A convolutional neural network-based impact location and energy detection method and system

Abstract

The present invention relates to the field of structural health monitoring technology, and in particular to a method and system for impact positioning and energy detection based on a convolutional neural network. The method adopts a convolutional neural network model to roughly locate the impact area, and based on this, accurately locates the specific impact position within the impact area through a centroid weighted algorithm based on DTW; at the same time, the energy of the impact response signal is used to characterize the magnitude of the impact energy. This method has accurate positioning, high impact inversion efficiency, and simple operation. It not only avoids the influence of structural complexity on positioning accuracy, but also effectively controls the impact energy estimation error. While reducing maintenance costs, it also effectively improves the efficiency of positioning and detection.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health monitoring, and in particular to an impact location and energy detection method and system based on a convolutional neural network. Background Art

[0002] Composite materials offer advantages such as high specific strength, large specific modulus, and structural design flexibility, offering significant advantages in lightweight manufacturing, economical operation and maintenance, and advanced design. Consequently, they are widely used in the aerospace sector. However, composite structures are inevitably subject to impact during manufacturing, service, and maintenance, which can cause various nearly invisible internal damage, degrading the load-bearing performance of the structure and seriously threatening the in-service safety of aerospace equipment. Therefore, online impact monitoring of composite structures to ensure equipment safety is an urgent engineering challenge.

[0003] Conventional nondestructive testing technologies currently used for impact damage detection include ultrasonic scanning, eddy current, and thermal imaging. These methods usually require point-by-point and area-by-area scanning and measurement of the structure, which is time- and economically expensive for large structures. Furthermore, they can only be monitored offline, making it difficult to meet the needs of airborne online monitoring. Existing impact location methods for complex composite structures generally use a location method based on the propagation velocity of stress waves, which involves extracting the time-domain features of stress waves and compensating for velocities in different propagation directions. This method has a large workload and its accuracy is greatly affected by the structure. The use of a system model-based method involves solving the transfer function through the impact response and constructing a system model. This method requires a large amount of mathematical calculations and parameter adjustments to achieve model convergence, resulting in low efficiency.

[0004] Impact energy, another characteristic parameter characterizing impact damage in composite structures, is crucial for assessing the structural condition of composite materials. Existing techniques typically estimate impact energy by inverting the time history of the impact load using a response model. However, this requires a complex system identification model for structures with dense ribs.

[0005] Therefore, based on the shortcomings of existing technologies and the needs of engineering applications, there is an urgent need for a method of impact positioning and energy detection that can more accurately, simply and effectively monitor the impact damage of composite structures online. Summary of the Invention

[0006] To address the shortcomings of the above-mentioned prior art methods for impact location and energy detection of composite materials, such as inaccurate positioning, low impact inversion efficiency, and complex operation, the present invention provides a method for impact location and energy detection based on a convolutional neural network, comprising the following steps:

[0007] The region processing step comprises disposing m sensors for receiving external impact signals on the composite material structure to be inspected, and dividing the surface of the composite material structure to be inspected into at least M regions.

[0008] The sample acquisition step is to perform impact tests on a plurality of training points and marking points in each of the areas to obtain a first sample database and a second sample database containing impact signals; and Impact the surface of the composite material structure and obtain the fixed energy according to the impact signal received by several sensors Characteristic value under impact .

[0009] In the model training step, a 1D-CNN neural network is constructed with an input of (l, m) and an output of (M, 1), where l is the signal length of each sensor; the first sample database is divided into a training set and a test set, and the sets are input into the 1D-CNN neural network for training and verification to obtain an impact event monitoring model.

[0010] In the rough positioning step, if an impact event occurs, several sensors will receive the impact signal S impact ; The impact signal S impact The input is fed into the impact event monitoring model, and then the impact area where the impact event occurred is determined based on the output results to complete the coarse positioning.

[0011] In the fine positioning step, based on the impact area determined by coarse positioning, the impact signal S is calculated according to the DTW algorithm. impact DTW distance to all markers in the impact area of ​​the second sample database ; the obtained The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm to achieve precise positioning.

[0012] Energy detection step, based on the impact signal S received by sensors in all areas impact Obtain energy characterization value Calculate the energy of the impact according to the following formula :

[0013] .

[0014] In one embodiment, the region processing step further includes numbering each region to obtain a signal label; and in the sample acquisition step, the impact signal in each region is mapped into a corresponding signal label for storage.

[0015] In one embodiment, in the sample acquisition step, the sample acquisition step of the first sample database is as follows: N (N≥1) training points are selected in each of the regions to perform impact tests to obtain corresponding impact signals, and then the impact signals on each of the markers in each of the regions are mapped into corresponding signal labels and stored in the first sample database;

[0016] The sample acquisition step of the second sample database is: selecting B marking points in each area and performing impact tests respectively to obtain corresponding impact signals, and then storing the impact signals obtained from the impact tests of each marking point in each area as samples.

[0017] In one embodiment, in the model training step, the samples in the first sample database are divided into a training set and a test set in a ratio of 7:3.

[0018] In one embodiment, in the fine positioning step, the obtained The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm, specifically including the following formula:

[0019] , ;

[0020] ;

[0021] in, Represents the weighting coefficient of all marked points in the impact area; , Respectively represent the coordinate positions of all marked points in the impact area; 、 That is the impact position coordinate of the impact event.

[0022] In one embodiment, in the sample acquisition step, the following formula is used to obtain the fixed energy Characteristic value under impact :

[0023] ;

[0024] ;

[0025] in, is the expression of the shock signal obtained by the sensor, is the energy value of the corresponding impact signal; is the number of sensors deployed on the composite material structure to be detected, For the The energy value of the impulse response signal corresponding to each sensor;

[0026] In the energy detection step, the energy characterization value when the impact event occurs is obtained by the following formula: :

[0027] ;

[0028] in, When the shock event occurs The energy value of the impulse response signal corresponding to each sensor.

[0029] In one embodiment, the energy correction coefficient acquisition step is further included. After performing a test on the surface of the composite material structure with a fixed energy E0, one of the regions is selected as a reference region, and the energy characterization value of the reference region is obtained. , calculate the energy correction coefficient of each area relative to the reference area , the formula is as follows:

[0030] ;

[0031] ;

[0032] in, is the number of sensors in the reference area, In the reference area The energy value of the impulse response signal corresponding to each sensor.

[0033] In one embodiment, an energy correction step is further included, wherein the impact area determined based on the rough positioning is corrected according to the energy correction coefficient. Obtain the energy correction coefficient of the impact area relative to the reference area And the corrected impact energy value is obtained by the following formula :

[0034] .

[0035] The present invention also provides an impact location and energy detection system based on a convolutional neural network, comprising a region processing module for arranging a plurality of sensors for receiving external impact signals on a composite material structure to be detected, and dividing the surface of the composite material structure to be detected into at least M regions;

[0036] The sample acquisition module is used to perform impact tests on a number of training points and marking points in each area to obtain a first sample database and a second sample database containing impact signals; and Impact the surface of the composite material structure and obtain the fixed energy according to the impact signal received by several sensors Characteristic value under impact ;

[0037] A model training module is used to construct a 1D-CNN neural network with an input of (l, m) and an output of (M, 1), where l is the signal length of each sensor; the first sample database is divided into a training set and a test set, and the sets are input into the 1D-CNN neural network for training and verification to obtain an impact event monitoring model;

[0038] Coarse positioning module, if a certain impact event occurs, some of the sensors will receive the impact signal S impact ; Used to convert the impact signal S impact Input into the impact event monitoring model, and then determine the impact area of ​​the impact event based on the output results to complete the coarse positioning;

[0039] The fine positioning module is used to calculate the impact signal S according to the DTW algorithm based on the impact area determined by the coarse positioning. impact The DTW distance L of all the markers in the impact area of ​​the second sample database i ; The obtained L i The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm to achieve precise positioning;

[0040] Energy detection module, used to detect the impact signal S received by sensors in all areas impact Obtain energy characterization value Calculate the energy of the impact according to the following formula :

[0041] .

[0042] In one embodiment, an energy correction module is further included for selecting one of the regions as a reference region after performing a test on the surface of the composite material structure with a fixed energy E0, and obtaining an energy characterization value of the reference region. , calculate the energy correction coefficient of each area relative to the reference area ;

[0043] Based on the impact area determined by coarse positioning, according to the energy correction coefficient Obtain the energy correction coefficient of the impact area relative to the reference area And the corrected impact energy value is obtained by the following formula :

[0044] .

[0045] Based on the above, compared with the existing technology, the convolutional neural network-based impact location and energy detection method provided by the present invention uses a convolutional neural network model to roughly locate the impact area. Based on this, a DTW-based centroid weighting algorithm is used to precisely locate the specific impact position within the impact area. The energy of the response signal is also used to characterize the impact energy. This method not only avoids the impact of structural complexity on positioning accuracy but also effectively controls the impact energy estimation error. This method reduces maintenance costs while effectively improving the efficiency of positioning and detection.

[0046] Other features and advantages of the present invention will be set forth in the following description, and in part will be obvious from the description, or may be learned by practicing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work. The positional relationships described in the drawings in the following description are based on the directions of the components drawn in the diagrams, unless otherwise specified.

[0048] Figure 1 A flowchart of the steps of the impact location and energy detection method provided by the present invention;

[0049] Figure 2 Flowchart of the coarse positioning steps;

[0050] Figure 3 Schematic diagram of the DTW calculation process;

[0051] Figure 4 Flowchart of precise positioning steps;

[0052] Figure 5 It is a flow chart of energy detection steps;

[0053] Figure 6 It is a flow chart of energy detection steps and energy correction steps;

[0054] Figure 7 A schematic diagram of the impact location and energy detection system provided by the present invention;

[0055] Figure 8 A schematic diagram of the structure of the regional processing steps for the composite stiffened panel;

[0056] Figure 9 Schematic diagram of 1D-CNN model training results;

[0057] Figure 10 Select the location of the marker point in the area;

[0058] Figure 11 Schematic diagram of the coarse positioning result;

[0059] Figure 12 Schematic diagram of the precise positioning results;

[0060] Figure 13 Schematic diagram comparing the impact energy estimation errors before and after regional correction compensation. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments; the technical features designed in different implementation modes of the present invention described below can be combined with each other as long as they do not conflict with each other; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0062] In the description of the present invention, it should be noted that all terms used in the present invention (including technical terms and scientific terms) have the same meanings as those generally understood by ordinary technicians in the field to which the present invention belongs, and should not be understood as limiting the present invention; it should be further understood that the terms used in the present invention should be understood to have the same meanings as these terms in the context of this specification and the relevant field, and should not be understood in an idealized or overly formal sense, unless explicitly defined as such in the present invention.

[0063] To solve the existing problems of impact location and impact energy detection methods for complex composite structures, such as the location results are greatly affected by the structure, the system modeling workload is large, and the impact inversion efficiency is low. Figure 1 The present invention provides a method for impact location and energy detection based on a convolutional neural network, which comprises at least the following steps:

[0064] The region processing step comprises disposing m sensors for receiving external impact signals on the composite material structure to be inspected, and dividing the surface of the composite material structure to be inspected into at least M regions.

[0065] It should be noted that the sensors can be piezoelectric sensors, acoustic emission sensors, and the like. The number of sensors deployed can be set based on actual needs and is not limited here. Preferably, to facilitate subsequent calculations and analysis, the surface of the composite material structure can be divided into M regions of equal or completely equal area.

[0066] The sample acquisition step is to perform impact tests on a plurality of training points and marking points in each of the areas to obtain a first sample database and a second sample database containing impact signals; and Impact the surface of the composite material structure and obtain the fixed energy according to the impact signal received by several sensors Characteristic value under impact .

[0067] In specific implementation, the sample acquisition steps of the first sample database are as follows: N (N≥1) training points are selected in each area and impact tests are performed to obtain corresponding impact signals, and then the impact signals at each training point in each area are mapped into corresponding signal labels and stored in the first sample database. The mapping relationship of the first sample database can be referred to as shown in the following table:

[0068]

[0069] The sample acquisition step of the second sample database is as follows: select B marking points in each area and perform impact tests to obtain corresponding impact signals, and then store the impact signals obtained from the impact tests of each marking point in each area as samples. The mapping relationship of the second sample database can be referred to the following table:

[0070]

[0071] The model training step involves constructing a 1D-CNN neural network with an input of (l, m) and an output of (M, 1), where l is the signal length of each sensor, m is the number of sensors, and M is the number of regions in the composite structure. The first sample database is divided into a training set and a test set, which are then fed into the 1D-CNN neural network for training and validation to develop an impact event monitoring model. It should be noted that conventional methods for training and validating convolutional neural networks are not discussed here. Preferably, the samples in the first sample database can be divided into a training set and a test set in a ratio of 7:3.

[0072] In the rough positioning step, if an impact event occurs, several sensors will receive the impact signal S impact ; The impact signal S impact The signal is input into the impact event monitoring model, and then the impact area where the impact event occurred is determined based on the output result to complete the coarse positioning; that is, the signal label output by the neural network is the impact area where the impact event occurred.

[0073] For precise positioning steps, please refer to Figure 4Based on the impact area determined by coarse positioning, the impact signal S is calculated according to the DTW algorithm. impact DTW distance to all markers in the impact area of ​​the second sample database ; the obtained The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm to achieve precise positioning.

[0074] In specific implementation, this embodiment uses the weighted centroid algorithm (WCA) to obtain the precise impact position within the impact area based on the rough positioning. It has the characteristics of simple calculation and strong adaptability. The specific steps are to first construct a rectangular coordinate system for each area, and then calculate the impact signal S using the DTW algorithm. impact DTW distance to all markers in the impact area of ​​the second sample database In this embodiment, the DTW (Dynamic Time Warping) algorithm is applied to the precise positioning of the impact positioning, mainly to compare the similarity of the time series of the two impact signals, that is, to calculate the Euclidean distance of the two time series. The specific calculation process is as follows: Figure 3 As shown, taking time series X and time series Y as an example, under the premise of aligning time series X and time series Y, the distance matrix between the points of the two sequences is calculated, and then the minimum value of the sum of the path distances calculated according to the DTW path mapping relationship in the path composed of p1~pk is found. This path distance is the DTW distance , where the expression is: ;in, Indicates the impact signal S generated at the impact position impact Impact signal with reference mark DTW distance between impact Impact signal indicating actual impact location; Indicates the first The impact signal of the marking point.

[0075] Finally, the obtained DTW distance The weight factor of the weighted centroid positioning algorithm is used to calculate the impact position coordinates (x, y) within the impact area, specifically including the following formula:

[0076] , ;

[0077] ;

[0078] in, Represents the weighting coefficient of all marked points in the impact area; , Respectively represent the coordinate positions of all marked points in the impact area; 、 That is the impact position coordinate of the impact event.

[0079] Through the above DTW distance The impact position coordinates calculated after weighting have high positioning accuracy and can accurately locate the impact position.

[0080] Energy detection steps, see Figure 5 , according to the impact signal S received by sensors in all areas impact Obtain energy characterization value Calculate the energy of the impact according to the following formula :

[0081] .

[0082] Where, and The acquisition can be obtained through the above sample acquisition step, that is: pre-fixed energy Impact the surface of the composite material structure and obtain the fixed energy according to the impact signal received by several sensors Characteristic value under impact , specifically obtained by the following formula at fixed energy Characteristic value under impact :

[0083] ;

[0084] ;

[0085] in, is the expression of the shock signal obtained by the sensor, is the energy value of the corresponding impact signal; is the number of sensors deployed on the composite material structure to be detected, For the The energy value of the impulse response signal corresponding to each sensor; According to The formula is used to calculate it.

[0086] In the formula, the energy representation value when the impact event occurs is obtained by the following formula: :

[0087] ;

[0088] in, When the shock event occurs The energy value of the impulse response signal corresponding to each sensor.

[0089] Finally, 、 、 Substitute the above information about the impact energy The formula can effectively estimate the impact energy of the impact area.

[0090] In order to generalize the above method to other impact areas, the present invention also provides a regional compensation correction method for impact energy, see Figure 6 , specifically including the energy correction coefficient acquisition step, after conducting a test of impacting the surface of the composite material structure with a fixed energy E0, selecting one area as a reference area, and obtaining the energy characterization value of the reference area , calculate the energy correction coefficient of each area relative to the reference area , the formula is as follows:

[0091] ;

[0092] ;

[0093] in, is the number of sensors in the reference area, In the reference area The energy value of the impulse response signal corresponding to each sensor.

[0094] Energy correction step, based on the impact area determined by coarse positioning, according to the energy correction coefficient Obtain the energy correction coefficient of the impact area relative to the reference area And the corrected impact energy value is obtained by the following formula :

[0095] .

[0096] The energy detection method of this embodiment uses an impact response signal to characterize the impact energy and makes corrections based on regional characteristics to obtain an accurate impact energy value. Compared with traditional methods, this method does not require the construction of a complex system response identification model. It has the advantages of a simple method, efficient impact energy inversion, and accurate energy detection.

[0097] The present invention also provides a convolutional neural network-based impact location and energy detection system, see Figure 7 , comprising a region processing module, for disposing m sensors for receiving external impact signals on the composite material structure to be detected, and dividing the surface of the composite material structure to be detected into at least M regions;

[0098] The sample acquisition module is used to perform impact tests on a number of training points and marking points in each area to obtain a first sample database and a second sample database containing impact signals; and Impact the surface of the composite material structure and obtain the fixed energy according to the impact signal received by several sensors Characteristic value under impact ;

[0099] A model training module is used to construct a 1D-CNN neural network with an input of (l, m) and an output of (M, 1), where l is the signal length of each sensor; the first sample database is divided into a training set and a test set, and the sets are input into the 1D-CNN neural network for training and verification to obtain an impact event monitoring model;

[0100] Coarse positioning module, if a certain impact event occurs, some of the sensors will receive the impact signal S impact ; Used to convert the impact signal S impact Input into the impact event monitoring model, and then determine the impact area of ​​the impact event based on the output results to complete the coarse positioning;

[0101] The fine positioning module is used to calculate the impact signal S according to the DTW algorithm based on the impact area determined by the coarse positioning. impact The DTW distance L of all the markers in the impact area of ​​the second sample database i ; The obtained L i The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm to achieve precise positioning;

[0102] Energy detection module, used to detect the impact signal S received by sensors in all areas impact Obtain energy characterization value Calculate the energy of the impact according to the following formula :

[0103] .

[0104] Preferably, an energy correction module may also be included for selecting one of the regions as a reference region after a test of impacting the composite material structure surface with a fixed energy E0, and obtaining the energy characterization value of the reference region. , calculate the energy correction coefficient of each area relative to the reference area ;

[0105] Then based on the impact area determined by coarse positioning, according to the energy correction coefficient Obtain the energy correction coefficient of the impact area relative to the reference area And the corrected impact energy value is obtained by the following formula :

[0106] .

[0107] In order to better illustrate the impact location and energy detection method based on convolutional neural network provided by the present invention, the implementation process of the method of the present invention is described in detail below using a specific composite material reinforced plate as an example:

[0108] (1) The monitoring area (360 mm × 360 mm) of a composite reinforced plate with a size of 480 mm × 480 mm is divided into 16 rectangular areas of equal area, such as Figure 8 As shown, they are numbered with letters A to P; piezoelectric sensors are arranged at the four boundary intersections of the monitoring area.

[0109] (2) In each region, 10 training points were selected to conduct 10 random shock tests and the region number was used as the signal label to construct the first sample database containing 16×10 groups of shock signals.

[0110] A rectangular coordinate system is constructed in each area, and 5 marking points are selected for impact testing to construct a second sample database for precise positioning, such as Figure 10 As shown, 5 marking points evenly distributed in the area are selected, namely NO.1, NO.2, NO.3, NO.4, and NO.5.

[0111] (3) Construct a 1D-CNN neural network with an input of (1600, 4) and an output of (16, 1), where 1600 is the signal length of each sensor, 4 is the number of sensors, and 16 is the number of areas divided by the monitored structure. The samples in the first sample database are divided into a training set and a test set according to a ratio of 7:3, which are used for the model generation training and model training effect verification process of the constructed 1D-CNN network respectively. The model training results are shown in the figure below. Figure 9 As shown in the figure, it can be seen that as the training cycle increases, the loss function of the model gradually decreases and the accuracy gradually increases. The model converges after about 35 training cycles, and the prediction accuracy reaches 100%.

[0112] (3) Randomly conduct 5 impact tests in each area to verify the reliability of the impact area identification. That is, according to the coarse positioning step provided by the present invention, the impact area is identified using the trained impact event monitoring model. The coarse positioning results are as follows: Figure 11 As shown in the figure, the horizontal axis represents the actual impact area and the vertical axis represents the identified impact position. Figure 11It can be seen that the method proposed in the present invention can effectively identify the impact area with an accuracy rate of up to 98.75% (79 / 80). The only recognition error was that the impact in area B was identified as area F.

[0113] (4) According to the precise positioning step provided by the present invention, the DTW algorithm and the weighted centroid positioning algorithm are used to identify the impact position of the impact area. The precise positioning result is as follows: Figure 12 As shown, from the distribution of positioning results on the structure, the impact position identified by the centroid weighted positioning algorithm is very close to the actual position, which further proves the effectiveness and accuracy of the positioning method provided by the present invention.

[0114] (5) If Figure 8 As shown, taking area G as a reference, a rubber ball with a height of 40mm is subjected to free fall impact in each area, and the correction coefficient of each area is calculated. Then, a rubber ball with a height of 80mm is subjected to free fall impact in area G, and the impact energy is tested and corrected according to the energy detection steps provided by the invention. The energy detection results are shown as follows: Figure 13 As shown, according to Figure 13 It can be seen that before correction and compensation, the average error of impact energy estimation was 61.46%, and the estimation error of individual areas such as A, N, I, and P even exceeded 60%. After correction and compensation, the overall average error was only 8.95%. Although the average error after compensation in individual areas was slightly greater than the average error before compensation, the overall energy estimation method after error compensation is more stable and can continuously output highly reliable results. Therefore, the energy detection step provided by the present invention has the advantages of high stability and lower average error, which is more meaningful for practical application scenarios.

[0115] It should also be noted that the method and system provided by this invention are applicable not only to composite flat panels but also to complex structures with dense ribs, such as aircraft fuselage panels and wing skins. When applied to aircraft structural maintenance, repair work can be confined to a limited area. Furthermore, energy estimation can provide a priori knowledge of composite material damage, thereby reducing maintenance costs and improving repair efficiency.

[0116] In summary, compared with existing technologies, the convolutional neural network-based impact location and energy detection method and system provided by the present invention offer accurate positioning, high impact inversion efficiency, and simple operation. It not only avoids the impact of structural complexity on positioning accuracy but also effectively controls impact energy estimation errors. While reducing maintenance costs, it also effectively improves positioning and detection efficiency, promising broad application prospects.

[0117] In addition, those skilled in the art should understand that, although there are many problems in the prior art, each embodiment or technical solution of the present invention may be improved in only one or several aspects, without having to simultaneously solve all the technical problems listed in the prior art or background art. Those skilled in the art should understand that any content not mentioned in a claim should not be construed as limiting the claim.

[0118] Although this document frequently uses terms such as sensor, training point, marker point, first sample database, second sample database, impact signal, DTW algorithm, 1D-CNN neural network, weighted centroid positioning algorithm, etc., it does not exclude the possibility of using other terms. These terms are used only to more conveniently describe and explain the essence of the present invention; interpreting them as any additional restrictions is contrary to the spirit of the present invention; the terms "first", "second", etc. (if any) in the description and claims of the embodiments of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A convolutional neural network-based impact location and energy detection method, characterized in that: The following steps are involved: an area processing step of disposing m sensors for receiving external impact signals on the composite material structure to be inspected, and dividing the surface of the composite material structure to be inspected into at least M areas; The sample acquisition step is to perform impact tests on a plurality of training points and marking points in each of the areas to obtain a first sample database and a second sample database containing impact signals; and Impact the surface of the composite material structure and obtain the fixed energy according to the impact signal received by several sensors Characteristic value under impact ; In the model training step, a 1D-CNN neural network is constructed with an input of (l, m) and an output of (M, 1), where l is the signal length of each sensor. The first sample database is divided into a training set and a test set, and the sets are input into the 1D-CNN neural network for training and verification to obtain a shock event monitoring model. In the rough positioning step, if an impact event occurs, several sensors will receive the impact signal S impact ; The impact signal S impact Input into the impact event monitoring model, and then determine the impact area of ​​the impact event based on the output results to complete the coarse positioning; In the fine positioning step, based on the impact area determined by coarse positioning, the impact signal S is calculated according to the DTW algorithm. impact DTW distance to all markers in the impact area of ​​the second sample database ; The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm to achieve precise positioning; Energy detection step, based on the impact signal S received by sensors in all areas impact Obtain energy characterization value Calculate the impact energy according to the following formula : The step of obtaining an energy correction coefficient is also included. After performing a test on the surface of the composite material structure with a fixed energy E0, one of the areas is selected as a reference area to obtain the energy characterization value of the reference area. , calculate the energy correction coefficient of each area relative to the reference area , the formula is as follows: ; ； in, is the number of sensors in the reference area, In the reference area The energy value of the impact signal corresponding to each sensor.

2. The impact location and energy detection method based on convolutional neural network according to claim 1, characterized in that: The region processing step further includes numbering each region to obtain a signal label; and in the sample acquisition step, the impact signal in each region is mapped into a corresponding signal label for storage.

3. The impact location and energy detection method based on convolutional neural network according to claim 2, characterized in that: In the sample acquisition step, the sample acquisition step of the first sample database is as follows: N (N≥1) training points are selected in each of the regions and impact tests are performed to obtain corresponding impact signals, and then the impact signals at each training point in each of the regions are mapped into corresponding signal labels and stored in the first sample database; The sample acquisition step of the second sample database is: selecting B marking points in each area and performing impact tests respectively to obtain corresponding impact signals, and then storing the impact signals obtained from the impact tests of each marking point in each area as samples.

4. The impact location and energy detection method based on convolutional neural network according to claim 1, characterized in that: In the model training step, the samples in the first sample database are divided into a training set and a test set in a ratio of 7:

3.

5. The impact location and energy detection method based on convolutional neural network according to claim 3, characterized in that: In the fine positioning step, the obtained The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm, specifically including the following formula: , ； ， ； in, Represents the weighting coefficient of all marked points in the impact area; , Respectively represent the coordinate positions of all marked points in the impact area; 、 is the impact position coordinate of the impact event.

6. The impact location and energy detection method based on convolutional neural network according to claim 1, characterized in that: In the sample acquisition step, the following formula is used to obtain the fixed energy Characteristic value under impact : ； ； in, is the expression of the shock signal obtained by the sensor, is the energy value of the corresponding impact signal; is the number of sensors deployed on the composite material structure to be detected, For the The energy value of the impact signal corresponding to each sensor; in the energy detection step, the energy representation value when the impact event occurs is obtained by the following formula : ； in, When the shock event occurs The energy value of the impact signal corresponding to each sensor.

7. The impact location and energy detection method based on convolutional neural network according to claim 1, characterized in that: The energy correction step is also included, based on the impact area determined by the rough positioning, according to the energy correction coefficient Obtain the energy correction coefficient of the impact area relative to the reference area And the corrected impact energy value is obtained by the following formula : 。 8. An impact location and energy detection system based on convolutional neural networks, characterized by: include an area processing module, configured to arrange m sensors for receiving external impact signals on the composite material structure to be inspected, and to divide the surface of the composite material structure to be inspected into at least M areas; The sample acquisition module is used to perform impact tests on a number of training points and marking points in each area to obtain a first sample database and a second sample database containing impact signals; and Impact the surface of the composite material structure and obtain the fixed energy according to the impact signal received by several sensors Characteristic value under impact ; A model training module is used to construct a 1D-CNN neural network with an input of (l, m) and an output of (M, 1), where l is the signal length of each sensor; the first sample database is divided into a training set and a test set, and the sets are input into the 1D-CNN neural network for training and verification to obtain an impact event monitoring model; Coarse positioning module, if a certain impact event occurs, some of the sensors will receive the impact signal S impact ; Used to convert the impact signal S impact Input into the impact event monitoring model, and then determine the impact area of ​​the impact event based on the output results to complete the coarse positioning; The fine positioning module is used to calculate the impact signal S according to the DTW algorithm based on the impact area determined by the coarse positioning. impact The DTW distance L of all the markers in the impact area of ​​the second sample database i ; The obtained L i The impact position coordinates (x, y) within the impact area are calculated as the weight factor of the weighted centroid positioning algorithm to achieve precise positioning; Energy detection module, used to detect the impact signal S received by sensors in all areas impact Obtain energy characterization value Calculate the impact energy according to the following formula : ； It also includes an energy correction module, which can be used to fix the energy E After the test of impacting the composite material structure surface, one of the areas is selected as the reference area, and the energy characterization value of the reference area is obtained. , calculate the energy correction coefficient of each area relative to the reference area , the formula is as follows: ; ； in, is the number of sensors in the reference area, In the reference area The energy value of the impact signal corresponding to each sensor.

9. The impact location and energy detection device based on convolutional neural network according to claim 8, characterized in that: The energy correction module is also used to determine the impact area based on the rough positioning, according to the energy correction coefficient Obtain the energy correction coefficient of the impact area relative to the reference area And the corrected impact energy value is obtained by the following formula : 。

Images