Damage localization method based on improved A* algorithm and BP neural network

Abstract

The invention discloses a damage positioning method based on an improved A * algorithm and a BP neural network. The method comprises the steps of definition and calculation of a direction-dependent wave velocity parameter OD-WV, search path generation of the improved anisotropic A * algorithm, pickup of an acoustic emission signal arrival time through a combined automatic pickup algorithm and a BP neural network training model based on an optimized data set. Compared with the prior art, the physical parameters are utilized to ensure scientificity of the data set, the model efficiency is improved through the optimized A * algorithm, and an efficient and practical solution is provided for health monitoring of the aerospace composite material structure.

Description

Technical Field

[0001] The present invention relates to the field of structural health monitoring and damage detection, in particular to a method based on anisotropic A * The composite material damage localization method based on the algorithm and BP neural network can be widely used in efficient health monitoring of aerospace composite structures. Background Art

[0002] Carbon fiber reinforced composites (CFRCs) are widely used in aircraft due to their light weight, high specific stiffness, good designability and excellent fatigue resistance. Online structural health monitoring of aircraft includes four steps: detection, location, classification and quantification. The detection mainly studies the accuracy and efficiency of impact damage. The progress in damage location assessment of anisotropic CFRCs structures provides opportunities for the development of structural health monitoring.

[0003] To address this problem, the booming multi-sensor acoustic emission technology provides a method for damage localization, which then guides pilots to make maintenance decisions and implement preventive maintenance. Kundu initially proposed a method to achieve damage localization in composite plates by minimizing a nonlinear error function. The localization efficiency was further improved by developing an improved error function. However, these methods are sensitive to the starting time and require a lot of time to calculate complex expressions. "Delta-T mapping", which involves generating contour maps of a constant Δt for each sensor pair and linearly interpolating between grid points, can identify intersections by overlaying their corresponding contours. This task takes a long time to repeat the lead-breaking experiment and does not take into account changes in meshing. In addition, although beamforming and sensor cluster methods can predict damage localization for weakly anisotropic structures without prior knowledge of wave speeds in different directions. However, the monitoring data limitations of sparse sensor clusters make it incompatible with the requirements of macroscopic aircraft components. Therefore, it is a challenge to develop an accurate algorithm for anisotropic CFRCs structures with direction-dependent physical information.

[0004] The emergence of machine learning, with its superior fault tolerance and fast processing characteristics, can quickly separate changes in sensitive features, bringing hope for identifying damage locations. However, the limitations of physical information-based and neural network-based methods in damage location include: (1) complex nonlinear problems that are difficult to model accurately, especially in anisotropic materials; (2) heavy reliance on large and high-quality data sets, inconvenient measurement conditions, unsuitable equations, and complex solution processes, making the finite element method (FEM) and manual repeated experiments inefficient in calculating the physical parameters of composite structures. Summary of the invention

[0005] The purpose of this invention is to provide a damage location method based on an improved A* algorithm and a BP neural network, which not only uses physical parameters to ensure the scientific nature of the data set, but also optimizes the A* algorithm to ensure the scientific nature of the data set. * The algorithm improves model efficiency and provides an efficient and practical solution for health monitoring of aerospace composite structures.

[0006] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0007] A damage location method based on improved A* algorithm and BP neural network, comprising:

[0008] Step 1, select a sample structure of the same material as the product to be tested, determine the propagation characteristics of the acoustic emission signal along each direction on the sample structure through an experimental method, and obtain the direction-related wave velocity parameter OD-WV and the minimum wave velocity; wherein the wave velocity parameter OD-WV refers to the propagation velocity of the acoustic emission signal Lamb wave in each polar angle direction in the monitoring area set on the sample structure;

[0009] Step 2, using the minimum wave speed to normalize the direction-related wave speed parameter OV-WV;

[0010] Step 3, gridding the monitoring area on the sample structure, setting the grids for arranging acoustic emission sensors in the detection area, and using the A* algorithm with the wave velocity parameter OV-WV introduced to calculate the shortest path between each grid to each grid where the acoustic emission sensor is located, so as to determine the arrival time of the acoustic emission signal obtained by each acoustic emission sensor when each grid is used as an acoustic emission source; taking the time difference after the arrival time determined by each acoustic emission sensor is subtracted from each other as a group of samples, and the label of the sample is the coordinate of the corresponding grid, so as to construct a training data set;

[0011] Step 4, constructing a BP neural network, using the samples in the training data set to train the BP neural network, and saving the trained BP neural network model;

[0012] Step 5, for the product to be tested, in actual application, the monitoring area and the acoustic emission sensor are first arranged on the product to be tested in the same way as on the sample structure, and the acoustic emission sensor is used to monitor the acoustic emission signal; when the monitoring area generates an acoustic emission signal due to structural damage or other reasons, the acoustic emission signal is captured by the acoustic emission sensor, and the arrival time of the acoustic emission signal to each acoustic emission signal sensor is determined by a joint automatic picking algorithm combining an adaptive threshold and an Akaike information criterion AIC, thereby determining the time difference of the arrival time and inputting it into the trained BP neural network model, and the result output by the BP neural network model is the coordinate of the location where the acoustic emission signal is generated.

[0013] Furthermore, the test method is as follows:

[0014] The monitoring area is set on the sample structure, and a polar coordinate system is established with the center of the monitoring area as the pole. The polar axis is defined as a ray from the pole to any direction of the edge of the sample structure. The polar angle is defined as the angle between any straight line passing through the pole and the polar axis. The polar axis corresponds to a polar angle of 0°.

[0015] The polar axis in the monitoring area is rotated counterclockwise at fixed intervals in the counterclockwise direction, so that the monitoring area is divided into multiple polar angle areas of the same size; an acoustic emission device is arranged at the pole, and then the wave velocity parameter OD-WV in each polar angle direction of the monitoring area is determined respectively, as follows:

[0016] The polar axis corresponds to the polar angle direction of 0°. Two acoustic emission sensors are arranged at intervals on the polar axis. Lamb wave signals are generated by the acoustic emission device. The acoustic emission signals are captured by the two acoustic emission sensors on the polar axis. The wave velocity parameter OD-WV=S / △t in the polar angle direction of 0° can be obtained by using the spacing S between the two acoustic emission sensors and the time difference △t between the Lamb wave signals reaching the two sensors.

[0017] At each position of polar axis rotation, acoustic emission sensors are arranged at intervals according to the same method, and the OD-WV in the corresponding polar angle direction is determined, thereby obtaining the OD-WV in each polar angle direction after division;

[0018] The OD-WV in different polar angle directions is fitted by using a cubic spline curve to obtain an OD-WV fitting curve; thus, the OD-WV in any polar angle direction can be determined by the OD-WV fitting curve.

[0019] Furthermore, the minimum wave velocity is used to normalize the direction-related wave velocity parameter OV-WV, which is expressed as:

[0020] α(θ)=v(θ) / v min

[0021] v(θ) represents the OD-WV in the direction of the polar angle θ, α(θ) represents the normalized OD-WV, and v min Indicates the minimum wave speed.

[0022] Furthermore, the A* algorithm introducing the wave velocity parameter OV-WV is as follows:

[0023] The acoustic emission sensors are arranged on the grids at the edge of the detection area; assuming that the acoustic emission signal is generated in each grid, the cost function f(n) in the path search process from a grid a to a grid c where the acoustic emission sensor is located based on the A* algorithm is expressed as:

[0024] f(n)=g(n)+wh(n)

[0025] Among them, w is the weight coefficient, g(n) represents the actual cost from grid a to the current search grid b, and h(n) represents the estimated cost of the shortest path from grid a to grid c where the acoustic emission sensor is located, which is expressed as follows:

[0026]

[0027] Among them, (x a ,y a ) represents the coordinates of grid a, (x b ,y b ) represents the coordinates of the current search grid b, (x c ,y c ) represents the coordinates of the grid c where the acoustic emission sensor is located, α(θ) represents the normalized OD-WV corresponding to the angle θ between the line connecting the grid a and the current search grid b and the polar axis;

[0028] After determining the shortest path d from grid a to grid c where the acoustic emission sensor is located, compare d with the minimum wave velocity v min The ratio of is taken as the arrival time of the acoustic emission signal of the acoustic emission sensor at grid c; in the same way, the arrival time from grid a to the grids where the other sensors are located is taken as a group of samples after the arrival time is subtracted from each other, and the corresponding label is the coordinate (x a ,y a ), so as to construct a training data set using the samples corresponding to each grid in the monitoring area.

[0029] Furthermore, the BP neural network includes an input layer, a hidden layer and an output layer, wherein the number of neurons in the input layer is consistent with the number of time differences in the sample, the network uses the hyperbolic tangent function Tanh as an activation function, and the samples of the input layer are processed by the hidden layer and then passed to the output layer by the hyperbolic tangent function Tanh;

[0030] When training the network, the mean square error loss function is used, and the gradient descent algorithm is used to adjust the weights and biases in each iteration.

[0031] Furthermore, the joint automatic picking algorithm determines the arrival time of the acoustic emission signal to each acoustic emission signal sensor as follows:

[0032] (1) After each acoustic emission sensor acquires an acoustic emission signal, the acoustic emission signal is preprocessed to determine the amplitude of the acoustic emission signal, and a signal threshold is set based on the amplitude;

[0033] (2) For the acoustic emission signal, the time series of the acoustic emission signal is recorded as T[1,T n ], indicating that the time length of the acoustic emission signal is from 1 to T nTime; the time point corresponding to the signal threshold is t n , then from the time series T[1,T n ] to extract 1 to t n The acoustic emission signal at time T[1,t n ] as a simplified signal;

[0034] (3) For the intercepted time series T[1,t n ], calculate the information criterion AIC value corresponding to each moment; where t n The AIC value at the moment AIC(t n ) is calculated as follows:

[0035] AIC(t n )=[t n ×lg[Var[T(1,t n )]]+(T n -t n -1)×lg(Var(T[1+t n ,T n ]))]

[0036] Among them, Var is a variance function, Var[T(1,t n )] represents 1 to t n The variance of each acoustic emission signal data point within the time range;

[0037] Therefore, in the time series intercepted from the acoustic emission signal acquired by each acoustic emission sensor, an AIC value will be calculated at each time point, so that an AIC curve can be obtained by fitting;

[0038] (4) Multiply the AIC value by a Billy factor K to reduce the interference of cluttered signals:

[0039]

[0040] Among them, max[T(1,t n )] represents the acoustic emission signal in the intercepted time series T[1,t n ], max(AIC) is the peak value of the AIC curve;

[0041] The minimum value obtained by multiplying each point on the AIC curve by the Billy factor K is the time when the acoustic emission signal reaches the acoustic emission sensor.

[0042] Furthermore, the signal threshold is half of the amplitude of the acoustic emission signal.

[0043] A terminal device comprises a processor, a memory and a computer program stored in the memory; the characteristic is that when the processor executes the computer program, the damage location method based on the improved A* algorithm and BP neural network is implemented.

[0044] A computer-readable storage medium stores a computer program; the characteristic of the computer program is that when the computer program is executed by a processor, the damage location method based on the improved A* algorithm and BP neural network is implemented.

[0045] Compared with the prior art, the present invention has the following technical features:

[0046] 1. This paper proposes an anisotropic A* algorithm with OD-WV to identify the search path between the source and the sensor. The theoretical basis for the fastest search path in anisotropic structures is established by introducing the OD-VC parameter, which changes the basic assumption of the shortest search path and breaks the corresponding relationship between the "shortest path" and the "fastest start" in isotropic materials.

[0047] 2. The present invention provides a joint automatic picking algorithm for determining the arrival time of the acoustic emission signal. Compared with the threshold value method, manual picking method, AIC algorithm and other methods commonly used now, this method can obtain the arrival time of the acoustic emission signal more accurately and efficiently. When the joint automatic picking algorithm is used, it is closest to the arrival time of manual picking and the calculation efficiency is significantly improved compared with other algorithms. Compared with the Fast-AIC algorithm, the time is shortened by about 50%.

[0048] 3. The AA*I-BPNN model formed by the present invention in combination with the anisotropic A* algorithm has a smaller training data set, generates damage localization with higher efficiency than the physical drive model, and has comparable accuracy with the reported LS method. This method not only avoids the long-term and high-cost offline finite element / experiment, but also provides considerable online potential for distinguishing damage classification and identifying damage extent.

[0049] 4. The sparse sensor matrix arranged only at the four corners of the monitoring area in the present invention effectively reduces the amount of data required for damage location, thereby breaking the limitation of data processing delay in online monitoring. It not only meets the online monitoring requirements of macroscopic CFRCs structures (<1min), but also realizes offline damage diagnosis of a batch of aircraft components with similar environmental conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of grid division and arrangement of acoustic emission sensors in a monitoring area on a sample structure in one embodiment of the present invention;

[0051] Figure 2A schematic diagram of the structure of a BP neural network in one embodiment of the present invention;

[0052] Figure 3 A schematic diagram of an acoustic emission signal acquired by an acoustic emission sensor and intercepted by using a signal threshold in one embodiment of the present invention;

[0053] Figure 4 It is the experimental result figure of Example 1;

[0054] Figure 5 This is a diagram of the experimental results of Example 2. DETAILED DESCRIPTION

[0055] Referring to the accompanying drawings, the present invention provides a damage location method based on an improved A* algorithm and a BP neural network, comprising the following steps:

[0056] Step 1: Select a sample structure with the same material as the product to be tested, determine the propagation characteristics of the acoustic emission signal on the sample structure in all directions through experimental methods, and obtain the orientation-dependent wave velocity (OD-WV) and the minimum wave velocity v min ; The wave velocity parameter OD-WV refers to the propagation speed of the acoustic emission signal Lamb wave in each polar angle direction in the monitoring area set on the sample structure.

[0057] Among them, the products to be tested can be, for example, composite material products such as aircraft wall panels and wings; the materials of these products have anisotropic characteristics, so this scheme determines the propagation characteristics of the acoustic emission signal through design experiments; the shape of the sample structure can be, for example, a plate-like structure.

[0058] The test method is as follows:

[0059] First, a sample structure is selected, which is consistent with the material of the product to be tested; a monitoring area is set on the sample structure, which can be set according to needs, such as a circle, rectangle, etc.; a polar coordinate system is established with the center of the monitoring area as the pole O, and the polar axis is defined as a ray from the pole to any direction of the edge of the sample structure, and the polar angle is defined as the angle between any straight line passing through the pole and the polar axis. The polar axis corresponds to a polar angle of 0°, such as Figure 1 shown.

[0060] The polar axis in the monitoring area is rotated counterclockwise at fixed intervals in the counterclockwise direction, so that the monitoring area is divided into multiple polar angle areas of the same size; for example, the monitoring area can be divided into 24 polar angle areas at intervals of 15°; an acoustic emission device is arranged at the pole, and then the wave velocity parameter OD-WV in each polar angle direction of the monitoring area is determined respectively, as follows:

[0061] Taking the polar axis as an example, the polar axis corresponds to the polar angle direction of 0°. Two acoustic emission sensors are arranged at intervals on the polar axis. The acoustic emission device is used to generate Lamb wave signals. The acoustic emission signals are captured by the two acoustic emission sensors on the polar axis. The distance S between the two acoustic emission sensors and the time difference △t between the Lamb wave signals reaching the two sensors can be used to obtain the wave velocity parameter OD-WV=S / △t in the polar angle direction of 0°.

[0062] At each position of polar axis rotation, acoustic emission sensors are arranged at intervals according to the same method as above, and the OD-WV in the corresponding polar angle direction is determined, thereby obtaining the OD-WV in each polar angle direction after division.

[0063] The OD-WV in different polar angle directions is fitted by using a cubic spline curve to obtain an OD-WV fitting curve; thus, the OD-WV in any polar angle direction can be determined by the OD-WV fitting curve.

[0064] The minimum wave speed v min It is the minimum value of OD-WV on the OD-WV fitting curve.

[0065] Step 2, using the minimum wave speed to normalize the direction-related wave speed parameter OV-WV, expressed as follows:

[0066] α(θ)=v(θ) / v min

[0067] v(θ) represents the OD-WV in the direction of the polar angle θ, and α(θ) represents the normalized OD-WV.

[0068] Step 3, gridding the monitoring area on the sample structure, setting the grids for arranging acoustic emission sensors in the detection area, and using the A* algorithm that introduces the wave velocity parameter OV-WV to calculate the shortest path between each grid to the grid where each acoustic emission sensor is located, so as to determine the arrival time of the acoustic emission signal obtained by each acoustic emission sensor when each grid is used as an acoustic emission source; taking the time difference after the arrival time determined by each acoustic emission sensor is subtracted from each other as a group of samples, and the label of the sample is the coordinate of the corresponding grid, so as to construct a training data set.

[0069] The size of the grid depends on the positioning accuracy requirements and the size of the monitoring area. For example, if the monitoring area is 1m×1m, it can be divided into 100×100 grids. Then the side length of each grid is 10mm and the positioning accuracy is also 10mm. The positioning accuracy can be further improved as the grid is subdivided.

[0070] The acoustic emission sensors are arranged on the grid at the edge of the detection area, and the specific number is at least three. In this embodiment, four acoustic emission sensors are set, which are arranged on the grids at the four corners of the monitoring area respectively; in actual detection applications, the acoustic emission sensors arranged on the product to be detected and the monitoring area should be consistent with this.

[0071] This scheme assumes that the acoustic emission signal is generated in each grid. Due to the characteristics of anisotropic materials, the shortest path between the acoustic emission signal and the acoustic emission sensor is not necessarily a direct line between the two. Therefore, this scheme proposes an A* algorithm that introduces the wave velocity parameter OV-WV to determine the shortest path. The cost function f(n) of the algorithm in the path search process from a grid a to a grid c where an acoustic emission sensor is located is expressed as:

[0072] f(n)=g(n)+wh(n)

[0073] Among them, w is the weight coefficient, which can be adjusted according to the distance from the grid to the acoustic emission sensor, g(n) represents the actual cost from grid a to the current search grid b, and h(n) represents the estimated cost of the shortest path from grid a to grid c where the acoustic emission sensor is located, which is expressed as follows:

[0074]

[0075] Among them, (x a ,y a ) represents the coordinates of grid a, (x b ,y b ) represents the coordinates of the current search grid b, (x c ,y c ) represents the coordinates of the grid c where the acoustic emission sensor is located, and α(θ) represents the normalized OD-WV corresponding to the angle θ between the line between the grid a and the current search grid b and the polar axis.

[0076] After determining the shortest path d from grid a to grid c where the acoustic emission sensor is located, compare d with the minimum wave velocity v min The ratio of is taken as the arrival time t of the acoustic emission signal of the acoustic emission sensor at the grid c c1 According to the same method, the arrival time t of the acoustic emission signals of the other three acoustic emission sensors in this embodiment can be obtained. c2、 t c3、 t c4 ; Therefore, by subtracting the arrival times from each other, we can get 6 time differences, namely t c1 -t c2 ,t c1 -t c3 ,t c1 -t c4 ,t c2 -tc3 ,t c2 -t c4 and t c3 -t c4 ; Take this set of time differences as a set of samples, and the corresponding labels are the coordinates of grid a (x a ,y a ), so as to construct a training data set using the samples corresponding to each grid in the monitoring area.

[0077] In this step, the improved anisotropic A* algorithm is used to establish the theoretical basis of the fastest search path in anisotropic structures by introducing the OD-VC parameter. This parameter changes the basic assumption of the shortest search path, breaks the corresponding relationship between the "shortest path" and the "fastest start" in isotropic materials, and obtains the theoretical time from each grid node to the sensor. Further, the time difference between the arrival times of the sensors can be calculated.

[0078] Step 4: construct a BP neural network, use the samples in the training data set to train the BP neural network, and save the trained BP neural network model (Anisotropic A*Informed-BP Neural Network, AA*I-BPNN).

[0079] like Figure 2 As shown, the BP neural network includes an input layer, a hidden layer and an output layer, wherein the number of neurons in the input layer is consistent with the number of time differences in the sample, which is 6 in this embodiment; the number of neurons in the hidden layer is set to 14 to ensure that the network captures the complex relationship between input features without overfitting; the number of neurons in the output layer is 2, so the network structure is 6-14-2. The network uses the hyperbolic tangent function Tanh as the activation function, and the samples of the input layer are passed to the output layer by the hyperbolic tangent function Tanh after being processed by the hidden layer.

[0080] When training the network, the mean square error loss function is used, and the gradient descent algorithm is used to adjust the weights and biases in each iteration; in order to accelerate the effective learning and convergence of the weights, the maximum number of training times is set to 500. When the loss function does not change for more than two iterations, the training is stopped to prevent overfitting.

[0081] Step 5, for the product to be tested, in actual application, the monitoring area and the acoustic emission sensor are first arranged on the product to be tested in the same way as on the sample structure, and the acoustic emission sensor is used to monitor the acoustic emission signal; when the monitoring area generates an acoustic emission signal due to structural damage or other reasons, the acoustic emission signal is captured by the acoustic emission sensor, and the arrival time of the acoustic emission signal to each acoustic emission signal sensor is determined by a joint automatic picking algorithm combining an adaptive threshold and an Akaike information criterion AIC, thereby determining the time difference of the arrival time and inputting it into the trained BP neural network model, and the result output by the BP neural network model is the coordinate of the location where the acoustic emission signal is generated.

[0082] Although the acoustic emission sensor has a corresponding timestamp when it captures the acoustic emission signal, it is affected by various factors such as environmental noise, and its time accuracy is not high enough. In addition, there are problems such as inconsistency in the time axes of various acoustic emission sensors. Therefore, if the time difference of the arrival time is directly calculated using its timestamp, it may affect the recognition accuracy.

[0083] The present invention provides a joint automatic picking algorithm to determine the arrival time of the acoustic emission signal to each acoustic emission signal sensor, which is as follows:

[0084] (1) After each acoustic emission sensor obtains the acoustic emission signal, the acoustic emission signal is preprocessed to determine the amplitude of the acoustic emission signal; and a signal threshold is set, which is half of the acoustic emission signal amplitude.

[0085] (2) For the acoustic emission signal, the time series of the acoustic emission signal is recorded as T[1,T n ], indicating that the time length of the acoustic emission signal is from 1 to T n Time; the time point corresponding to the signal threshold is t n , then from the time series T[1,T n ] to extract 1 to t n The acoustic emission signal at time T[1,t n ] as a simplified signal.

[0086] like Figure 3 The actual arrival time of an acoustic emission signal is the onset time, which is the value corresponding to the red point. In order to obtain the specific position of this value, it is usually necessary to calculate the AIC value, but the amount of calculation is huge. In order to reduce the amount of data calculated, this scheme adopts an adaptive threshold, that is, Figure 2 For the green line in the middle, only the part before the intersection of the line and the signal needs to be calculated. Therefore, the original calculation from 1 to T n Each data point only needs to calculate 1 to t n data.

[0087] (3) For the intercepted time series T[1,t n ], calculate the information criterion AIC value corresponding to each moment; where t n The AIC value at the moment AIC(t n ) is calculated as follows:

[0088] AIC(t n )=[t n ×lg[Var[T(1,t n )]]+(T n -t n -1)×lg(Var(T[1+t n ,T n ]))]

[0089] Among them, Var is a variance function, Var[T(1,t n )] represents 1 to t n The variance of each acoustic emission signal data point within the time range.

[0090] Therefore, in the time series intercepted from the acoustic emission signal acquired by each acoustic emission sensor, an AIC value will be calculated at each time point, so that an AIC curve can be obtained by fitting.

[0091] (4) Multiply the AIC value by a Billy factor K to reduce the interference of cluttered signals:

[0092]

[0093] Among them, max[T(1,t n )] represents the acoustic emission signal in the intercepted time series T[1,t n ], max(AIC) is the peak of the AIC curve.

[0094] Therefore, the minimum value obtained by multiplying each point on the AIC curve by the Billy factor K is the time when the acoustic emission signal reaches the acoustic emission sensor.

[0095] After determining the arrival time of the acoustic emission signal in each acoustic emission sensor, the arrival times are subtracted from each other to obtain a set of time differences, which are input into the trained BP neural network model to locate the emission source of the acoustic emission signal.

[0096] Embodiment 1:

[0097] This example is verified by using CFRCs composite material samples: the experiment uses a T300CFRCs sample with a size of 600mm×600mm×2mm, and divides the area of ​​400mm×400mm into 20×20 grids, with each grid node as a possible damage point; the acoustic emission signal is triggered by the excitation sensor. The experimental results of the training data generated by the A* algorithm are shown in the attached figure. Figure 4 As shown, the experimental coordinates, based on the anisotropy A * The prediction results of the algorithm and the traditional A * The algorithm's predictions are represented by red circles, green pentagons, and blue triangles, respectively.

[0098] Experiments show that the predicted results are close to the actual position, with a maximum error of less than 2.80 cm, an average error of 1.46 cm (2.58%), and a positioning time of 3.14 ms. The maximum error of the positioning results based on the traditional A* algorithm is 6.32 cm, and the average error is 3.85 cm (larger than the sensor diameter). These results show that the proposed A* algorithm is more accurate than the traditional A* algorithm. * The I-BPNN model produces accurate and efficient prediction and positioning, bringing potential possibilities for the development of online structural health monitoring.

[0099] Embodiment 2:

[0100] In this example, the aviation component verification is as follows: a 4mm thick aviation composite component is selected, the shape is an irregular curved surface, and the monitoring area is 600mm×600mm. A small hammer is used to perform an impact test at 25 random points, and the signal is recorded and verified. * The applicability of the I-BPNN model. The experimental results are shown in the attached Figure 5 The average error is 2.60cm, the positioning time is 3.32ms, the relative error is only 3.07%, and the error is controlled within 5%, which meets the actual accuracy requirements of the project. The statistical positioning results show that the error in the upper half of the monitoring area is slightly larger than that in the lower half of the monitoring area. This is because the curvature of the engineering structure gradually increases to form an obvious arc in the upper half of the monitoring area. The results show that the present invention has good applicability in different types of composite materials and complex geometric structures, meeting the actual engineering application requirements.

[0101] It should be pointed out that the AA*I-BPNN model provided by this scheme is also applicable to complex structures containing empty areas. At the same time, as the number of sensors increases, the positioning accuracy will be further improved.

[0102] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A damage location method based on improved A* algorithm and BP neural network, characterized in that: include: Step 1, select a sample structure of the same material as the product to be tested, determine the propagation characteristics of the acoustic emission signal along each direction on the sample structure through an experimental method, and obtain the direction-related wave velocity parameter OD-WV and the minimum wave velocity; wherein the wave velocity parameter OD-WV refers to the propagation velocity of the acoustic emission signal Lamb wave in each polar angle direction in the monitoring area set on the sample structure; Step 2, using the minimum wave speed to normalize the direction-related wave speed parameter OV-WV; Step 3, gridding the monitoring area on the sample structure, setting the grids for arranging acoustic emission sensors in the detection area, and using the A* algorithm with the wave velocity parameter OV-WV introduced to calculate the shortest path between each grid to each grid where the acoustic emission sensor is located, so as to determine the arrival time of the acoustic emission signal obtained by each acoustic emission sensor when each grid is used as an acoustic emission source; taking the time difference after the arrival time determined by each acoustic emission sensor is subtracted from each other as a group of samples, and the label of the sample is the coordinate of the corresponding grid, so as to construct a training data set; Step 4, constructing a BP neural network, using the samples in the training data set to train the BP neural network, and saving the trained BP neural network model; Step 5, for the product to be tested, in actual application, the monitoring area and the acoustic emission sensor are first arranged on the product to be tested in the same way as on the sample structure, and the acoustic emission sensor is used to monitor the acoustic emission signal; when the monitoring area generates an acoustic emission signal due to structural damage or other reasons, the acoustic emission signal is captured by the acoustic emission sensor, and the arrival time of the acoustic emission signal to each acoustic emission signal sensor is determined by a joint automatic picking algorithm combining an adaptive threshold and an Akaike information criterion AIC, thereby determining the time difference of the arrival time and inputting it into the trained BP neural network model, and the result output by the BP neural network model is the coordinate of the location where the acoustic emission signal is generated.

2. The damage location method based on the improved A* algorithm and BP neural network according to claim 1 is characterized in that: The test method is as follows: The monitoring area is set on the sample structure, and a polar coordinate system is established with the center of the monitoring area as the pole. The polar axis is defined as a ray from the pole to any direction of the edge of the sample structure. The polar angle is defined as the angle between any straight line passing through the pole and the polar axis. The polar axis corresponds to a polar angle of 0°. The polar axis in the monitoring area is rotated counterclockwise at fixed intervals in the counterclockwise direction, so that the monitoring area is divided into multiple polar angle areas of the same size; an acoustic emission device is arranged at the pole, and then the wave velocity parameter OD-WV in each polar angle direction of the monitoring area is determined respectively, as follows: The polar axis corresponds to the polar angle direction of 0°. Two acoustic emission sensors are arranged at intervals on the polar axis. Lamb wave signals are generated by the acoustic emission device. The acoustic emission signals are captured by the two acoustic emission sensors on the polar axis. The wave velocity parameter OD-WV=S / △t in the polar angle direction of 0° can be obtained by using the spacing S between the two acoustic emission sensors and the time difference △t between the Lamb wave signals reaching the two sensors. At each position of polar axis rotation, acoustic emission sensors are arranged at intervals according to the same method, and the OD-WV in the corresponding polar angle direction is determined, thereby obtaining the OD-WV in each polar angle direction after division; The OD-WV in different polar angle directions is fitted by using a cubic spline curve to obtain an OD-WV fitting curve; thus, the OD-WV in any polar angle direction can be determined by the OD-WV fitting curve.

3. The damage location method based on the improved A* algorithm and BP neural network according to claim 1 is characterized in that: The minimum wave speed is used to normalize the direction-related wave speed parameter OV-WV, which is expressed as: α(θ)=v(θ) / v min v(θ) represents the OD-WV in the direction of the polar angle θ, α(θ) represents the normalized OD-WV, and v min Indicates the minimum wave speed.

4. The damage location method based on improved A* algorithm and BP neural network according to claim 1 is characterized in that: The A* algorithm that introduces the wave velocity parameter OV-WV is as follows: The acoustic emission sensors are arranged on the grids at the edge of the detection area; assuming that the acoustic emission signal is generated in each grid, the cost function f(n) in the path search process from a grid a to a grid c where the acoustic emission sensor is located based on the A* algorithm is expressed as: f(n)=g(n)+wh(n) Among them, w is the weight coefficient, g(n) represents the actual cost from grid a to the current search grid b, and h(n) represents the estimated cost of the shortest path from grid a to grid c where the acoustic emission sensor is located, which is expressed as follows: Among them, (x a ,y a ) represents the coordinates of grid a, (x b ,y b ) represents the coordinates of the current search grid b, (x c ,y c ) represents the coordinates of the grid c where the acoustic emission sensor is located, α(θ) represents the normalized OD-WV corresponding to the angle θ between the line connecting the grid a and the current search grid b and the polar axis; After determining the shortest path d from grid a to grid c where the acoustic emission sensor is located, compare d with the minimum wave velocity v min The ratio of is taken as the arrival time of the acoustic emission signal of the acoustic emission sensor at grid c; in the same way, the arrival time from grid a to the grids where the other sensors are located is taken as a group of samples after the arrival time is subtracted from each other, and the corresponding label is the coordinate (x a ,y a ), so as to construct a training data set using the samples corresponding to each grid in the monitoring area.

5. The damage location method based on improved A* algorithm and BP neural network according to claim 1 is characterized in that: The BP neural network includes an input layer, a hidden layer and an output layer, wherein the number of neurons in the input layer is consistent with the number of time differences in the sample, the network uses the hyperbolic tangent function Tanh as an activation function, and the samples of the input layer are processed by the hidden layer and passed to the output layer by the hyperbolic tangent function Tanh; When training the network, the mean square error loss function is used, and the gradient descent algorithm is used to adjust the weights and biases in each iteration.

6. The damage location method based on improved A* algorithm and BP neural network according to claim 1 is characterized in that: The joint automatic picking algorithm determines the arrival time of the acoustic emission signal to each acoustic emission signal sensor as follows: (1) After each acoustic emission sensor acquires an acoustic emission signal, the acoustic emission signal is preprocessed to determine the amplitude of the acoustic emission signal, and a signal threshold is set based on the amplitude; (2) For the acoustic emission signal, the time series of the acoustic emission signal is recorded as T[1,T n ], indicating that the time length of the acoustic emission signal is from 1 to T n Time; the time point corresponding to the signal threshold is t n , then from the time series T[1,T n ] to extract 1 to t n The acoustic emission signal at time T[1,t n ] as a simplified signal; (3) For the intercepted time series T[1,t n ], calculate the information criterion AIC value corresponding to each moment; where t n The AIC value at the moment AIC(t n ) is calculated as follows: AIC(t n )=[t n ×lg[Var[T(1,t n )]]+(T n -t n -1)×lg(Var(T[1+t n ,T n ]))] Among them, Var is a variance function, Var[T(1,t n )] represents 1 to t n The variance of each acoustic emission signal data point within the time range; Therefore, in the time series intercepted from the acoustic emission signal acquired by each acoustic emission sensor, an AIC value will be calculated at each time point, so that an AIC curve can be obtained by fitting; (4) Multiply the AIC value by a Billy factor K to reduce the interference of cluttered signals: Among them, max[T(1,t n )] represents the acoustic emission signal in the intercepted time series T[1,t n ], max(AIC) is the peak value of the AIC curve; The minimum value obtained by multiplying each point on the AIC curve by the Billy factor K is the time when the acoustic emission signal reaches the acoustic emission sensor.

7. The damage location method based on improved A* algorithm and BP neural network according to claim 6 is characterized in that: The signal threshold is half of the acoustic emission signal amplitude.

8. A terminal device comprising a processor, a memory and a computer program stored in the memory; characterized in that: When the processor executes the computer program, it implements the damage localization method based on the improved A* algorithm and BP neural network according to any one of claims 1 to 7.

9. A computer-readable storage medium, wherein a computer program is stored in the medium; characterized in that: When the computer program is executed by a processor, the damage localization method based on the improved A* algorithm and BP neural network according to any one of claims 1 to 7 is implemented.

Images