Impact location method, readable storage medium and device based on convolutional neural network and centroid weighting

Abstract

The present invention provides a shock location method based on convolutional neural network and centroid weighting, which includes setting a sensor, dividing a monitored structure into M areas and numbering them, performing N shocks in each divided area, building a sample database, obtaining a trained convolutional neural network model based on the sample database, using the model for shock event monitoring, and when a shock event is detected, using the signal collected by the sensor as the input of the model, establishing a rectangular coordinate system in the area where the shock event occurs, setting a reference mark point in the area, and calculating the DTW distance L between the shock signal and the reference mark point in the area where the shock event is located. i , L i The reciprocal of is taken as the weighting coefficient W i Substitute the centroid positioning formula to locate the impact position in the impact area. This positioning method has the advantages of high efficiency, little influence of structural characteristics on the positioning results, and no need for high-density sensor deployment.

Description

Technical Field

[0001] The present invention relates to the technical field of impact positioning of complex composite material structures of aircraft, and in particular to an impact positioning method based on convolutional neural network and centroid weighting, a readable storage medium and a device. Background Art

[0002] Composite materials have the advantages of high specific strength, large specific modulus, and designable structure. They have obvious advantages in lightweight manufacturing, economical operation and maintenance, and advanced structural design. Therefore, they are widely used in the aerospace field. However, composite materials will inevitably be hit by foreign objects such as bird strikes and flying rocks during service, which will cause various almost invisible damages inside the composite materials, resulting in the degradation of the load-bearing performance of the structure, which seriously threatens the in-service safety of aerospace equipment.

[0003] Conventional nondestructive testing technologies currently used for impact damage detection include ultrasonic scanning, eddy current, thermal imaging, etc. These methods usually require point-by-point and area-by-area scanning and measurement of the structure, which is time-consuming and economically expensive for large structures and difficult to meet the needs of airborne online monitoring.

[0004] Some people have proposed to monitor impact events in real time through a sensor network integrated on the composite structure, and to locate the impact based on the impact response signal, thereby limiting subsequent inspection and maintenance work to a limited local area, which has a positive effect on the in-service safety and operational maintenance of the aircraft structure.

[0005] Most of the existing impact location methods for large and complex composite structures are based on high-density sensor networks for impact monitoring and impact location through corresponding algorithms. Among them, the location method based on stress wave propagation velocity involves time domain feature extraction of stress waves and velocity compensation in different propagation directions. It has a large workload and its accuracy is greatly affected by the structure. The location method based on the system model involves solving the transfer function through the impact response and building the system model. It requires a lot of mathematical calculations and parameter adjustments to make the model converge, which is inefficient. Summary of the invention

[0006] In order to solve the problems mentioned in the above background technology in the existing impact positioning methods for complex composite structures, the positioning results are greatly affected by the complex structure, the system modeling workload is large, and the positioning efficiency is low. The present invention provides an impact positioning method based on convolutional neural network and centroid weighting. Compared with the existing technology, this method can give full play to the advantages of 1D-CNN (one-dimensional convolutional neural network) in solving structural linear responses, and realize the adaptive construction of the relationship between complex structure input and response output without excessive manual intervention. It has the advantages of high efficiency, little influence of structural characteristics on positioning results, and no need for high-density sensor deployment.

[0007] The present invention provides a shock location method based on convolutional neural network and centroid weighting, which comprises the following steps:

[0008] S100: a sensor for receiving an external impact signal is arranged on the monitored structure, the monitored structure is divided into M areas, and the M areas are numbered to obtain a number of each area;

[0009] S200: Perform N shocks in each divided area, where N is greater than or equal to 1 and is a positive integer, and store the number of each area as a label of the shock response signal to construct a sample database D1 containing M×N groups of signals;

[0010] S300: Divide the sample database D1 into a training set and a test set of the 1D-CNN neural network according to a preset ratio, perform learning and training, and obtain a convolutional neural network model that has been trained and verified by the test set;

[0011] S400: using the convolutional neural network model obtained in step S300 to monitor impact events of complex composite materials. When an impact event is detected, the signal collected by the sensor is used as the input of the convolutional neural network model. The output label number of the convolutional neural network model is the area where the impact event occurs.

[0012] S500: Establish a rectangular coordinate system in the area where the impact event occurs, and set a reference mark point B in the area i Calculate the DTW distance L between the impact signal and the reference marker in the area where the impact event is located i , the calculation formula of DTW distance is L i =DTW(S k ,S i ), where L i Represents the response signal S k With the response signal S i The DTW distance between k Represents the impact response signal, S i Indicates the Bth i The response signal of the reference mark point;

[0013] S600: The DTW distance L calculated in step S500 is i The reciprocal of is taken as the weighting coefficient W i , W i Substitute the center of mass positioning formula to locate the impact position (x, y) within the impact area. The center of mass positioning formula is as follows:

[0014]

[0015]

[0016] Among them, W i Indicates the reference mark point B in the impact zone i The weighting coefficient, x i and i They respectively represent the horizontal and vertical coordinate positions of the reference mark point Bi.

[0017] In some embodiments, the sensor is a piezoelectric sensor or an acoustic emission sensor.

[0018] In some embodiments, in step S100, the monitored structure is divided into M rectangular areas of equal area.

[0019] In some embodiments, in step S300, the preset ratio is 7:3.

[0020] In some embodiments, in step S300, the method for constructing a 1D-CNN neural network includes: constructing a 1D-CNN neural network with an input of (l, m) and an output of (M, 1), wherein l is the signal length of each sensor, m is the number of sensors, and M is the number of regions into which the monitored structure is divided.

[0021] In some embodiments, in step S500 , a rectangular coordinate system is established with the lower left corner of the area where the impact event occurs as the origin.

[0022] In some embodiments, in step S500, reference marking points B1, B2, B3, and B4 are set in the area, and the reference marking points B1, B2, B3, and B4 are respectively located at the four corners of the area.

[0023] In some embodiments, in step S500, a reference mark point B is set in the region i After that, a collision test is performed on each reference mark point and the signal of each reference mark point is stored.

[0024] The present invention also provides a computer-readable storage medium, on which is stored a program for implementing a shock localization method based on a convolutional neural network and centroid weighting, and the program for implementing a shock localization method based on a convolutional neural network and centroid weighting is executed by a processor to implement the steps of the shock localization method based on a convolutional neural network and centroid weighting as described in any of the above embodiments.

[0025] The present invention also provides a device, which includes: a memory for storing an embedded software program; a processor for executing the embedded software program stored in the memory, and when the embedded program is executed, the steps of the impact positioning method based on convolutional neural network and centroid weighting described in any of the above embodiments are implemented.

[0026] The present invention provides an impact positioning method, a readable storage medium and a device based on convolutional neural network and centroid weighting, which can give full play to the advantages of 1D-CNN (one-dimensional convolutional neural network) in solving structural linear response, realize the adaptive construction of the relationship between complex structural input and response output, and do not require excessive manual intervention. It has the advantages of high efficiency, little influence of structural characteristics on positioning results, and no need for high-density sensor deployment.

[0027] Other features and beneficial effects of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other beneficial effects of the present invention can be achieved and obtained by the structures specifically pointed out in the description, claims, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces 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 creative work. The positional relationships described in the drawings in the following description are based on the directions in which the components are drawn in the diagrams, unless otherwise specified.

[0029] Figure 1 It is a flow chart of the impact location method based on convolutional neural network and centroid weighting of the present invention;

[0030] Figure 2 This is a schematic diagram of the rough positioning process based on the 1D-CNN neural network;

[0031] Figure 3 This is a schematic diagram of the precise positioning process based on centroid weighting;

[0032] Figure 4 It is a schematic diagram of the marking positions within the impact area;

[0033] Figure 5 It is a schematic diagram of the DTW calculation process;

[0034] Figure 6 It is a schematic diagram of the impact verification point selection;

[0035] Figure 7 is a schematic diagram of the rough positioning and recognition results;

[0036] Figure 8 It is a schematic diagram of precise positioning and identification results;

[0037] Fig. 9It is a schematic diagram of the composite material tank and sensor layout;

[0038] Fig.10 It is a schematic diagram of the impact recognition accuracy of setting a sensor;

[0039] Fig.11 It is a schematic diagram of the impact recognition accuracy when two sensors are set;

[0040] Fig.12 It is a schematic diagram of the impact recognition accuracy when four sensors are set;

[0041] Fig.13 It is a schematic diagram of the impact recognition accuracy when six sensors are set up. DETAILED DESCRIPTION

[0042] 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, rather than all 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.

[0043] In the description of the present invention, it should be understood that the terms "center", "lateral", "up", "down", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more. In addition, the term "including" and any variation thereof all mean "at least including".

[0044] See also Figure 1 , Figure 1: is a flow chart of the impact location method based on convolutional neural network and centroid weighting of the present invention. In order to achieve at least one of the above advantages or other advantages, an embodiment of the present invention provides an impact location method based on convolutional neural network and centroid weighting. The impact location method based on convolutional neural network and centroid weighting first roughly locates the impact area based on the convolutional neural network model, and then accurately locates the impact point in the roughly located impact area by centroid weighting. As shown in the figure, the impact location method may include the following steps:

[0045] S100: a sensor for receiving an external impact signal is arranged on the monitored structure, the monitored structure is divided into M areas, and the M areas are numbered to obtain a number of each area;

[0046] S200: Perform N shocks in each divided area, where N is greater than or equal to 1 and is a positive integer, and store the number of each area as a label of the shock response signal to construct a sample database D1 containing M×N groups of signals;

[0047] S300: Divide the sample database D1 into a training set and a test set of the 1D-CNN neural network according to a preset ratio, perform learning and training, and obtain a convolutional neural network model that has been trained and verified by the test set;

[0048] S400: using the convolutional neural network model obtained in step S300 to monitor impact events of complex composite materials. When an impact event is detected, the signal collected by the sensor is used as the input of the convolutional neural network model. The output label number of the convolutional neural network model is the area where the impact event occurs.

[0049] S500: Establish a rectangular coordinate system in the area where the impact event occurs, and set a reference mark point B in the area i Calculate the DTW distance L between the impact signal and the reference marker in the area where the impact event is located i , the calculation formula of DTW distance is L i =DTW(S k ,S i ), where L i Represents the response signal S k With the response signal S i The DTW distance between k Represents the impact response signal, S i Indicates the Bth i The response signal of the reference mark point is set in the area. i After that, each reference mark point B iPerform a collision test and store the signals of each reference mark point to calculate the collision signal in the impact event and the reference mark point B in the impact area. i DTW distance between the signals;

[0050] S600: The DTW distance L calculated in step S500 is i The reciprocal of is taken as the weighting coefficient W i , W i Substitute the center of mass positioning formula to locate the impact position (x, y) within the impact area, where the center of mass positioning formula is as follows:

[0051]

[0052] Among them, W i Indicates the reference mark point B in the impact zone i The weighting coefficient, x i and i They respectively represent the horizontal and vertical coordinate positions of the reference mark point Bi.

[0053] In one embodiment, if Figure 2 As shown, a rough positioning process is provided:

[0054] First, a sensor for receiving external impact signals is arranged on the monitored structure (the sensor may be, for example, a piezoelectric sensor, an acoustic emission sensor, etc.), and then the monitored structure is divided into M rectangular areas of equal area and these rectangular areas are numbered to obtain the number of each area.

[0055] Secondly, training points are selected in each divided area to perform N (N≥1) shocks, and the area number is stored as the label of the shock response signal, and finally a sample database D1 containing M×N groups of signals is constructed (as shown in Table 1 below).

[0056] Table 1

[0057]

[0058] Next, a 1D-CNN neural network with input (l, m) and output (M, 1) is constructed, where l is the signal length of each sensor, m is the number of sensors, and M is the number of areas divided by the monitored structure. The sample database D1 is divided into a training set and a test set according to a 7:3 (preset ratio), which are used for model generation training and model training effect verification of the constructed 1D-CNN neural network respectively. The impact area positioning based on 1D-CNN is to use 1D-CNN to process the time series composed of each sensor signal in the impact event, classify and identify the time series of the impact response signal through 1D-CNN, and realize the impact area positioning by combining the divided impact area.

[0059] Finally, the trained and test set verified convolutional neural network model is applied to the impact event monitoring of complex composite structures. When an impact event occurs, the signal collected by the sensor network is used as the input of the convolutional neural network model, and the label number output by the convolutional neural network model is the area where the impact event occurs.

[0060] Based on the rough positioning of the impact event area, the centroid weighted positioning algorithm is used to obtain the precise impact position in the impact identification area. The centroid weighted algorithm is based on the centroid positioning algorithm. The DTW (Dynamic Time Wrapping) method is used to calculate the DTW distance (dynamic time warping distance) between the impact signal and the reference point inside the impact area to perform centroid weighting. Figure 3 As shown, a precise positioning process is provided:

[0061] First, if Figure 4 As shown, a rectangular coordinate system is established with the lower left corner of each area as the origin, and B1 to B4 are selected inside the area as reference marking points for impact testing. Preferably, the reference marking points B1, B2, B3, and B4 are respectively located at the four corners of the rectangular area, and a sample database D2 for precise positioning is constructed (as shown in Table 2 below).

[0062] Table 2

[0063]

[0064] Secondly, when an impact event occurs, the 1D-CNN network is used to perform rough positioning, identify the area where the impact event is located, and calculate the DTW distance L between the impact signal and the reference markers B1 to B4 in the area where the impact event is located. i Each test mark point B1~B4 will first perform a collision test and store the signal of the mark point. The calculation process of DTW distance is as follows Figure 5 As shown, its purpose is to find the time series from p1 to p under the premise of aligning the two time series. kComposition path c p (X, Y) makes the path distance and L calculated according to the DTW path mapping relationship minimum. Figure 4 Take the k collision verification point as an example, the DTW distance L between the collision signal of this point and the selected reference mark point i for:

[0065] L i =DTW(S k ,S i ),i=1..4

[0066] Among them, L i The response signal S generated by the k-point impact k With reference mark point B i The response signal S i The DTW distance between k represents the response signal of k-point impact; S i Indicates the Bth i A response signal marking a reference point.

[0067] Finally, the calculated DTW distance value L i The reciprocal of is taken as the weighting coefficient W i Substitute the center of mass positioning formula to calculate the impact position (x, y) within the impact area to achieve accurate positioning of the impact event.

[0068] The centroid positioning formula is as follows:

[0069]

[0070] Among them, W i represents the weighting coefficient corresponding to the reference mark points B1 to B4 in the impact zone, x i and i They represent the horizontal and vertical coordinates of the reference marking points B1 to B4 respectively, and (x, y) is the position coordinates of the collision event.

[0071] The following will further illustrate this with reference to a positioning case, so as to more intuitively understand the impact positioning method based on convolutional neural network and centroid weighting of the present invention.

[0072] Firstly, the monitoring area (360mm×360mm) of a composite reinforced plate with a size of 480mm×480mm is divided into 16 rectangular areas of equal area, and the 16 areas are numbered with letters A to P. Piezoelectric sensors are arranged at the four boundary intersections of the monitoring area, and then 10 training points are selected in each area for 10 random impact tests. The area number is used as the signal label, and a sample database D1 containing 16×10 groups of impact signals is constructed.

[0073] Secondly, a 1D-CNN neural network with input (1600,4) and output (16,1) is constructed, where 1600 is the signal length of each sensor, 4 is the number of sensors, and 16 is the number of regions divided by the monitored structure. The functional architecture of each layer of the 1D-CNN network is as follows:

[0074] Input layer: Each data represents the response signal of four sensors in an impact event, that is, 4 time series with a length of 1600, and the input vector size is (1600, 4). For the convenience of data processing, the input data is flattened into a vector of length 6400, and then restored to its original size (1600, 4) in the first layer of the network.

[0075] The first and second convolutional layers: A convolution kernel can only learn one feature. Here, 16 convolution kernels of size 8 are defined respectively, so each convolutional layer can obtain 16 different features in advance.

[0076] The first maximum pooling layer: In order to reduce the complexity of the output data and reduce the overfitting phenomenon of the model, a pooling layer is added after the convolution layer. Here, a maximum pooling layer of size 2 is selected, which means that the output matrix size after this layer is only half of its input.

[0077] The remaining convolutional layers are pooling layers: In order to learn higher-level features, 6 convolutional layers and corresponding maximum pooling layers are added here.

[0078] Average pooling layer: An average pooling layer is added here to further avoid overfitting. Unlike the maximum pooling layer, the average pooling layer takes the mean of the weights.

[0079] Dropout layer: The Dropout layer randomly assigns zero weights to neurons in the network. Here, a ratio of 0.5 is selected, which means that 50% of the neurons will have zero weights. This can reduce the sensitivity of the network to small changes in input data and improve the robustness of the model.

[0080] Fully connected layer: The last layer will transform the output into a vector of length 16, corresponding to 16 label regions. Softmax is used as the activation function, which makes the sum of the final 16 vector output values ​​​​corresponding to 1, so that the 16 values ​​​​correspond to the probability of the predicted object being the corresponding category.

[0081] Again, the database D1 is divided into a training set and a test set according to a ratio of 7:3, which are used for model generation training of the built 1D-CNN network and model training effect verification process respectively. 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%. Then, 5 collisions are randomly carried out in each area, and the trained network is used to identify the area where the collision event is located to verify the accuracy of the model's coarse positioning.

[0082] Finally, a rectangular coordinate system is constructed in each divided area, and reference markers are selected for collision to construct a sample database D2 for precise positioning; Figure 6 As shown in Figure 1, select positions No. 1 to No. 5 in the divided area for impact test and perform positioning according to the above method. Finally, the rough positioning result is as follows: Figure 7 As shown in the figure, the horizontal axis represents the actual impact area, and the vertical axis represents the identified impact position. Figure 7 It can be seen that this method can effectively identify the impact area with an accuracy of 98.75% (79 / 80). The only recognition error was that the impact in area B was identified as area F. Figure 8 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 proves the effectiveness of the positioning method provided by the present invention.

[0083] It should be noted that the rough positioning method provided by the present invention is applicable not only to the above-mentioned composite stiffened panels, but also to impact positioning of complex curved surface structures such as aircraft fuselage panels and composite tanks.

[0084] like Fig. 9 As shown, the cylinder section of the tank is divided into a rectangular area of ​​8 rows and 5 columns, and 6 piezoelectric sensors S1 to S6 are arranged. Ten random impacts are carried out in all areas to collect samples to build a sample database for training the 1D-CNN network, and then five random impact tests are carried out in each area and the trained 1D-CNN network is used to verify the impact positioning. In order to illustrate the superiority of the method provided by the present invention in sensor layout, the positioning effects of the networks trained with different sensor data are given below.

[0085] See also Figures 10 to 13 , Figures 10 to 13The following are schematic diagrams of the impact recognition accuracy when 1, 2, 4, and 6 sensors are set respectively. The data used by the single sensor is sensor S1, the dual sensor uses S1 and S5, the quad sensor uses S1, S3, S4, and S6, and the six sensors are S1 to S6. As can be seen from the figure, the impact recognition accuracy of the method provided by the invention varies under different sensor layouts, but overall, except for the single sensor layout, the average recognition accuracy is above 90% under other layout schemes. In particular, the four-sensor layout has the highest accuracy of 96.5%. This is because compared with the single / dual sensor layout scheme, the four sensors can cover the monitored area more evenly and obtain more structural impact response signals. As for the six-sensor scheme, although the amount of perceived structural impact information is more, the input of too high a dimension means that the structure of the neural network model is more complex, resulting in limited generalization ability. Therefore, for a monitored structure, when designing the sensor layout scheme, the number of sensors should be reduced while the sensors cover all parts as much as possible according to the structural characteristics. A suitable sensor layout scheme can not only control the monitoring cost, but also enable the model to obtain better monitoring effects.

[0086] An embodiment of the present invention provides a computer-readable storage medium, on which is stored a program for implementing a shock location method based on a convolutional neural network and centroid weighting, wherein the program for implementing a shock location method based on a convolutional neural network and centroid weighting is executed by a processor to implement the steps of the shock location method based on a convolutional neural network and centroid weighting as described in any of the above embodiments.

[0087] An embodiment of the present invention provides a device, comprising a memory and a processor. The memory is used to store an embedded software program. The processor is used to execute the embedded software program stored in the memory, and when the embedded program is executed, the steps of the impact location method based on convolutional neural network and centroid weighting described in any of the above embodiments are implemented.

[0088] In summary, the present invention provides an impact positioning method, a readable storage medium and a device based on convolutional neural network and centroid weighting, which can give full play to the advantages of 1D-CNN (one-dimensional convolutional neural network) in solving structural linear response, realize the adaptive construction of the relationship between complex structural input and response output, without excessive manual intervention, and has the advantages of high efficiency, little influence of structural characteristics on positioning results, and no need for high-density sensor layout. In addition, the impact positioning method provided by the present invention is not only applicable to structures such as composite panels and metal plates, but also to complex structures such as aircraft fuselage panels and composite tanks.

[0089] 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 can be improved in only one or several aspects, without having to solve all the technical problems listed in the prior art or background technology at the same time. Those skilled in the art should understand that the content not mentioned in a claim should not be used as a limitation on the claim.

[0090] 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned 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 shock location method based on convolutional neural network and centroid weighting, characterized in that: The impact location method based on convolutional neural network and centroid weighting comprises the following steps: S100: a sensor for receiving an external impact signal is arranged on the monitored structure, the monitored structure is divided into M areas, and the M areas are numbered to obtain a number of each area; S200: Perform N shocks in each divided area, where N is greater than or equal to 1 and is a positive integer, and store the number of each area as a label of the shock response signal to construct a sample database D1 containing M×N groups of signals; S300: Divide the sample database D1 into a training set and a test set of the 1D-CNN neural network according to a preset ratio, perform learning and training, and obtain a convolutional neural network model that has been trained and verified by the test set; S400: Using the convolutional neural network model obtained in step S300 to monitor impact events of complex composite materials. When an impact event is detected, the signal collected by the sensor is input into the convolutional neural network model. The output label number of the convolutional neural network model is the area where the impact event occurs. S500: Establish a rectangular coordinate system in the area where the impact event occurs, and set a reference mark point B in the area i Calculate the DTW distance L between the impact signal and the reference marker in the area where the impact event is located i , the calculation formula of DTW distance is L i =DTW(S k ,S i ), where L i Represents the response signal S k With the response signal S i The DTW distance between k Represents the impact response signal, S i Indicates the Bth i The response signal of the reference mark point; S600: The DTW distance L calculated in step S500 is i The reciprocal of is taken as the weighting coefficient W i , W i Substitute the center of mass positioning formula to locate the impact position (x, y) within the impact area. The center of mass positioning formula is as follows: Among them, W i Indicates the reference mark point B in the impact zone i The weighting coefficient, x i and i Respectively represent the horizontal and vertical coordinate positions of the reference mark point Bi; Wherein, in step S300, the method for constructing a 1D-CNN neural network includes: 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 into which the monitored structure is divided; in step S500, a rectangular coordinate system is established with the lower left corner of the region where the impact event occurs as the origin; in step S500, reference marking points B1, B2, B3, and B4 are set in the region, and the reference marking points B1, B2, B3, and B4 are respectively located at the four corners of the region; When an impact event occurs, the 1D-CNN network is first used to perform rough positioning, identify the area where the impact event is located, and calculate the DTW distance L between the impact signal and the reference markers B1 to B4 in the area where the impact event is located. i Each test mark point B1~B4 will first perform a collision test and store the signal of the mark point. In the case of aligning the time series, find the k Composition path c p (X, Y) minimizes the path distance and L calculated according to the DTW path mapping relationship.

2. The impact location method based on convolutional neural network and centroid weighting according to claim 1, characterized in that: The sensor is a piezoelectric sensor or an acoustic emission sensor.

3. The impact location method based on convolutional neural network and centroid weighting according to claim 1, characterized in that: In step S100, the monitored structure is divided into M rectangular areas of equal area.

4. The impact location method based on convolutional neural network and centroid weighting according to claim 1, characterized in that: In step S300, the preset ratio is 7:

3.

5. The impact location method based on convolutional neural network and centroid weighting according to claim 1, characterized in that: In step S500, a reference mark point B is set in the area i After that, a collision test is performed on each reference mark point and the signal of each reference mark point is stored.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program for implementing a shock localization method based on a convolutional neural network and centroid weighting, and the program for implementing a shock localization method based on a convolutional neural network and centroid weighting is executed by a processor to implement the steps of the shock localization method based on a convolutional neural network and centroid weighting as described in any one of claims 1 to 5.

7. A device for impact location method based on convolutional neural network and centroid weighting, characterized in that: The device of the impact location method based on convolutional neural network and centroid weighting includes: A memory for storing embedded software programs; A processor is used to execute the embedded software program stored in the memory, and when the embedded software program is executed, the steps of the impact localization method based on convolutional neural network and centroid weighting described in any one of claims 1 to 5 are implemented.

Images