CN115326935A - Impact positioning method based on convolutional neural network and centroid weighting, readable storage medium and device - Google Patents
Impact positioning method based on convolutional neural network and centroid weighting, readable storage medium and device Download PDFInfo
- Publication number
- CN115326935A CN115326935A CN202211056971.9A CN202211056971A CN115326935A CN 115326935 A CN115326935 A CN 115326935A CN 202211056971 A CN202211056971 A CN 202211056971A CN 115326935 A CN115326935 A CN 115326935A
- Authority
- CN
- China
- Prior art keywords
- impact
- neural network
- convolutional neural
- area
- centroid
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 54
- 238000003860 storage Methods 0.000 title claims description 12
- 238000012544 monitoring process Methods 0.000 claims abstract description 11
- 230000004044 response Effects 0.000 claims description 31
- 239000002131 composite material Substances 0.000 claims description 18
- 238000012549 training Methods 0.000 claims description 16
- 230000004807 localization Effects 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 11
- 238000009863 impact test Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 description 10
- 238000011176 pooling Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 239000013598 vector Substances 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 239000003550 marker Substances 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000009659 non-destructive testing Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000035939 shock Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000001931 thermography Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/14—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object using acoustic emission techniques
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N19/00—Investigating materials by mechanical methods
- G01N19/08—Detecting presence of flaws or irregularities
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4481—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Acoustics & Sound (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Signal Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides an impact positioning method based on a convolutional neural network and centroid weighting, which comprises the steps of arranging a sensor, dividing a monitored structure into M regions and numbering the regions, carrying out N times of impacts in each divided region, constructing a sample database, obtaining a trained convolutional neural network model based on the sample database, using the model for impact event monitoring, taking a signal acquired by the sensor as the input of the model when an impact event is monitored, establishing a rectangular coordinate system in the region where the impact event occurs, and arranging a reference in the regionMarking point, calculating DTW distance L between the impact signal and reference marking point in the area of impact event i Is prepared by mixing L i Is taken as a weighting coefficient W i And substituting the center of mass positioning formula to position the impact position in the impact area. The positioning method has the advantages of high efficiency, small influence of structural characteristics on positioning results and no need of high-density sensing arrangement.
Description
Technical Field
The invention relates to the technical field of impact positioning of complex composite material structures of aircrafts, in particular to an impact positioning method based on a convolutional neural network and centroid weighting, a readable storage medium and equipment.
Background
The composite material structure has the advantages of high specific strength, large specific modulus, designable structure and the like, and has obvious advantages in the aspects of light-weight structure manufacturing, structure operation and maintenance economy, advanced structure design and the like. And thus are widely used in the aerospace field. However, the composite material structure is inevitably impacted by foreign matters such as birds and flying stones in the service process to form various almost invisible damages in the composite material, so that the bearing performance of the structure is degraded, and the in-service safety of the aerospace equipment is seriously threatened.
Conventional non-destructive testing techniques currently used for impact damage testing include ultrasonic scanning, eddy current, thermal imaging, and the like. These methods generally require point-by-point, area-by-area scanning measurements of the structure, are time and economic costly for large structures, and are difficult to meet onboard on-line monitoring requirements.
It has been proposed to monitor the impact event in real time via a sensor network integrated on the composite material structure and to perform impact localization based on the impact response signal, so as to limit subsequent overhaul 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.
The existing impact positioning method aiming at the large-scale complex composite material structure mostly carries out impact monitoring based on a high-density sensing network and realizes impact positioning through a corresponding fixed algorithm. The positioning method based on the stress wave propagation speed relates to time domain feature extraction of stress waves and speed compensation in different propagation directions, and has the advantages of large workload and large structural influence on precision. The positioning method based on the system model also relates to the processes of solving the transfer function and constructing the system model through the impulse response, and a large amount of mathematical calculation and parameter adjustment are needed to make the model convergent, so that the efficiency is low.
Disclosure of Invention
The method aims to solve the problems that the positioning result is greatly influenced by a complex structure, the system modeling workload is large and the positioning efficiency is low in the existing impact positioning method for the complex composite material structure in the background technology. Compared with the prior art, the impact positioning method based on the convolutional neural network and the centroid weighting can fully play the advantages of the 1D-CNN (one-dimensional convolutional neural network) in the aspect of solving the structural linear response, realizes the self-adaptive construction of the complex structural input and response output relation, does not need excessive manual intervention, and has the advantages of high efficiency, small influence of structural characteristics on the positioning result, and no need of high-density sensing arrangement.
The invention provides an impact positioning method based on a convolutional neural network and centroid weighting, which comprises the following steps:
s100: arranging a sensor for receiving an external impact signal on a monitored structure, dividing the monitored structure into M areas, and numbering the M areas to obtain the number of each area;
s200: performing N times of impact in each divided region, wherein N is greater than or equal to 1 and is a positive integer, and storing the number of each region as a label of an impact response signal to construct a sample database D1 containing M multiplied by N groups of signals;
s300: dividing the sample database D1 into a training set and a test set of the 1D-CNN neural network according to a preset proportion, and performing learning training to obtain a trained convolutional neural network model verified by the test set;
s400: the convolutional neural network model obtained in the step S300 is used for monitoring the impact event of the complex composite material structure, when the impact event is monitored, a signal acquired by a sensor is used as the input of the convolutional neural network model, and the output tag number of the convolutional neural network model is the area where the impact event occurs at this time;
s500: a rectangular coordinate system is established in the area where the impact event occurs, and a reference mark point B is arranged in the area i Calculating the DTW distance L between the impact signal and the reference mark point in the area where the impact event is located i The calculation formula of the DTW distance is L i ＝DTW(S k ,S i ) Wherein L is i Representing the response signal S k And a response signal S i DTW distance between, S k Response signal, S, indicating impact i Indicating the B-th in-impact region of recognition i A response signal of each reference mark point;
s600: the DTW distance L calculated in step S500 i As the weighting coefficient W i W is to be i Substituting the centroid location formula into the centroid location formula to locate the impact location (x, y) within the impact region, the centroid location formula is as follows:
wherein, W i The representation corresponds to the reference mark point B in the impact area i Weighting coefficient of (a), x i And y i Respectively indicate the abscissa and ordinate positions of the reference mark point Bi.
In some embodiments, the sensor is a piezoelectric sensor or an acoustic emission sensor.
In some embodiments, in step S100, the monitored structure is divided into M rectangular regions with equal area.
In some embodiments, in step S300, the preset ratio is 7.
In some embodiments, in step S300, the method for constructing the 1D-CNN neural network includes: and constructing a 1D-CNN neural network with input of (l, M) and output of (M, 1), wherein l is the signal length of each sensor, M is the number of the sensors, and M is the number of the areas divided by the monitored structure.
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.
In some embodiments, in step S500, the reference mark point B is set in the area 1 、B 2 、B 3 、B 4 Reference mark point B 1 、B 2 、B 3 、B 4 Located at the 4 corners of the area, respectively.
In some embodiments, in step S500, a reference marker point B is set within the area i And then, performing impact test on each reference mark point and storing the signal of each reference mark point.
The present invention also provides a computer-readable storage medium having stored thereon a program for implementing a convolutional neural network and centroid weighting based impulse positioning method, the program being executed by a processor to implement the steps of the convolutional neural network and centroid weighting based impulse positioning method as described in any of the above embodiments.
The present invention also provides an apparatus comprising: a memory for storing an embedded software program; a processor for executing the embedded software program stored in the memory, and the embedded program when executed implements the steps of the impact localization method based on convolutional neural network and centroid weighting described in any of the above embodiments.
The invention provides an impact positioning method based on a convolutional neural network and centroid weighting, a readable storage medium and equipment, which can fully play the advantages of a 1D-CNN (one-dimensional convolutional neural network) in the aspect of solving the structural linear response, realize the self-adaptive construction of the input and response output relationship of a complex structure, do not need excessive manual intervention, and have the advantages of high efficiency, small influence of the structural characteristics on the positioning result and no need of high-density sensing layout.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts; in the following description, the drawings are described with reference to the illustrated directions of components unless otherwise specified.
FIG. 1 is a schematic flow chart of the impact localization method based on convolutional neural network and centroid weighting according to the present invention;
FIG. 2 is a schematic diagram of a coarse positioning process based on a 1D-CNN neural network;
FIG. 3 is a schematic diagram of a centroid weighting based fine positioning process;
FIG. 4 is a schematic illustration of the location of a marker within the impact region;
FIG. 5 is a schematic diagram of a DTW calculation process;
FIG. 6 is a schematic view of bump verification point selection;
FIG. 7 is a schematic diagram of a coarse positioning recognition result;
FIG. 8 is a schematic diagram of the fine positioning recognition result;
FIG. 9 is a schematic view of a composite reservoir and sensor layout;
FIG. 10 is a schematic of impact recognition accuracy with one sensor set;
FIG. 11 is a schematic view of impact recognition accuracy with two sensors;
FIG. 12 is a graph illustrating impact recognition accuracy with four sensors;
fig. 13 is a diagram showing the impact recognition accuracy in which six sensors are provided.
Detailed Description
In order to make the objects, 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 with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments; the technical features devised in the different embodiments of the invention described below can be combined with each other as long as they do not conflict with each other; all other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it is to be understood that the terms "center", "lateral", "up", "down", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, indicate orientations and positional relationships based on those shown in the drawings, are only for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the device or component in question must have a particular orientation or be constructed and operated in a particular orientation, and therefore, should not be taken as limiting the present invention. Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless otherwise specified. In addition, the term "comprises" and any variations thereof mean "including at least".
Referring to fig. 1, fig. 1 is a schematic flow chart of an impulse positioning method based on a convolutional neural network and centroid weighting according to the present invention. To achieve at least one of the above advantages or other advantages, an embodiment of the present invention provides an impulse localization method based on a convolutional neural network and centroid weighting. The impact positioning method based on the convolutional neural network and the centroid weighting is characterized in that an impact area is roughly positioned in a mode based on a convolutional neural network model, and then an impact point is accurately positioned in the roughly positioned impact area in a centroid weighting mode. As shown in the figure, the impact localization method may include the steps of:
s100: arranging a sensor for receiving an external impact signal on a monitored structure, dividing the monitored structure into M areas, and numbering the M areas to obtain the number of each area;
s200: performing N times of impact in each divided region, wherein N is greater than or equal to 1 and is a positive integer, storing the number of each region as a label of an impact response signal, and constructing a sample database D1 containing M multiplied by N groups of signals;
s300: dividing a sample database D1 into a training set and a test set of the 1D-CNN neural network according to a preset proportion, and performing learning training to obtain a trained convolutional neural network model verified by the test set;
s400: the convolutional neural network model obtained in the step S300 is used for monitoring an impact event of a complex composite material structure, when the impact event is monitored, a signal acquired by a sensor is used as the input of the convolutional neural network model, and the output tag number of the convolutional neural network model is the area where the impact event occurs;
s500: a rectangular coordinate system is established in the area where the impact event occurs, and a reference mark point B is arranged in the area i Calculating the DTW distance L between the impact signal and the reference mark point in the area of the impact event i The calculation formula of the DTW distance is L i ＝DTW(S k ,S i ) Wherein L is i Indicating the response signal S k And a response signal S i DTW distance between, S k Response signal, S, indicating impact i Indicating the identified impact region within the B-th i A response signal of each reference mark point; wherein, a reference mark point B is arranged in the area i Then, each reference mark point B is needed i Performing impact test and storing the signals of the reference mark points so as to calculate the impact signal in the impact event and the reference mark point B in the impact area i Is sent toThe DTW distance between the signs;
s600: the DTW distance L calculated in step S500 i As the weighting coefficient W i A 1 is prepared from W i And substituting a centroid locating formula to locate the impact position (x, y) within the impact area, wherein the centroid locating formula is as follows:
wherein, W i The representation corresponds to the reference mark point B in the impact area i Weighting coefficient of (1), x i And y i Respectively indicate the abscissa and ordinate positions of the reference mark point Bi.
In one embodiment, as shown in fig. 2, a positioning process for coarse positioning is provided:
first, a sensor (for example, a piezoelectric sensor, an acoustic emission sensor, or the like) for receiving an external impact signal is arranged on a monitored structure, and then the monitored structure is divided into M rectangular regions having equal areas and the rectangular regions are numbered, thereby obtaining the number of each region.
Next, selecting a training point in each divided region to perform N (N ≧ 1) times of impact, storing the region number as a tag of an impact response signal, and finally constructing a sample database D1 (shown in table 1 below) containing M × N groups of signals.
TABLE 1
And then constructing a 1D-CNN neural network with input of (l, M) and output of (M, 1), wherein l is the signal length of each sensor, M is the number of the sensors, and M is the number of the areas divided by the monitored structure. And (4) dividing the sample database D1 into a training set and a test set according to the ratio of (7) (preset proportion), and respectively using the training set and the test set for the model generation training and the model training effect verification process of the built 1D-CNN neural network. The impact area positioning based on the 1D-CNN is to process a time sequence formed by each sensor signal in an impact event by using the 1D-CNN, classify and identify the time sequence of the impact response signal through the 1D-CNN, and realize the impact area positioning by combining with the divided impact area.
And finally, applying the trained convolutional neural network model verified by the test set to the monitoring of the impact event of the complex composite material structure, and when the impact event occurs, taking the signal acquired by the sensor network as the input of the convolutional neural network model, wherein the tag number output by the convolutional neural network model is the area where the impact event occurs.
And on the basis of roughly positioning the area where the impact event occurs, acquiring an accurate impact position in the impact identification area through a centroid weighting positioning algorithm. The centroid weighting algorithm is based on a centroid positioning algorithm, and calculates a DTW distance (Dynamic Time warping distance) between an impact signal and an internal reference point of an impact region by a DTW (Dynamic Time warping) method to perform centroid weighting. As shown in fig. 3, a fine positioning process is provided:
first, as shown in fig. 4, a rectangular coordinate system is established with the lower left corner of each region as the origin, and B is selected inside the region 1 ～B 4 The impact test is carried out as a reference mark point, preferably the reference mark point B 1 、B 2 、B 3 、B 4 Located at 4 corners of the rectangular area, respectively, a sample database D2 for fine positioning is constructed (as shown in table 2 below).
TABLE 2
Secondly, when an impact event occurs, firstly, coarse positioning is carried out based on the 1D-CNN network, the area where the impact event is located is identified, and the impact signal and the reference mark point B in the area where the impact event is located are calculated 1 ～B 4 DTW distance betweenL i . Each examination mark point B 1 ～B 4 The impact test is performed first and the signal of the marking point is stored. The DTW distance calculation process is shown in FIG. 5, and the purpose is to find the distance p from the two time sequences under the premise of aligning the two time sequences 1 ～p k Component path c p (X, Y) minimizes the sum of the calculated path distances L from the DTW path mapping relationship. Taking the k-hit verification point in FIG. 4 as an example, the point hit signal is a distance L from the DTW of the selected reference mark point i Comprises the following steps:
L i ＝DTW(S k ,S i ),i＝1..4
wherein L is i Representing the response signal S generated by a k-point impact k And a reference mark point B i In response to signal S i DTW distance therebetween; s k A response signal representing a k-point impact; s i Indicating the B-th in-impact region of recognition i A response signal marking the reference point.
Finally, the DTW distance value L obtained by calculation is used i As the weighting coefficient W i And substituting the calculated impact position (x, y) into a mass center positioning formula to realize the accurate positioning of the impact event.
The centroid location formula is as follows:
wherein, W i The representation corresponds to the reference mark point B in the impact area 1 ～B 4 Weighting coefficient of (a), x i And y i Respectively represent reference mark points B 1 ～B 4 Is the horizontal and vertical coordinate position of (x, y) is the position coordinate of the impact event.
The following description will be further described in conjunction with a localization case to facilitate a more intuitive understanding of the impulse localization method based on convolutional neural network and centroid weighting of the present invention.
Firstly, dividing a monitoring area (360 mm multiplied by 360 mm) of a composite material stiffened plate with the block size of 480mm multiplied by 480mm into 16 rectangular areas with equal areas, and numbering the 16 areas by letters A-P; piezoelectric sensors are arranged at four boundary intersection points of a monitoring area, 10 training points are selected in each area to perform 10 times of random impact tests, and area numbers are used as labels of signals, so that a sample database D1 containing 16 x 10 groups of impact signals is constructed.
Secondly, a 1D-CNN neural network with inputs of (1600, 4) and outputs of (16, 1) is constructed, wherein 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 functional architecture of each layer of the 1D-CNN network is as follows:
an input layer: each piece of data represents the response signals of four sensors in a shock event, i.e. 4 time series of length 1600, with an input vector of size (1600, 4). To facilitate data processing, the input data is flattened into vectors of length 6400, which are then restored to the original size at the first layer of the network (1600, 4).
First and second convolutional layers: only one characteristic can be learned by one convolution kernel, and 16 convolution kernels with the size of 8 are defined, so that each convolution layer can obtain 16 different characteristics in advance.
First largest pooling layer: in order to reduce the complexity of the output data and reduce the model overfitting phenomenon, a pooling layer is added after the convolutional layer, and a maximum pooling layer with the size of 2 is selected, so that the size of an output matrix passing through the layer is only half of the size of the input matrix.
The rest of the convolutional layers are the pooling layers: to learn higher level features, 6 convolutional layers and corresponding max pooling layers are added here.
Average pooling layer: an average pooling layer is added to avoid overfitting, and the average pooling layer is an average value of the weights, which is different from the maximum pooling layer.
Dropout layer: the Dropout layer randomly assigns zero weight to the neurons in the network. A ratio of 0.5 is chosen here, meaning that 50% of the neurons will be zero weight. By the method, the sensitivity of the network to small changes of input data can be reduced, and the robustness of the model can be improved.
Full connection layer: the last layer will change the output to a length-16 vector, corresponding to 16 label regions. Using Softmax as the activation function, it brings the final corresponding 16 vector output values to a sum of 1, so that the 16 values correspond to the probability that the predicted object is of the corresponding class, respectively.
And thirdly, dividing the database D1 into a training set and a testing set according to the 7. With the increase of the training period, the loss function of the model is gradually reduced, the accuracy rate is gradually increased, the model is converged after about 35 training periods, and the prediction accuracy rate reaches 100%. And then, randomly performing 5 times of impact in each area, using the trained network for identifying the area where the impact event is positioned, and verifying the accuracy of the coarse positioning of the model.
Finally, constructing a rectangular coordinate system in each divided area, selecting a reference mark point for collision, and constructing a sample database D2 for fine positioning; as shown in FIG. 6, the positions of NO.1 to NO.5 were selected in the divided regions for the impact test, and the positioning was performed according to the method described above. Finally, the coarse positioning results are shown in FIG. 7, with the abscissa representing the actual impact region and the ordinate representing the identified impact location. As can be seen from fig. 7, the method can achieve effective identification of the impact region, the accuracy rate is 98.75% (79/80), and the only identification error is to identify the impact of the B region as the F region. The precise positioning result is shown in fig. 8, and from the distribution of the positioning result on the structure, the impact position identified by the centroid weighting positioning algorithm is very close to the actual position, which proves the effectiveness of the positioning method provided by the invention.
It should be noted that the rough positioning method provided by the invention is applicable to impact positioning of complex curved surface structures such as aircraft fuselage wall plates, composite material storage tanks and the like, in addition to the composite material stiffened plate. The following description will be given taking a composite material storage tank as an example.
As shown in fig. 9, the barrel section of the reservoir is divided into rectangular regions of 8 rows and 5 columns, and 6 piezoelectric sensors S1 to S6 are arranged. And (3) acquiring samples by random impact for 10 times in all areas to construct a sample database for training the 1D-CNN network, then performing impact tests for 5 times randomly in each area, and performing impact positioning verification by using the trained 1D-CNN network. In order to illustrate the superiority of the method provided by the present invention in the sensor layout, the following respectively shows the positioning effect of the network trained by using different sensor data.
Referring to fig. 10 to 13, fig. 10 to 13 are schematic diagrams of impact recognition accuracy of sensors with 1, 2, 4 and 6 sensors respectively. Data used for single sensing is that of the sensor S1, data used for double sensing is that of the sensors S1 and S5, data used for four sensing is that of the sensors S1, S3, S4 and S6, and data used for six sensing is that of the sensors S1 to S6. It can be seen from the figure that the impact recognition accuracy rate of the method provided by the invention is different under different sensing layouts, but the average recognition accuracy rate is over 90% under other layout schemes except for a single sensing layout as a whole. Particularly, the highest accuracy rate of the four-sensor arrangement is 96.5%. This is because, compared to the single/dual sensing arrangement, the four sensors can cover the monitored area more uniformly, and obtain more structural impulse response signals. While the six-sensing scheme impacts more information quantity on the sensing structure, the input with too high dimensionality means that the neural network model structure is more complex, and the generalization capability of the neural network model is limited. Thus, for a structure to be monitored, the sensor arrangement should be designed to reduce the number of sensors while covering as many parts as possible according to the structural characteristics of the sensors. The appropriate sensor arrangement scheme can control the monitoring cost and enable the model to obtain a better monitoring effect.
An embodiment of the present invention provides a computer-readable storage medium having stored thereon a program for implementing a convolutional neural network and centroid weighting based impulse localization method, the program being executed by a processor to implement the steps of the convolutional neural network and centroid weighting based impulse localization method as described in any of the above embodiments
An embodiment of the invention provides an apparatus that includes a memory and a processor. The memory is used for storing the embedded software program. The processor is used for executing the embedded software program stored in the memory, and the embedded program is executed to realize the steps of the impact location method based on the convolutional neural network and the centroid weighting in any embodiment.
In summary, the invention provides an impact location method, a readable storage medium and a device based on a convolutional neural network and centroid weighting, which can fully exert the advantages of a 1D-CNN (one-dimensional convolutional neural network) in the aspect of solving structural linear response, realize self-adaptive construction of complex structural input and response output relations, do not need excessive manual intervention, and have the advantages of high efficiency, little influence of the location result on structural characteristics, and no need of high-density sensing layout. In addition, the impact positioning method provided by the invention is not only suitable for structures such as composite material plates and metal plates, but also suitable for complex structures such as aircraft fuselage wall plates and composite material storage boxes.
In addition, it will be appreciated by those skilled in the art that, although there may be many problems with the prior art, each embodiment or aspect of the present invention may be improved only in one or several respects, without necessarily simultaneously solving all the technical problems listed in the prior art or in the background. It will be understood by those skilled in the art that nothing in a claim should be taken as a limitation on that claim.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and these modifications or substitutions do not depart from the spirit of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. An impact positioning method based on a convolutional neural network and centroid weighting is characterized in that: the impact positioning method based on the convolutional neural network and the centroid weighting comprises the following steps:
s100: arranging a sensor for receiving an external impact signal on a monitored structure, dividing the monitored structure into M regions, and numbering the M regions to obtain the number of each region;
s200: performing N times of impact in each divided region, wherein N is greater than or equal to 1 and is a positive integer, and storing the number of each region as a label of an impact response signal to construct a sample database D1 containing M multiplied by N groups of signals;
s300: dividing the sample database D1 into a training set and a test set of the 1D-CNN neural network according to a preset proportion, and performing learning training to obtain a trained convolutional neural network model verified by the test set;
s400: the convolutional neural network model obtained in the step S300 is used for monitoring an impact event of a complex composite material structure, when the impact event is monitored, a signal acquired by a sensor is input into the convolutional neural network model, and an output tag number of the convolutional neural network model is an area where the impact event occurs;
s500: a rectangular coordinate system is established in the area where the impact event occurs, and a reference mark point B is arranged in the area i Calculating the DTW distance L between the impact signal and the reference mark point in the area where the impact event is located i The DTW distance is calculated by the formula L i ＝DTW(S k ,S i ) Wherein L is i Indicating the response signal S k And a response signal S i DTW distance between, S k Response signal, S, indicating impact i Indicating the B-th in-impact region of recognition i A response signal of each reference mark point;
s600: the DTW distance L calculated in step S500 i Is taken as a weighting coefficient W i W is to be i Substituting the centroid localization formula to locate the impact location (x, y) within the impact region, the centroid localization formula is as follows:
wherein, W i Indicating a point B corresponding to a reference mark in the impact area i Weighting coefficient of (a), x i And y i Respectively indicate the abscissa and ordinate positions of the reference mark point Bi.
2. The convolutional neural network and centroid weighting based impact localization method of claim 1, wherein: the sensor is a piezoelectric sensor or an acoustic emission sensor.
3. The convolutional neural network and centroid weighting based impulse positioning method of claim 1, wherein: in step S100, the monitored structure is divided into M rectangular regions with equal areas.
4. The convolutional neural network and centroid weighting based impulse positioning method of claim 1, wherein: in step S300, the preset ratio is 7.
5. The convolutional neural network and centroid weighting based impact localization method of claim 1, wherein: in step S300, the method for constructing the 1D-CNN neural network includes: and (3) constructing a 1D-CNN neural network with the input of (l, M) and the output of (M, 1), wherein l is the signal length of each sensor, M is the number of the sensors, and M is the number of the areas divided by the monitored structure.
6. The convolutional neural network and centroid weighting based impulse positioning method of claim 1, wherein: 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.
7. The convolutional neural network and centroid weighting based impulse positioning method of claim 1, wherein: in step S500, a reference mark point B is set in the area 1 、B 2 、B 3 、B 4 Reference mark point B 1 、B 2 、B 3 、B 4 Located at the 4 corners of the area, respectively.
8. The convolutional neural network and centroid weighting based impulse positioning method of claim 1, wherein: in step S500, a reference mark point B is set within the area i And then, performing impact test on each reference mark point and storing a signal of each reference mark point.
9. A computer-readable storage medium characterized by: the computer readable storage medium having stored thereon a program for implementing a convolutional neural network and centroid weighting based impulse location method, the program being executed by a processor to implement the steps of the convolutional neural network and centroid weighting based impulse location method as claimed in any one of claims 1 to 8.
10. An apparatus, characterized by: the apparatus comprises:
a memory for storing an embedded software program;
a processor for executing an embedded software program stored in the memory, and which when executed performs the steps of the convolutional neural network and centroid weighting based impulse positioning method of any of the above claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211056971.9A CN115326935A (en) | 2022-08-29 | 2022-08-29 | Impact positioning method based on convolutional neural network and centroid weighting, readable storage medium and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211056971.9A CN115326935A (en) | 2022-08-29 | 2022-08-29 | Impact positioning method based on convolutional neural network and centroid weighting, readable storage medium and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115326935A true CN115326935A (en) | 2022-11-11 |
Family
ID=83928572
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211056971.9A Pending CN115326935A (en) | 2022-08-29 | 2022-08-29 | Impact positioning method based on convolutional neural network and centroid weighting, readable storage medium and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115326935A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115790720A (en) * | 2022-11-30 | 2023-03-14 | 大连理工大学 | Health monitoring system and method for reusable aerospace low-temperature liquid oxygen composite material storage box structure |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108645498A (en) * | 2018-04-28 | 2018-10-12 | 南京航空航天大学 | Impact Location Method based on phase sensitivity light reflection and convolutional neural networks deep learning |
| WO2019238976A1 (en) * | 2018-06-15 | 2019-12-19 | Université de Liège | Image classification using neural networks |
| CN111721450A (en) * | 2020-06-22 | 2020-09-29 | 山东大学 | Impact positioning recognition device and method |
| CN113358028A (en) * | 2021-06-09 | 2021-09-07 | 长春工业大学 | Impact load positioning method and system based on fiber bragg grating sensor network |
| CN114061474A (en) * | 2021-11-15 | 2022-02-18 | 山东大学 | Low-speed impact positioning method for composite material structure of implanted fiber grating sensor |
| CN114189809A (en) * | 2021-11-15 | 2022-03-15 | 华东师范大学 | Indoor positioning method based on convolutional neural network and high-dimensional 5G observation characteristics |
-
2022
- 2022-08-29 CN CN202211056971.9A patent/CN115326935A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108645498A (en) * | 2018-04-28 | 2018-10-12 | 南京航空航天大学 | Impact Location Method based on phase sensitivity light reflection and convolutional neural networks deep learning |
| WO2019238976A1 (en) * | 2018-06-15 | 2019-12-19 | Université de Liège | Image classification using neural networks |
| CN111721450A (en) * | 2020-06-22 | 2020-09-29 | 山东大学 | Impact positioning recognition device and method |
| CN113358028A (en) * | 2021-06-09 | 2021-09-07 | 长春工业大学 | Impact load positioning method and system based on fiber bragg grating sensor network |
| CN114061474A (en) * | 2021-11-15 | 2022-02-18 | 山东大学 | Low-speed impact positioning method for composite material structure of implanted fiber grating sensor |
| CN114189809A (en) * | 2021-11-15 | 2022-03-15 | 华东师范大学 | Indoor positioning method based on convolutional neural network and high-dimensional 5G observation characteristics |
Non-Patent Citations (4)
| Title |
|---|
| 李毅;金守峰;尹加杰;仝梦园;陈阳;: "基于Kinect V2的筒子纱作业机器人视觉定位方法", 纺织高校基础科学学报, no. 03, 30 September 2020 (2020-09-30) * |
| 王奕惟 等: "基于全航段QAR数据和卷积神经网络的航空发动机状态辨识", 航空动力学报, vol. 36, no. 7, 31 July 2021 (2021-07-31) * |
| 陈亚宇 等: "基于深度学习与图像处理的废弃物分类与定位方法", 科学技术与工程, vol. 21, no. 21, 30 September 2021 (2021-09-30) * |
| 黄知超;梁国丽;朱芳来;: "微芯片焊点快速识别与定位方法", 半导体技术, no. 03, 3 March 2020 (2020-03-03) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115790720A (en) * | 2022-11-30 | 2023-03-14 | 大连理工大学 | Health monitoring system and method for reusable aerospace low-temperature liquid oxygen composite material storage box structure |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685152B (en) | Image target detection method based on DC-SPP-YOLO | |
| CN110660052B (en) | Hot-rolled strip steel surface defect detection method based on deep learning | |
| CN110276269B (en) | Remote sensing image target detection method based on attention mechanism | |
| CN116644608B (en) | Real sea area ship motion forecasting method and system based on marine environment data | |
| CN103776416B (en) | The indirect measurement method that a kind of antenna structure based on strain transducer is out of shape | |
| CN103353988B (en) | Allos SAR scene Feature Correspondence Algorithm performance estimating method | |
| CN113808174B (en) | Radar small target tracking method based on full convolution network and Kalman filtering | |
| CN115326935A (en) | Impact positioning method based on convolutional neural network and centroid weighting, readable storage medium and device | |
| CN108960421A (en) | The unmanned surface vehicle speed of a ship or plane online forecasting method based on BP neural network of improvement | |
| CN113838010B (en) | Intelligent composite material damage detection method based on infrared thermal wave and convolutional neural network | |
| CN104930980A (en) | Extraction device and method | |
| CN111340748A (en) | Battery defect identification method and device, computer equipment and storage medium | |
| CN116542912A (en) | Flexible body bridge vibration detection model with multi-target visual tracking function and application | |
| CN115795874A (en) | Dynamic explosive point calculation method based on target structure response | |
| Yanchen | Sonar image target detection and recognition based on convolution neural network | |
| CN116128883A (en) | Photovoltaic panel quantity counting method and device, electronic equipment and storage medium | |
| CN112084860A (en) | Target object detection method and device and thermal power plant detection method and device | |
| CN115410102A (en) | SAR image airplane target detection method based on combined attention mechanism | |
| CN110988872A (en) | Method for rapidly identifying health state of wall body detected by unmanned aerial vehicle-mounted through-wall radar | |
| CN108983187B (en) | Online radar target identification method based on EWC | |
| Tabian et al. | Impact Detection on Composite Plates Based on Convolution Neural Network | |
| CN117350096A (en) | Multi-type sensor layout optimization method for load performance evaluation under driving of particle swarm optimization algorithm | |
| CN115458088A (en) | Impact positioning and energy detection method and system based on convolutional neural network | |
| CN114879263B (en) | Three-dimensional magnetic gradient tensor inversion method and device based on convolutional neural network | |
| CN114580239B (en) | Bridge damage identification method considering uncertainty |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |