damage probability distribution generation and multi-scale verification module by using a multi-modal data fusion and transfer learning technology.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive testing of concrete structures, and particularly to a method for real-time positioning of concrete damage based on laser induction and dynamic wave velocity correction. Background Art

[0002] During the long-term use of concrete structures, due to the influence of various natural factors (such as temperature change, humidity change, freeze-thaw cycle, etc.) and human factors (such as vehicle load, seismic action, etc.), internal damage may occur, such as cracks, holes, etc. If these damages are not discovered and treated in time, they may gradually expand, ultimately affecting the safety and durability of the concrete structure. Therefore, it is crucial to effectively detect and locate the damage of the concrete structure.

[0003] Currently, common concrete structure damage detection methods mainly include ultrasonic testing method, rebound method, ground penetrating radar method, etc. The ultrasonic testing method infers the internal defect situation of concrete by measuring parameters such as the propagation speed, wave amplitude, and frequency of ultrasonic waves in concrete. However, this method has high requirements for the technical level of operators, and different operators may obtain different test results, and the detection accuracy for complex structures and deep defects is limited. The rebound method estimates the strength of concrete based on the rebound value of the concrete surface, and indirectly judges the internal damage situation of the concrete. However, this method is easily affected by the surface state of the concrete (such as flatness, humidity, etc.), and can only reflect the situation within a certain depth range of the concrete surface, and it is difficult to accurately detect internal deep damage. The ground penetrating radar method uses the propagation characteristics of electromagnetic waves in concrete to detect internal defects, but its test results are greatly affected by the differences in the electromagnetic characteristics of concrete materials, and its ability to identify some subtle damages is insufficient.

[0004] In addition, most of the existing damage detection technologies mainly focus on the damage detection of the surface or shallow part of the concrete structure, and the detection effect of the internal damage of the concrete with cavities and special-shaped structures is not good. When facing complex concrete structures, it is difficult for traditional methods to achieve precise positioning and comprehensive evaluation of damage. At the same time, due to the complex and changeable environment where the concrete structure is located, the detection process is easily affected by various noise interferences, resulting in a reduction in the accuracy and reliability of the detection signal, affecting the accuracy of damage detection and positioning. Moreover, most of the current detection methods lack the ability to monitor the real-time dynamic process of damage development, and cannot timely grasp the change situation of damage, which is not conducive to the long-term health monitoring of the concrete structure. In summary, the existing technologies have disadvantages such as low detection accuracy, large influence of environmental interference, poor adaptability to complex structures, and lack of real-time dynamic monitoring ability in the damage detection and positioning of concrete structures, and there is an urgent need for a more efficient, accurate, and comprehensive concrete structure damage monitoring and positioning technology. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a real-time concrete damage localization method based on laser induction and dynamic wave velocity correction, and the specific steps are as follows, characterized in that:

[0006] Step 1: Construct a laser-induced damage test platform and a sensor network, build a three-dimensional solid model of concrete with cavities and special-shaped structures, configure a high-energy pulsed laser as a damage induction device, and achieve precise positioning of the laser focus on the model surface through a three-dimensional moving platform; deploy a phase-demodulated fiber laser acoustic emission sensor array on the model surface, and adopt a non-uniform array strategy to form a full-domain coverage acoustic emission monitoring network;

[0007] Step 2: Realize synchronous acquisition and preprocessing of high-frequency acoustic emission signals, synchronously capture acoustic emission signals through a multi-channel coupler phase demodulation system, and suppress environmental noise by combining an adaptive wavelet threshold denoising algorithm;

[0008] Step 3: Reconstruct the dynamic wave velocity field and optimize the propagation path. Based on the spatio-temporal reference information of laser-induced damage, establish a dynamic wave velocity attenuation model for the concrete cavity structure; use the basic theta* path planning algorithm to calculate the shortest time-delay path for sound waves to propagate around the cavity in real time, and synchronously combine the ART algebraic iterative algorithm to reconstruct the wave velocity field in the damage area to achieve dynamic correction of wave velocity values;

[0009] Step 4: Multi-modal deep learning damage localization analysis, construct a convolutional-graph neural network joint model, and the input layer synchronously receives the acoustic emission time-domain waveform and the corrected wave velocity value; strengthen damage feature extraction through an attention mechanism to solve the signal distortion problem caused by the anisotropy of composite materials, and the output layer generates the probability distribution of damage coordinates.

[0010] The real-time concrete damage localization method based on laser induction and dynamic wave velocity correction of the present invention has the following beneficial effects. The technical effects of the present invention are as follows:

[0011] 1. By building a three-dimensional solid model of concrete with cavities and special-shaped structures, configuring a high-energy pulsed laser as a damage induction device, and combining a three-dimensional moving platform to achieve precise positioning of the laser focus on the model surface, the present invention can accurately induce damage at a predetermined position, providing a reliable damage source for subsequent monitoring.

[0012] 2. The present invention establishes a dynamic wave velocity attenuation model for concrete cavity structures based on the spatio-temporal reference information of laser-induced damage, which can accurately describe the influence of laser-induced damage on the acoustic wave propagation velocity in concrete. By using the basic theta* path planning algorithm to calculate the shortest time-delay path for acoustic waves to propagate around the cavity in real time, and combining with the ART algebraic iteration algorithm to reconstruct the wave velocity field in the damaged area and realize the dynamic correction of wave velocity values, it can not only clearly present the changes in the wave velocity field in the damaged area, but also provide more accurate wave velocity information for damage location.

[0013] 3. The present invention constructs a convolutional-graph neural network joint model. The input layer synchronously receives the acoustic emission time-domain waveform and the corrected wave velocity value, and strengthens the extraction of damage features through the attention mechanism, effectively solving the problem of signal distortion caused by the anisotropy of composite materials. The output layer generates the probability distribution of damage coordinates and is optimized through the transfer learning strategy, improving the generalization ability of the model and the accuracy of damage location. The multi-scale damage verification module can accurately determine effective damage through cross-verification with the dynamically corrected wave velocity field, further improving the reliability of damage location. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flowchart of the present invention;

[0015] Figure 2 is a schematic diagram of multi-modal deep learning damage location analysis of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] The present invention will be further described in detail below in conjunction with the drawings and the specific embodiments:

[0017] The present invention proposes a real-time concrete damage location method based on laser induction and dynamic wave velocity correction, constructs a laser-induced damage test platform and a sensor network, realizes the acquisition and preprocessing of high-frequency acoustic emission signals, conducts the reconstruction of the dynamic wave velocity field and the optimization of the propagation path, and uses multi-modal deep learning to realize damage location analysis, which can accurately and real-time monitor and locate concrete damage. The flowchart of the invention is as Figure 1 shown, and the steps of the present invention will be introduced in detail below.

[0018] Step 1: Construct a laser-induced damage test platform and a sensor network

[0019] Build a three-dimensional solid model of concrete containing cavities and special-shaped structures, configure a high-energy pulsed laser as a damage induction device, and realize the precise positioning of the laser focus on the model surface through a three-dimensional moving platform; deploy a phase demodulation type fiber optic laser acoustic emission sensor array on the model surface, and adopt a non-uniform array strategy to form an acoustic emission monitoring network covering the whole area.

[0020] Step 1.1, configure a high-energy pulsed laser as a damage induction device:

[0021] Use a high-energy pulsed laser with a wavelength of λ = 1064 nm and a single-pulse energy E ≥ 200 mJ. Bind the high-energy pulsed laser to a three-dimensional moving platform, precisely control the three-dimensional moving platform, accurately position the laser focus at a specified position on the model surface, and induce damage at the predetermined position.

[0022] Step 1.2, Deploy a phase-demodulated fiber optic laser acoustic emission sensor array:

[0023] Adopt a phase-demodulated fiber optic laser acoustic emission sensor array, which has 8 channels. Use a non-uniform arraying strategy to densely arrange sensor nodes at a spacing of 10 cm in the structurally weak areas. Arrange sensor nodes at a spacing of 50 cm in other areas. Form an acoustic emission monitoring network that covers the entire area. This network can real-time monitor the acoustic emission signals generated during the laser-induced damage process of the concrete model and transmit the signals to the subsequent signal processing system for analysis.

[0024] Step 2: Achieve synchronous acquisition and preprocessing of high-frequency acoustic emission signals. Synchronously capture the acoustic emission signals through a multi-channel coupler phase demodulation system, and combine the adaptive wavelet threshold denoising algorithm to suppress environmental noise.

[0025] Step 2.1, Synchronous capture of high-frequency acoustic emission signals:

[0026] Use a multi-channel 3×3 coupler phase demodulation system to collect acoustic emission signals. The sampling rate of this system is 10 million times per second, and the system synchronously captures acoustic emission signals in the 500 kHz frequency band. The signals in the 500 kHz frequency band contain information related to damage. Analyze the acoustic emission signals in the 500 kHz frequency band through the following steps.

[0027] Step 2.2, Environmental noise suppression:

[0028] Adopt an adaptive wavelet threshold denoising algorithm with Daubechies9 wavelet basis to suppress environmental noise. Perform wavelet transform on the signal, decompose the signal into different frequency sub-bands, and then process the wavelet coefficients according to the different characteristics of noise and signal in the wavelet domain. Let the acquired noisy emission signal be x(t), and this signal is the superposition of the real signal s(t) and the noise n(t), that is:

[0029] x(t) = s(t) + n(t).

[0030] Perform wavelet transform on x(t) to obtain the wavelet coefficients W x (j,k), where j represents the scale and k represents the translation.

[0031] According to the adaptive wavelet threshold algorithm, calculate the threshold λ at each scale jj Perform threshold processing on the wavelet coefficients to obtain the denoised wavelet coefficients

[0032]

[0033] Finally, for the denoised wavelet coefficients perform inverse wavelet transform to obtain the denoised acoustic emission signal

[0034] Step 3: Reconstruct the dynamic wave speed field and optimize the propagation path

[0035] Based on the spatio-temporal reference information of laser-induced damage, establish a dynamic wave speed attenuation model for the concrete cavity structure; use the basic theta* path planning algorithm to calculate the shortest time-delay path for sound waves to propagate around the cavity in real time, and synchronously combine the ART algebraic iteration algorithm to reconstruct the wave speed field in the damaged area to achieve dynamic correction of the wave speed value.

[0036] Step 3.1, establish the dynamic wave speed attenuation model:

[0037] Based on the spatio-temporal reference information of laser-induced damage, establish a dynamic wave speed attenuation model for the concrete cavity structure. Laser-induced damage will change the material properties of the concrete structure, thereby affecting the propagation speed of sound waves in it. The spatio-temporal reference information contains the key information of the time and location of laser-induced damage.

[0038] Let the initial wave speed of sound waves in the concrete in the undamaged state be v0. After laser-induced damage, due to the existence of damage, the wave speed will decay. Let the wave speed decay and the damage degree be D, 0 ≤ D ≤ 1, D = 0 indicates no damage, and D = 1 indicates complete damage. Establish the dynamic wave speed attenuation model:

[0039] v = v0·(1 - α·D)

[0040] where v is the wave speed of sound waves in the concrete after laser-induced damage, and α is the attenuation coefficient related to the material properties and damage type. Its value is determined by monitoring and analyzing the laser-induced damage process and combining the physical properties of the material.

[0041] Step 3.2, calculate the shortest time-delay path for sound waves to propagate around the cavity:

[0042] Use the basic theta* path planning algorithm to calculate the shortest time-delay path for sound waves to propagate around the cavity in real time, and the calculation frequency is 100 Hz. First, discretize the space of the concrete cavity structure to construct a graph structure composed of nodes and edges. Let the distance between node i and node j be d ij , and according to the dynamic wave speed attenuation model, calculate the wave speed v on this path ij .

[0043] The time required for the sound wave to propagate from node i to node j is:

[0044]

[0045] The goal of the basic theta* path planning algorithm is to find a path path with the shortest total propagation time T from the starting node s to the target node t in this graph structure, where the total propagation time T is expressed as:

[0046]

[0047] By continuously searching and evaluating the total propagation time of different paths, the shortest time-delay path for the sound wave to propagate around the cavity is determined.

[0048] Step 3.3, reconstruct the wave speed field of the damaged area:

[0049] Reconstruct the wave speed field of the damaged area through the ART algebraic iterative algorithm. Let x be the wave speed field vector to be reconstructed, and y be the acoustic emission signal data vector measured by the sensor. The measurement matrix A represents the relationship between the wave speed field and the acoustic emission signal, and we can get:

[0050] y = Ax + ε

[0051] where ε is the measurement noise.

[0052] The iterative process of the ART algorithm is as follows:

[0053] At the k-th iteration, for the i-th measurement equation:

[0054]

[0055] where, a ij represents the contribution degree of the j-th dimension of the wave speed field vector to the i-th component of the acoustic emission signal, x j is the j-th component of the wave speed field vector x, ε i is the i-th element in the noise vector ε, m is the dimension of the wave speed field vector, y i is the i-th component in the measured acoustic emission signal data vector y, and update the estimated value x (k +1) of the wave speed field vector x at the (k + 1)-th iteration:

[0056]

[0057] where, represents the estimated value of the j-th component in the wave speed field vector x at the (k + 1)-th iteration, represents the estimated value of the j-th component in the wave speed field vector x at the k-th iteration, aii is the element on the main diagonal of the measurement matrix A. By iteration, the estimated wave velocity field gradually approaches the true wave velocity field, thus realizing the reconstruction of the wave velocity field in the damaged area.

[0058] Step 3.4: Implement dynamic correction of wave velocity values:

[0059] Through the above reconstructed wave velocity field, dynamic correction of wave velocity values is realized. According to the reconstructed wave velocity field, the wave velocity values at each position are updated in real time, and the corrected wave velocity values reflecting the laser-induced damage of the concrete structure are obtained.

[0060] Step 3.5: Generate thermogram data of the wave velocity field

[0061] Map the dynamically corrected wave velocity values to a three-dimensional grid according to spatial coordinates, assign gray levels according to the magnitude of the wave velocity values, and generate thermogram data M(i, j, z) of the wave velocity field, where (i, j, z) represents the coordinates of the three-dimensional plane grid after discretization of the concrete structure surface. The specific method is as follows:

[0062] Discretize the concrete structure surface into a grid with a resolution of 1 mm 3 ; According to the corrected wave velocity values in Step 3.4, fill the wave velocity values of each pixel point in the grid through an interpolation algorithm; normalize the wave velocity values to the range of 0 and 255 to generate gray thermogram data M(i, j, z).

[0063] Step 4: Multi-modal deep learning damage location analysis

[0064] Construct a convolutional-graph neural network joint model. The input layer synchronously receives the acoustic emission time-domain waveform and the corrected wave velocity values; through the attention mechanism, the extraction of damage features is strengthened to solve the problem of signal distortion caused by the anisotropy of composite materials. The output layer generates the probability distribution of damage coordinates. The schematic diagram of multi-modal deep learning damage location analysis is as Figure 2 shown.

[0065] Step 4.1: Construct the multi-modal data fusion input layer:

[0066] The input layer synchronously receives the acoustic emission time-domain waveform and the thermogram data modality of the wave velocity field. The acoustic emission time-domain waveform is the signal denoised in Step 2.2 with a sampling rate of 10 MHz, a time window length of 1 ms, and a data dimension of 1×10 4 . The thermogram data of the wave velocity field is the gray thermogram M(i, j, z) from Step 3.5, with a resolution of 1 mm 3 , normalized to 0 - 255, and a data dimension of 1000×1000×1000.

[0067] The acoustic emission waveform passes through a 1D convolutional layer with a convolution kernel size of 64 and a step size of 8, and is reduced to 1×1250, and then mapped to a feature vector f through a fully connected layer. ae .

[0068] The wave velocity field heat map is passed through a 3D convolutional network with a convolution kernel size of 5×5×5 and a number of channels of 32, 64, and 128, respectively, to extract spatial features and output the feature vector f vf .

[0069] Bimodal feature f ae and f vf Fusion via Cross-Attention Mechanism:

[0070]

[0071] Where Q = f ae , K=V=f vf , d k is the key vector dimension adjustment parameter, softmax is the softmax function, and the output fusion feature f fusion .

[0072] Step 4.2, damage probability distribution generation and transfer learning optimization:

[0073] Step 4.2.1, the output layer generates a probability distribution map of damage coordinates

[0074] Fusion feature f fusion Mapped to a 1024×1024 resolution grid through a fully connected layer and upsampled to 1mm through bilinear interpolation 2 / pixel. Use softmax with temperature parameter to generate probability distribution P(i,j,z):

[0075]

[0076] Among them, τ is 0.1, exp is an exponential function with a constant e as the base, z(i,j,z) and z(i',j',z') are the fusion features f fusion After being mapped by the fully connected layer, the values ​​at positions (i, j, z) and (i', j', z') are the values ​​covering all possible pixel coordinate points in the entire area to be analyzed, and the damage coordinate probability distribution map P(i, j, z) is obtained.

[0077] Step 4.2.2, transfer learning strategy:

[0078] In the pre-training phase, training is done on the data set, and the loss function is KL divergence:

[0079] L pre =∑D KL (Pgt ||P pred )

[0080] Among them, L pre is the loss function in the pre-training stage, P gt is the true damage coordinate probability distribution, P pred is the predicted damage coordinate probability distribution, D KL is the KL divergence.

[0081] In the fine-tuning stage, freeze the underlying parameters of the convolutional-graph neural network joint model, and only optimize the attention layer and the output layer. Add L2 regularization to the loss function:

[0082] L task = L pre + λ||θ|| 2

[0083] Among them, λ is 0.001.

[0084] Step 4.3, multi-scale damage verification module:

[0085] The output probability distribution map is post-processed morphologically to remove noise points with an area < 1mm 2 . Cross-verify with the dynamic correction wave velocity field in Step 3.4. If the overlap degree between the wave velocity abnormal area and the probability peak area > 90%, it is determined as an effective damage.

[0086] The above is only a preferred embodiment of the present invention, and it is not a limitation to the present invention in any other form. Any modification or equivalent change made according to the technical essence of the present invention still belongs to the scope protected by the present invention.

Claims

1. A real-time concrete damage location method based on laser-induced and dynamic wave velocity correction, the specific steps are as follows, and it is characterized in that: Step 1: Construct a laser-induced damage test platform and a sensor network, build a three-dimensional solid concrete model with cavities and special-shaped structures, configure a high-energy pulsed laser as a damage induction device, and achieve precise positioning of the laser focus on the model surface through a three-dimensional moving platform; deploy a phase-demodulated fiber laser acoustic emission sensor array on the model surface, adopt a non-uniform array strategy to form an acoustic emission monitoring network covering the entire area; Step 2: Realize synchronous acquisition and preprocessing of high-frequency acoustic emission signals, synchronously capture acoustic emission signals through a multi-channel coupler phase demodulation system, and suppress environmental noise by combining an adaptive wavelet threshold denoising algorithm; Step 3: Reconstruct the dynamic wave velocity field and optimize the propagation path. Based on the spatio-temporal reference information of laser-induced damage, establish a dynamic wave velocity attenuation model for the concrete cavity structure; use the basic theta* path planning algorithm to calculate the shortest time-delay path of sound waves propagating around the cavity in real time, and synchronously combine the ART algebraic iterative algorithm to reconstruct the wave velocity field in the damage area to achieve dynamic correction of wave velocity values; Step 4: Multi-modal deep learning damage location analysis, construct a convolutional-graph neural network joint model, and the input layer synchronously receives the acoustic emission time-domain waveform and the corrected wave velocity value; strengthen damage feature extraction through the attention mechanism to solve the signal distortion problem caused by the anisotropy of composite materials, and the output layer generates the probability distribution of damage coordinates.

2. The real-time concrete damage location method based on laser induction and dynamic wave velocity correction according to claim 1, wherein: The construction of the laser-induced damage test platform and the sensor network in Step 1 can be expressed as: Step 1.1, Configure a high-energy pulsed laser as a damage induction device: Adopt a high-energy pulsed laser with a wavelength of 1064 nm and a single-pulse energy greater than 200 mJ; bind the high-energy pulsed laser to the three-dimensional moving platform, precisely control the three-dimensional moving platform, accurately position the laser focus to the specified position on the model surface, and induce damage at the predetermined position; Step 1.2, Deploy a phase-demodulated fiber laser acoustic emission sensor array: Adopt a phase-demodulated fiber laser acoustic emission sensor array with 8 channels. Adopt a non-uniform array strategy, densely arrange sensor nodes at a spacing of 10 cm in the structurally weak areas; arrange sensor nodes at a spacing of 50 cm in other areas; form an acoustic emission monitoring network covering the entire area, and continuously monitor the acoustic emission signals generated during the laser-induced damage process of the concrete model, and transmit the signals to the subsequent signal processing system for analysis.

3. The real-time positioning method for concrete damage based on laser induction and dynamic wave velocity correction according to claim 1, wherein: The realization of synchronous acquisition and preprocessing of high-frequency acoustic emission signals in Step 2 can be expressed as follows: Step 2.1, Synchronous capture of high-frequency acoustic emission signals: Use a multi-channel 3×3 coupler phase demodulation system to collect acoustic emission signals; the sampling rate of this system is 10 million times per second, and the system synchronously captures acoustic emission signals in the 500 kHz frequency band; the signals in the 500 kHz frequency band contain information related to damage, and the acoustic emission signals in the 500 kHz frequency band are analyzed through the following steps; Step 2.2, Environmental noise suppression: An adaptive wavelet threshold denoising algorithm using the Daubechies9 wavelet basis is employed to suppress environmental noise. The signal is subjected to wavelet transform, decomposed into different frequency sub-bands, and then the wavelet coefficients are processed according to the different characteristics of noise and signal in the wavelet domain. Let the collected noisy emission signal be x(t), which is the superposition of the true signal s(t) and noise n(t), i.e.: x(t) = s(t) + n(t); Perform a wavelet transform on \(x(t)\) to obtain the wavelet coefficient \(W\) x (j,k), where \(j\) represents the scale and \(k\) represents the translation; According to the adaptive wavelet threshold algorithm, calculate the threshold λ at each scale j j ; perform threshold processing on the wavelet coefficients to obtain the denoised wavelet coefficients Finally, perform the inverse wavelet transform on the denoised wavelet coefficients to obtain the denoised acoustic emission signal 4. The real-time positioning method for concrete damage based on laser induction and dynamic wave velocity correction according to claim 1, characterized in that: The dynamic wave velocity field reconstruction and propagation path optimization in step 3 can be expressed as follows: Step 3.1, establish a dynamic wave velocity attenuation model: Based on the spatio-temporal reference information of laser-induced damage, a dynamic wave velocity attenuation model of the concrete cavity structure is established; Laser-induced damage will change the material properties of the concrete structure, thereby affecting the propagation speed of sound waves in it; The spatio-temporal reference information contains the key information of the time and location where laser-induced damage occurs. Let the initial wave velocity of sound waves in the concrete in the undamaged state be v0. After laser-induced damage, due to the existence of damage, the wave velocity will decay; Let the wave velocity decay and the damage degree be D, 0 ≤ D ≤ 1, D = 0 represents no damage, D = 1 represents complete damage, and establish a dynamic wave velocity attenuation model: v = v0·(1 - α·D) where v is the wave velocity of sound waves in the concrete after laser-induced damage, and α is the attenuation coefficient related to material properties and damage types. Its value is determined by monitoring and analyzing the laser-induced damage process and combining the physical properties of the material. Step 3.2, calculate the shortest time-delay path for sound waves to propagate around the cavity: The basic theta* path planning algorithm is used to calculate the shortest time-delay path for sound waves propagating around the cavity in real time, and the calculation frequency is 100 Hz. First, the space of the concrete cavity structure is discretized to construct a graph structure composed of nodes and edges. Let the distance between node i and node j be d ij , and according to the dynamic wave speed attenuation model, the wave speed v on this path is calculated ij ; Then the time required for sound waves to propagate from node i to node j is: The goal of the basic theta* path planning algorithm is to find a path path with the shortest total propagation time T from the starting node s to the target node t in this graph structure, where the total propagation time T is expressed as: By continuously searching and evaluating the total propagation time of different paths, the shortest time-delay path for sound waves to propagate around the cavity is determined. Step 3.3, reconstruct the wave velocity field in the damaged area: Reconstruct the wave velocity field in the damaged area through the ART algebraic iterative algorithm; Let x be the wave velocity field vector to be reconstructed, and y be the acoustic emission signal data vector measured by the sensor; The measurement matrix A represents the relationship between the wave velocity field and the acoustic emission signal, and we can get: y = Ax + ε where ε is the measurement noise; The iterative process of the ART algorithm is as follows: At the k-th iteration, for the i-th measurement equation: Among them, a ij represents the contribution degree of the j-th dimension of the wave velocity field vector to the i-th component of the acoustic emission signal, x j is the j-th component of the wave velocity field vector x, ε i is the i-th element in the noise vector ε, m is the dimension of the wave velocity field vector, y i is the i-th component in the measured acoustic emission signal data vector y, and update the estimated value x of the wave velocity field vector x at the (k + 1)-th iteration (k+1) : Among them, represents the estimated value of the j-th component in the wave speed field vector x at the (k + 1)-th iteration, represents the estimated value of the j-th component in the wave speed field vector x at the k-th iteration, a ii is the element on the main diagonal of the measurement matrix A. By iteration, the estimated wave speed field gradually approaches the true wave speed field, thereby realizing the reconstruction of the wave speed field in the damage area; Step 3.4, realize the dynamic correction of wave velocity values: Through the reconstructed wave velocity field above, realize the dynamic correction of wave velocity values. According to the reconstructed wave velocity field, the wave velocity values at each position are updated in real time to obtain the corrected wave velocity values that can reflect the laser-induced damage of the concrete structure. Step 3.5: Generate wave velocity field heat map data Map the dynamically corrected wave velocity values to a three-dimensional grid according to the spatial coordinates, and assign gray levels according to the wave velocity value sizes to generate the wave velocity field heat map data M(i,j,z), where (i,j,z) represents the coordinates of the three-dimensional plane grid after discretization of the concrete structure surface; The specific method is as follows: Discretize the surface of the concrete structure into a grid with a resolution of 1 mm 3 ; According to the corrected wave velocity value in step 3.4, fill the wave velocity value of each pixel point in the grid through an interpolation algorithm; normalize the wave velocity value to the range of 0 and 255 to generate grayscale thermogram data M(i, j, z).

5. The real-time concrete damage location method based on laser induction and dynamic wave velocity correction according to claim 1, wherein: The multi-modal deep learning damage location analysis in step 4 can be expressed as follows: Step 4.1, construct the multi-modal data fusion input layer: The input layer synchronously receives the modal data of the acoustic emission time-domain waveform and the wave velocity field thermal map data. The acoustic emission time-domain waveform is the signal after denoising in step 2.2 The sampling rate is 10 MHz, the time window length is 1 ms, and the data dimension is 1×10 4 ; The wave velocity field thermal map data is the grayscale thermal map M(i,j,z) from step 3.5, with a resolution of 1 mm 3 , normalized to 0-255, and the data dimension is 1000×1000×1000; The acoustic emission waveform passes through a 1D convolutional layer with a convolutional kernel size of 64 and a stride of 8, reducing the dimension to 1×1250, and then is mapped to the feature vector f through a fully connected layer ae ; The wave velocity field heat map uses a 3D convolutional network with a convolutional kernel size of 5×5×5 and channel numbers of 32, 64, and 128 to extract spatial features and output the feature vector f vf ; Bimodal feature f ae and f vf are fused through the cross-attention mechanism: where Q = f ae , K = V = f vf , d k is the key vector dimension adjustment parameter, softmax is the softmax function, and the output fusion feature is f fusion ; Step 4.2, generate the damage probability distribution and optimize it with transfer learning: Step 4.2.1, the output layer generates the damage coordinate probability distribution map Fused feature f fusion It is mapped to a 1024×1024 resolution grid through a fully connected layer and upsampled to 1 mm / pixel by bilinear interpolation; 2 The softmax with temperature parameter is used to generate the probability distribution P(i, j, z): where τ is 0.1, exp is the exponential function with the constant e as the base, z(i, j, z) and z(i', j', z') are the values at positions (i, j, z) and (i', j', z') after being mapped by the fully connected layer, and (i', j', z') covers all possible pixel coordinate points within the entire area to be analyzed, and the damage coordinate probability distribution map P(i, j, z) is obtained; fusion The values at positions (i, j, z) and (i', j', z') after being mapped by the fully connected layer, where (i', j', z') covers all possible pixel coordinate points within the entire area to be analyzed, and the damage coordinate probability distribution map P(i, j, z) is obtained; Step 4.2.2, transfer learning strategy: In the pre-training stage, train on the dataset, and the loss function is the KL divergence: L pre = ∑D KL (P gt ||P pred ) Among them, L pre is the loss function in the pre-training stage, P gt is the probability distribution of the true damage coordinates, P pred is the probability distribution of the predicted damage coordinates, D KL is the KL divergence; In the fine-tuning stage, freeze the underlying parameters of the convolutional-graph neural network joint model, only optimize the attention layer and the output layer, and add L2 regularization to the loss function: L task = L pre + λ||θ|| 2 where λ is 0.001; Step 4.3, multi-scale damage verification module: The output probability distribution map is post-processed morphologically to remove noise points with an area <1 mm 2 . It is cross-validated with the dynamic correction wave velocity field in step 3.

4. If the overlap degree between the wave velocity anomaly area and the probability peak area is >90%, it is determined as an effective damage.

Images