CN117347378A - Composite material damage identification method based on optical fiber measurement and neural network - Google Patents
Composite material damage identification method based on optical fiber measurement and neural network Download PDFInfo
- Publication number
- CN117347378A CN117347378A CN202311045468.8A CN202311045468A CN117347378A CN 117347378 A CN117347378 A CN 117347378A CN 202311045468 A CN202311045468 A CN 202311045468A CN 117347378 A CN117347378 A CN 117347378A
- Authority
- CN
- China
- Prior art keywords
- data
- damage
- strain
- optical fiber
- neural network
- 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.)
- Granted
Links
- 230000006378 damage Effects 0.000 title claims abstract description 136
- 239000013307 optical fiber Substances 0.000 title claims abstract description 75
- 239000002131 composite material Substances 0.000 title claims abstract description 49
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 39
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000005259 measurement Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000012545 processing Methods 0.000 claims abstract description 33
- 238000012544 monitoring process Methods 0.000 claims abstract description 19
- 238000004088 simulation Methods 0.000 claims abstract description 18
- 230000036541 health Effects 0.000 claims abstract description 13
- 230000009467 reduction Effects 0.000 claims abstract description 12
- 238000009826 distribution Methods 0.000 claims abstract description 11
- 238000003062 neural network model Methods 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 238000013075 data extraction Methods 0.000 claims abstract description 5
- 239000010410 layer Substances 0.000 claims description 18
- 229920000049 Carbon (fiber) Polymers 0.000 claims description 12
- 239000004917 carbon fiber Substances 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 239000004744 fabric Substances 0.000 claims description 7
- 238000005452 bending Methods 0.000 claims description 6
- 239000011229 interlayer Substances 0.000 claims description 6
- 238000007906 compression Methods 0.000 claims description 5
- 238000004519 manufacturing process Methods 0.000 claims description 5
- 238000003672 processing method Methods 0.000 claims description 5
- 238000005336 cracking Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 239000011347 resin Substances 0.000 claims description 3
- 229920005989 resin Polymers 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000003068 static effect Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 description 16
- 239000000463 material Substances 0.000 description 7
- 208000027418 Wounds and injury Diseases 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 208000014674 injury Diseases 0.000 description 6
- 239000000835 fiber Substances 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 239000002344 surface layer Substances 0.000 description 2
- 229920001169 thermoplastic Polymers 0.000 description 2
- 239000004416 thermosoftening plastic Substances 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000001739 density measurement Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000013100 final test Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000008376 long-term health Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/16—Measuring arrangements characterised by the use of optical techniques for measuring the deformation in a solid, e.g. optical strain gauge
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8854—Grading and classifying of flaws
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/26—Composites
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Chemical & Material Sciences (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Pathology (AREA)
- Immunology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Signal Processing (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention discloses a composite material damage identification method based on optical fiber measurement and a neural network, which belongs to the technical field of composite material structure health monitoring and comprises finite element simulation and data extraction of a damage-containing structure, strain data processing and training data set generation, establishment of a standardized UNet neural network model, data acquisition in a loading state to obtain dense strain measurement data uniformly distributed along an optical fiber path, two-dimensional mapping interpolation processing of measured strain data to obtain two-dimensional strain field distribution of measured data, processing of the measured strain field data to form a strain field cloud image, processing of image data to be the same image size as a training set, damage identification of the measured strain cloud image by utilizing the trained UNet neural network, damage identification of a real structure by utilizing simulated data set training, and noise reduction processing of a damage identification result to obtain a more accurate and clear damage identification image.
Description
Technical Field
The invention discloses a composite material damage identification method based on optical fiber measurement and a neural network, and belongs to the technical field of composite material structure health monitoring.
Background
Composite materials are widely used in various military and aerospace structures due to their relatively high specific strength and specific stiffness. However, the service environment of the aerospace structure is bad, various damages are easy to generate in the service process of the structure, and the safe operation of the structure is threatened. And the dispersion and large damage form of the composite material are often difficult to predict. Therefore, research on structural health monitoring technology of the carbon fiber composite material is carried out, and the method has important significance in guaranteeing safe operation of the structure, reducing maintenance cost and the like. The distributed optical fiber sensor can sense the temperature and the strain change of the structure and has the advantages of small volume, electromagnetic interference resistance, high spatial resolution and the like. The optical fiber can be used for measuring the measuring points at 1mm intervals along the optical fiber path, can be buried into the composite material, can not greatly affect the material performance, and is widely applied to the field of composite material structure health monitoring.
The structural damage can change the strain field around the damage position, the distributed optical fiber sensor reflects the damage condition of the structure through measuring the structural strain, but the complexity of the real structure and the dispersibility of the composite material make it difficult to give damage judgment standards according to a theoretical model, and a great deal of data are often required to be analyzed by professionals, so that the consumed labor cost is high. In order to reduce the monitoring cost and improve the monitoring quality, an intelligent damage judgment method aiming at distributed optical fiber health monitoring needs to be developed. The UNet neural network is widely applied to the field of medical image diagnosis, can assist doctors to carry out rapid disease diagnosis, and has good abnormality (pathological and injury) recognition capability. The long-term health monitoring of the structure is also a structural health diagnosis process, and the high-density optical fiber layout method is combined with the UNet neural network model, so that the debonding and crack damage of the composite material plate are automatically identified and positioned by utilizing the high-density strain information obtained by the optical fibers and the pixel-level data classification capability of the UNet network.
Disclosure of Invention
The invention aims to provide a composite material damage identification method based on optical fiber measurement and a neural network, which aims to solve the problems that in the prior art, structural damage identification requires a professional to analyze a large amount of data and consumes large labor cost.
The composite material damage identification method based on optical fiber measurement and a neural network comprises the following steps:
s1, finite element simulation and data extraction of a damage-containing structure, wherein the damage-containing composite material structure is simulated through ABAQUS finite element software to obtain strain field data under various damage states
S2, strain data processing and training data set generation, wherein the strain data processing and training data set generation are carried out in the step S1Processing to form a strain field cloud picture, changing the size of a data image, and marking the strain field image to form a training data set;
s3, establishing a standardized UNet neural network model, and performing recognition training on various damage features in the S2 training data set to obtain network models capable of recognizing different damage states;
s4, carrying out high-density distributed optical fiber layout on the composite material sample with damage, and carrying out data acquisition in a loading state to obtain dense strain measurement data epsilon uniformly distributed along an optical fiber path measured ；
S5, performing two-dimensional mapping on the measured strain data obtained in the step S4, and performing interpolation processing to obtain two-dimensional strain field distribution of the measured data
S6, processing the actually measured strain field data obtained in the S5 to form a strain field cloud picture, and processing the image data into the image size same as that of the training set;
s7, performing damage identification on the actually measured strain cloud image obtained in the S6 by using the UNet neural network trained in the S3, and finally, identifying damage of a real structure by using simulated data set training;
s8, carrying out noise reduction treatment on the damage identification result obtained in the step S7 to obtain a more accurate and clear damage identification image.
S1 comprises the following steps:
establishing a finite element model of a carbon fiber resin matrix composite board, presetting interlayer debonding damage and surface crack damage when the composite material is laid, and extracting surface strain field data of a structure but in the directionModeling by parameterizationMultiple sets of strain field data for the lossy structure are obtained for different lesion types and lesion locations.
S2 comprises the following steps: and (3) converting the strain field data of the lossy structure obtained in the step (S1) into image information displayed by a strain field cloud picture in Matlab, adjusting the image size to be a set neural network input size through an image processing method, and generating a data pixel label for supervising and training a UNet neural network model according to the data simulation state.
S3 comprises the following steps:
generating a standard UNet network model in Matlab, selecting a data input size according to identification positioning accuracy, setting the identification types of the UNet network to be 3 types of debonding, cracking and health aiming at debonding and crack damage, performing network training by using the processed strain field data of the damage-containing structure, and identifying strain characteristics of different damages by the network after training is completed;
establishing a UNet neural network by adopting a Matlab platform, wherein the UNet neural network comprises downsampling, upsampling and jump connection;
the compression process is convolution and downsampling to reduce the image size to extract the features of the shallow, the compressed network structure includes three blocks, each block including a convolution of 3*3 using the Relu activation function, a 2 x 2 pooling layer with a stride of 2;
the decoding process is to obtain deep features through deconvolution and up-sampling, and the decoded network structure comprises three program blocks, wherein each program block comprises deconvolution operation of 2 x 2 with a step length of 2, and two convolution of 3*3 with a Relu activation function;
the compression process and the decoding process are connected through jump layer, and the image is thinned by combining deep and shallow features, and prediction segmentation is carried out according to the obtained feature map;
the last layer is classified through convolution of 1x1, and a sigmoid activation function is followed to generate a two-dimensional fault probability image;
the UNet neural network was set to an input size of 128 x 128 and trained on a MatLab platform using a standard training dataset.
S4 comprises the following steps:
manufacturing a prefabricated debonded and crack damaged composite material plate, wherein the debonded damage is in the form of interlayer pre-buried release cloth, the crack damage is prefabricated in the form of surface cutter scribing, and the high-density distributed optical fibers with 5mm intervals are arranged in a spiral mode so that the optical fiber paths cover the monitoring area;
static loading is carried out on the composite material, and strain monitoring is carried out by adopting a distributed optical fiber to obtain epsilon measured The measured strain data is strain data of a plurality of measuring points measured along the optical fiber path.
The high-density distributed optical fiber layout in the S4 meets the requirement of the minimum bending radius of the optical fiber.
S5 comprises the following steps:
epsilon obtained in S4 measured Performing two-dimensional mapping to obtain two-dimensional plane distribution of strain data, performing interpolation processing on the strain data on the plane to obtain a strain field two-dimensional distribution result with a spatial resolution of 1mm
The noise reduction processing in S7 performs noise reduction processing on the result by setting a threshold, and designates the data result with the damage probability less than 0.5 as healthy data.
Compared with the prior art, the invention has the following beneficial effects: the neural network training database provided by the invention is generated in a finite element modeling mode, and the training database is generated or expanded in a simulation mode, so that the training cost can be effectively reduced; the structural damage identification method adopts a distributed optical fiber to acquire structural strain, the distributed optical fiber can perform strain measurement with 1mm spatial resolution along an optical fiber path, and the more the number of measuring points is, the more damage information contained in a real structural damage strain field obtained through interpolation is; meanwhile, the distributed optical fiber sensor has small volume, light weight and electromagnetic interference resistance, and the surface adhesion can not influence the performance of the structure, so that the distributed optical fiber sensor is an ideal sensor for monitoring the structural health; the structural damage identification method adopts a distributed optical fiber spiral layout mode to realize high-density optical fiber path layout, the distributed optical fiber can achieve high-density measurement along the layout path, but the layout path distance is larger under the influence of the minimum bending radius of the optical fiber, and the spiral layout mode can realize high-density optical fiber layout with 5mm distance, so that the interpolation strain field precision is further improved.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
fig. 2 is a diagram of a UNet standard neural network structure according to an embodiment of the present invention;
FIG. 3 is a schematic representation of a finite element model composite layup according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a debonding state in a damaged state of a finite element model structure according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a crack state in a damaged state of a finite element model structure according to an embodiment of the present invention;
FIG. 6 is a strain cloud diagram of an x-direction debonded state of a finite element model according to an embodiment of the present invention under different damage states;
FIG. 7 is a strain cloud chart of an x-direction crack state under different damage states of a finite element model according to an embodiment of the present invention;
FIG. 8 is a diagram of a test piece high density fiber lay debonded state routing scheme in accordance with an embodiment of the present invention;
FIG. 9 is a diagram of a test piece high density fiber lay crack state path scenario in an embodiment of the present invention;
FIG. 10 is a graph showing the debonding state of measured strain data for a distributed optical fiber according to an embodiment of the present invention;
FIG. 11 is a graph of crack state of measured strain data for a distributed optical fiber in accordance with an embodiment of the present invention;
FIG. 12 is a two-dimensional map of strain debonding state curves according to an embodiment of the present invention;
FIG. 13 is a two-dimensional map of strain crack state curves according to an embodiment of the present invention;
FIG. 14 is a graph of neural network training accuracy in accordance with an embodiment of the present invention;
FIG. 15 shows a strain cloud image and damage identification situation of a debonded test piece according to an embodiment of the present invention, where (a) is a strain field cloud image, (b) is a damage identification result, and (c) is a damage identification result after noise reduction;
fig. 16 shows strain cloud images and damage recognition cases of a crack test piece according to an embodiment of the present invention, where (a) strain field cloud images, (b) damage recognition results, and (c) damage recognition results after noise reduction.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions in the present invention will be clearly and completely described below, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, the invention provides a composite material structure damage identification method based on optical fiber measurement and a neural network, which comprises the following steps:
(1) Establishing a finite element model of a carbon fiber resin matrix composite board, presetting interlayer debonding damage and surface crack damage when the composite material is laid, and extracting surface strain field data of a structure but in the directionAnd obtaining multiple groups of strain field data of the lossy structure with different damage types and damage positions by means of parametric modeling. The method solves the problem that a large number of data sets are needed for neural network training, obtains a large number of data through finite element simulation, reduces the difficulty and cost of data acquisition, and has high practicability.
(2) Converting the strain field value obtained in the step (1) into image information displayed by a strain field cloud picture in Matlab, adjusting the image size to be a set neural network input size through an image processing method, and generating a data pixel label for supervising and training a UNet neural network model according to a data simulation state.
(3) Generating a standard UNet network model in Matlab, selecting a data input size according to identification positioning accuracy, and setting the identification type of the UNet network to be 3 types (debonding, cracking and health) aiming at debonding and crack damage; and (3) performing network training by using the damage-containing database generated in the step (2), and identifying strain characteristics of different damages by a network after training is completed.
(4) And manufacturing a prefabricated debonded and crack damaged composite material plate, wherein the debonded damage is in the form of interlayer pre-buried release cloth, and the crack damage is prefabricated in the form of surface cutter scribing. And high-density distributed optical fiber layout with 5mm spacing is performed in a spiral layout mode, so that an optical fiber path covers a monitoring area. Static loading is carried out on the composite material, and strain monitoring is carried out by adopting a distributed optical fiber to obtain measured strain data epsilon measured The measured strain data is strain data of a plurality of measuring points measured along the optical fiber path.
(5) The measured strain data epsilon obtained in the step (4) is processed measured Performing two-dimensional mapping to obtain two-dimensional plane distribution of strain data, performing interpolation processing on the strain data on the plane to obtain a strain field two-dimensional distribution result with a spatial resolution of 1mm
(6) And (3) converting the strain field value obtained in the step (5) into image information displayed by a strain field cloud picture in Matlab, and adjusting the image size to be the set neural network input size through an image processing method.
(7) And (3) inputting the actually measured strain cloud image of the structure in the step (6) into the UNet neural network trained in the step (3) for damage identification, and finally realizing damage identification of the real structure by using a simulation database.
(8) And (3) the damage identification result obtained in the step (6) has misjudgment of partial pixels, so that the noise of the identification image is more, the result can be sharpened in a mode of setting a threshold value, the data result with the damage judgment probability smaller than 0.5 is designated as health data, and the data result with the damage judgment probability larger than or equal to 0.5 is designated as the damage with the maximum corresponding probability, so that the more accurate and clear damage identification image is obtained. Therefore, the damage of the composite material structure is monitored, and the method has important significance for the safe operation of the composite material structure.
The embodiment adopts a carbon fiber composite material plate with the size of 100 mm by 200mm, adopts T300/PC unidirectional prepreg for orthogonal pavement, and adopts an orthogonal symmetrical pavement with the pavement angle of [0/90], 20 layers are paved in total, and the thickness is 1.5mm. The debonding damage is simulated in a form of pre-buried release cloth, the release cloth is arranged between 2-3 layers of carbon fibers, and the debonding size is 20mm; the prefabricated form of crack damage is to cut carbon fiber at the corresponding position of the prepreg by a knife, so that the strain characteristics generated by real cracks can be simulated after the main bearing fiber is broken, the crack length is 30mm, and the crack depth is 3 layers of prepregs on the surface layer.
S1, finite element simulation and data extraction of a lossy structure, wherein the finite element simulation adopts a 100-200 mm flat plate structure with the same size as that of a test piece in an embodiment, and the structure type is a shell structure. The composite lay-up simulation was performed in a continuous shell fashion, with an orthosymmetric lay-up angle of 0/90 being shown in FIG. 3, a single layer thickness of 0.15mm, 20 total layers, and the material properties of each carbon fiber lay-up being shown in Table 1.
TABLE 1 Material Properties of each carbon fiber layup
| E1/Mpa | E2/Mpa | μ12 | G12/Mpa | G13/Mpa | G23/Mpa |
| 149000 | 8000 | 0.34 | 5930 | 5930 | 3227 |
The damage form is set in a material rigidity attenuation form, the debonding damage is set in a region with the size of 20mm in a layer 3, and the material rigidity of the layer is reduced to simulate local rigidity attenuation caused by debonding; the crack damage is set as a non-penetrating crack with a germinated surface, the crack length is 30mm along the width direction of the test piece, the material rigidity of the surface 3 layers within the width range of 1mm is reduced, and the local influence of the crack is simulated. Two forms of injury are shown in figures 4 and 5.
The finite element simulation loading mode is cantilever bending loading, and as the distributed optical fiber can only measure the strain along the length direction of the optical fiber, the data extraction is carried out on the strain field (E11) in the structure x direction in combination with the subsequent test scheme, as shown in fig. 6 and 7. The strain cloud image shows that the debonding damage and the crack damage show different data characteristics, but the boundary characteristics of the debonding damage are not obvious, the damage boundary is difficult to judge artificially, and the UNet network can solve the problem well.
The finite element simulation solves the problem that the neural network training data set is difficult to acquire, reduces the cost of manpower and material resources for acquiring different damage data through experiments, has the consistency with a real structure, and also provides larger requirements for a finite element model. According to data analysis, the finite element model can better reflect the strain results of the real composite material plate in the debonding and crack damage states. In a parametric modeling manner, 500 groups of simulation data sets are generated for different load sizes and damage positions.
S2, strain data processing and training data set generation, converting the strain field value obtained in the step 1 into image information displayed by a strain field cloud picture in Matlab, and expanding the strain field cloud picture data into 1000 groups through data expansion operation (translation, rotation and symmetry). And simultaneously, the image size is adjusted to 128×128 pixels of the neural network input size by an image processing method. And generating a corresponding pixel data tag set according to the structural damage state corresponding to the data, and using the pixel data tag set for supervising and training the UNet neural network model.
S3, training a UNet network capable of identifying damage characteristics of the composite material, and creating a UNet neural network with the same standard as that of the structure of the figure 2 by adopting a Matlab platform, wherein the network structure is mainly divided into three parts: downsampling, upsampling, and skip connection. On the left is the compression process, i.e. reducing the image size by convolution and downsampling, extracting some of the features that are shallow. Three blocks are included, each including a convolution of 3*3 (using the Relu activation function), a 2 x 2 pooling layer with a stride of 2. The right part is the decoding process, i.e. some deep features are obtained by deconvolution and upsampling. Three blocks are included, each including a 2 x 2 deconvolution operation with a stride of 2, and then finally by convolution of two 3*3 (using the Relu activation function). And combining deep and shallow features in a jump layer connection mode in the middle, refining the image, and carrying out prediction segmentation according to the obtained feature map. The last layer is classified by convolution of 1x1, followed by a sigmoid activation function, generating a two-dimensional tomographic probability image.
The neural network was set to 128 x 128 in input size, UNet was trained using the standard training dataset previously generated, training was performed on a MatLab platform, the accuracy curve of the training process is shown in fig. 14, and the final accuracy of the network training was 97.62%.
S4, manufacturing a prefabricated damaged thermoplastic carbon fiber composite material plate, wherein a real structure test piece is 100 mm in 200mm size which is the same as a finite element, orthogonal paving is performed by adopting T700/PC unidirectional prepreg, and the prepreg is a thermoplastic composite material. The debonding damage is simulated in a form of pre-buried release cloth, the release cloth is arranged between 2-3 layers of carbon fibers, and the debonding size is 20mm; the prefabricated form of crack damage is to cut carbon fiber at the corresponding position of the prepreg by a knife, so that the strain characteristics generated by real cracks can be simulated after the main bearing fiber is broken, the crack length is 30mm, and the crack depth is 3 layers of prepregs on the surface layer. The test piece curing process is as follows: and (3) placing the laid prepreg into a die, closing the film, placing the die into a hot press, heating to 250-260 ℃, applying 2-4 MPa pressure, cooling after 3-4 min, and demoulding after the temperature is reduced to 80 ℃.
High-density optical fiber layout and strain data acquisition, in order to obtain a strain field with a more real structure surface, the distributed optical fiber for measuring the strain is required to be laid in a high density. The clamp and the optical fiber bending in the test process can only carry out high-density optical fiber layout on the local range of a test piece, the optical fiber monitoring range is an 80-80 mm section covering a damaged area, the adjacent interval of the optical fibers is 5mm, the distance between strain measuring points on the optical fiber path is set to be 1mm during measurement, the optical fibers are laid in a spiral translation mode, and specific layout forms of different damage types are shown in fig. 8 and 9.
The test data acquisition device uses an odiSIA distributed optical fiber acquisition system of LUNA company, and the test piece is in a cantilever bending mode. The equipment is fully connected and debugged, baseline is collected as the initial state of the structure, in order to eliminate the influence of internal stress in the test piece processing and manufacturing process and possible assembly gaps in the clamp mounting process on the test, 3 loading tests are carried out, the last data are taken for analysis, and the strain data epsilon are actually measured measured The curves are shown in fig. 10 and 11. The graph shows that the strain curve has higher signal-to-noise ratio, no larger data noise occurs, the data result is more reliable, and the feasibility of the high-density optical fiber layout scheme is verified.
S5, actually measured strain data processing is carried out, and a one-dimensional strain curve acquired by the optical fiber is mapped into a two-dimensional plane according to the actual path of the optical fiber. Simultaneously, interpolating the strain data into strain lattices with 1mm intervals by adopting a cubic spline interpolation mode to obtain the two-dimensional distribution of the strain field generated by the measured data
S6, generating and processing strain field cloud patterns, and obtaining strain cloud patterns of different injuries in the monitoring area through an image processing mode, wherein the strain field cloud patterns comprise strain characteristics of the different injuries. The damage characteristics obtained by the experiment are basically consistent with the finite element simulation, the internal debonding damage is shown as local-range strain reduction in a surface strain field, and the change range can be observed in the graph; the surface crack damage is shown as 0 at the boundary of the surface crack in the structural strain, but the optical fiber sensor does not break during optical fiber monitoring, the tensile deformation of the optical fiber at the crack is increased, the range of optical fiber data near the crack is increased, and the strain response caused by the structural rigidity attenuation adopted during finite element simulation is consistent. And according to the final test data form required to be identified, processing a large amount of training data obtained by simulation into a strain cloud image with the range of 80mm containing the damaged area, and processing the strain cloud image into 128 x 128 pixel picture data for identifying and inputting a UNet network. Mapping a one-dimensional strain curve acquired by the optical fiber into a two-dimensional plane according to the actual path of the optical fiber, wherein the distribution of the strain curve in the two-dimensional plane is shown in fig. 12 and 13. By means of image processing, strain cloud charts of different injuries in the monitoring area range are obtained, and strain characteristics of the different injuries are contained in the strain field cloud charts, as shown in fig. 15 (a) and fig. 16 (a).
S7, performing damage identification by using the UNet network, inputting the processed actually measured strain cloud picture into the trained UNet network for performing damage identification, and generating a pixel tag identification result by the UNet network. The UNet network accurately identifies the strain characteristics of debonding and cracks, the damage type is judged very accurately, and misjudgment does not occur, so that the identification accuracy is high. But there is a false positive of some pixels and the boundary data accuracy is low. And inputting the processed actual measurement strain cloud image into a trained UNet network for damage identification, wherein the UNet network generates a pixel tag identification result, and the result is intuitively displayed in a pixel color as shown in fig. 15 (b) and 16 (b).
S8, noise reduction processing is carried out on the identification result, and misjudgment of partial pixels exists after the UNet network identifies damage, so that more noise of the identification image is caused, and noise reduction can be carried out on the result in a mode of setting a threshold value according to the situation. The specific method is that the data result with the damage judgment probability smaller than 0.5 is designated as health data, the data result with the damage judgment probability larger than or equal to 0.5 is designated as the damage with the maximum corresponding probability, a more accurate and clear damage identification image is obtained, and the comparison situation of the noise-reduced identification result and the real result is shown in fig. 15 (c) and fig. 16 (c).
TABLE 2 quantitative determination errors
The maximum boundary of the damage determination was used as a determination criterion for the size of the damage, and the center of the range was used as the damage position, and the quantitative determination error was obtained as shown in table 2. In the embodiment, the maximum positioning error of the UNet neural network identification method on different types of damage is 3.61mm, the maximum error of the damage range is 7mm, and the identification result is good.
The parts which are not described in the invention can be realized by adopting or referring to the prior art.
The above embodiments are only for illustrating the technical aspects of the present invention, not for limiting the same, and although the present invention has been described in detail 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 be modified or some or all of the technical features may be replaced with other technical solutions, which do not depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. The composite material damage identification method based on optical fiber measurement and a neural network is characterized by comprising the following steps of:
s1, finite element simulation and data extraction of a damage-containing structure, wherein the damage-containing composite material structure is simulated through ABAQUS finite element software to obtain strain field data under various damage states
S2, strain data processing and training data set generation, wherein the strain data processing and training data set generation are carried out in the step S1Processing to formChanging the size of a data image according to the strain field cloud picture, and marking the strain field image to form a training data set;
s3, establishing a standardized UNet neural network model, and performing recognition training on various damage features in the S2 training data set to obtain network models capable of recognizing different damage states;
s4, carrying out high-density distributed optical fiber layout on the composite material sample with damage, and carrying out data acquisition in a loading state to obtain dense strain measurement data epsilon uniformly distributed along an optical fiber path measured ；
S5, performing two-dimensional mapping on the measured strain data obtained in the step S4, and performing interpolation processing to obtain two-dimensional strain field distribution of the measured data
S6, processing the actually measured strain field data obtained in the S5 to form a strain field cloud picture, and processing the image data into the image size same as that of the training set;
s7, performing damage identification on the actually measured strain cloud image obtained in the S6 by using the UNet neural network trained in the S3, and finally, identifying damage of a real structure by using simulated data set training;
s8, carrying out noise reduction treatment on the damage identification result obtained in the step S7 to obtain a more accurate and clear damage identification image.
2. The method for identifying damage to a composite material based on optical fiber measurement and neural network according to claim 1, wherein S1 comprises:
establishing a finite element model of a carbon fiber resin matrix composite board, presetting interlayer debonding damage and surface crack damage when the composite material is laid, and extracting surface strain field data of a structure but in the directionAnd obtaining multiple groups of strain field data of the lossy structure with different damage types and damage positions through a parameterized modeling mode.
3. The method for identifying damage to a composite material based on optical fiber measurement and neural network according to claim 1, wherein S2 comprises: and (3) converting the strain field data of the lossy structure obtained in the step (S1) into image information displayed by a strain field cloud picture in Matlab, adjusting the image size to be a set neural network input size through an image processing method, and generating a data pixel label for supervising and training a UNet neural network model according to the data simulation state.
4. The method for identifying damage to a composite material based on optical fiber measurement and neural network according to claim 1, wherein S3 comprises:
generating a standard UNet network model in Matlab, selecting a data input size according to identification positioning accuracy, setting the identification types of the UNet network to be 3 types of debonding, cracking and health aiming at debonding and crack damage, performing network training by using the processed strain field data of the damage-containing structure, and identifying strain characteristics of different damages by the network after training is completed;
establishing a UNet neural network by adopting a Matlab platform, wherein the UNet neural network comprises downsampling, upsampling and jump connection;
the compression process is convolution and downsampling to reduce the image size to extract the features of the shallow, the compressed network structure includes three blocks, each block including a convolution of 3*3 using the Relu activation function, a 2 x 2 pooling layer with a stride of 2;
the decoding process is to obtain deep features through deconvolution and up-sampling, and the decoded network structure comprises three program blocks, wherein each program block comprises deconvolution operation of 2 x 2 with a step length of 2, and two convolution of 3*3 with a Relu activation function;
the compression process and the decoding process are connected through jump layer, and the image is thinned by combining deep and shallow features, and prediction segmentation is carried out according to the obtained feature map;
the last layer is classified through convolution of 1x1, and a sigmoid activation function is followed to generate a two-dimensional fault probability image;
the UNet neural network was set to an input size of 128 x 128 and trained on a MatLab platform using a standard training dataset.
5. The method for identifying damage to a composite material based on optical fiber measurement and neural network according to claim 1, wherein S4 comprises:
manufacturing a prefabricated debonded and crack damaged composite material plate, wherein the debonded damage is in the form of interlayer pre-buried release cloth, the crack damage is prefabricated in the form of surface cutter scribing, and the high-density distributed optical fibers with 5mm intervals are arranged in a spiral mode so that the optical fiber paths cover the monitoring area;
static loading is carried out on the composite material, and strain monitoring is carried out by adopting a distributed optical fiber to obtain epsilon measured The measured strain data is strain data of a plurality of measuring points measured along the optical fiber path.
6. The method for identifying damage to a composite material based on optical fiber measurement and neural network according to claim 1, wherein the high-density distributed optical fiber layout in S4 meets the requirement of minimum bending radius of the optical fiber.
7. The method for identifying damage to a composite material based on optical fiber measurement and neural network according to claim 1, wherein S5 comprises:
epsilon obtained in S4 measured Performing two-dimensional mapping to obtain two-dimensional plane distribution of strain data, performing interpolation processing on the strain data on the plane to obtain a strain field two-dimensional distribution result with a spatial resolution of 1mm
8. The method for identifying damage to composite materials based on optical fiber measurement and neural network according to claim 1, wherein the noise reduction processing in S7 performs noise reduction processing on the result by setting a threshold, and designates the data result with the damage probability less than 0.5 as health data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311045468.8A CN117347378B (en) | 2023-08-18 | 2023-08-18 | Composite material damage identification method based on optical fiber measurement and neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311045468.8A CN117347378B (en) | 2023-08-18 | 2023-08-18 | Composite material damage identification method based on optical fiber measurement and neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117347378A true CN117347378A (en) | 2024-01-05 |
| CN117347378B CN117347378B (en) | 2024-08-13 |
Family
ID=89356382
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311045468.8A Active CN117347378B (en) | 2023-08-18 | 2023-08-18 | Composite material damage identification method based on optical fiber measurement and neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117347378B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105352816A (en) * | 2015-11-26 | 2016-02-24 | 中国航空工业集团公司沈阳飞机设计研究所 | Composite structure failure prediction analytical method |
| CN110095470A (en) * | 2019-04-15 | 2019-08-06 | 北京航空航天大学 | A kind of Crack Damage quantitative approach based on fiber-optic grating sensor |
| CN111024821A (en) * | 2019-12-30 | 2020-04-17 | 大连理工大学 | Composite material storage box health monitoring system and method |
| CN112504808A (en) * | 2020-11-02 | 2021-03-16 | 北京空天技术研究所 | Aircraft thermal protection system damage diagnosis method based on machine learning algorithm |
| CN113484418A (en) * | 2021-07-09 | 2021-10-08 | 大连理工大学 | Damage cooperative diagnosis technology based on multi-frequency domain response signals |
| KR20220011456A (en) * | 2020-07-21 | 2022-01-28 | 울산과학기술원 | Method of detecting the location of damage for composite using machine learning |
| WO2022077605A1 (en) * | 2020-10-15 | 2022-04-21 | 青岛理工大学 | Wind turbine blade image-based damage detection and localization method |
| CN114970240A (en) * | 2022-04-28 | 2022-08-30 | 上海交通大学 | Method and equipment for rapidly evaluating load state of multi-phase composite structure image |
| US20220292338A1 (en) * | 2021-03-09 | 2022-09-15 | Chevron U.S.A. Inc. | Geomechanics Informed Machine Intelligence |
| CN115579086A (en) * | 2022-10-18 | 2023-01-06 | 天津大学 | Metal material crack propagation path prediction method |
| CN115575104A (en) * | 2022-09-16 | 2023-01-06 | 浙江大学 | Fan blade damage rapid detection method based on inverse finite element reconstruction image recognition |
| CN115859708A (en) * | 2022-11-16 | 2023-03-28 | 大连理工大学 | Honeycomb sandwich structure damage model correction method based on distributed optical fiber measurement |
| US20230099872A1 (en) * | 2021-09-27 | 2023-03-30 | Northwestern University | Image correlation for end-to-end displacement and strain measurement |
-
2023
- 2023-08-18 CN CN202311045468.8A patent/CN117347378B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105352816A (en) * | 2015-11-26 | 2016-02-24 | 中国航空工业集团公司沈阳飞机设计研究所 | Composite structure failure prediction analytical method |
| CN110095470A (en) * | 2019-04-15 | 2019-08-06 | 北京航空航天大学 | A kind of Crack Damage quantitative approach based on fiber-optic grating sensor |
| CN111024821A (en) * | 2019-12-30 | 2020-04-17 | 大连理工大学 | Composite material storage box health monitoring system and method |
| KR20220011456A (en) * | 2020-07-21 | 2022-01-28 | 울산과학기술원 | Method of detecting the location of damage for composite using machine learning |
| WO2022077605A1 (en) * | 2020-10-15 | 2022-04-21 | 青岛理工大学 | Wind turbine blade image-based damage detection and localization method |
| CN112504808A (en) * | 2020-11-02 | 2021-03-16 | 北京空天技术研究所 | Aircraft thermal protection system damage diagnosis method based on machine learning algorithm |
| US20220292338A1 (en) * | 2021-03-09 | 2022-09-15 | Chevron U.S.A. Inc. | Geomechanics Informed Machine Intelligence |
| CN113484418A (en) * | 2021-07-09 | 2021-10-08 | 大连理工大学 | Damage cooperative diagnosis technology based on multi-frequency domain response signals |
| US20230099872A1 (en) * | 2021-09-27 | 2023-03-30 | Northwestern University | Image correlation for end-to-end displacement and strain measurement |
| CN114970240A (en) * | 2022-04-28 | 2022-08-30 | 上海交通大学 | Method and equipment for rapidly evaluating load state of multi-phase composite structure image |
| CN115575104A (en) * | 2022-09-16 | 2023-01-06 | 浙江大学 | Fan blade damage rapid detection method based on inverse finite element reconstruction image recognition |
| CN115579086A (en) * | 2022-10-18 | 2023-01-06 | 天津大学 | Metal material crack propagation path prediction method |
| CN115859708A (en) * | 2022-11-16 | 2023-03-28 | 大连理工大学 | Honeycomb sandwich structure damage model correction method based on distributed optical fiber measurement |
Non-Patent Citations (3)
| Title |
|---|
| XIAOXI QU: "Various static loading condition monitoring of carbon fiber composite cylinder with integrated optical fiber sensors", 《OPTICAL FIBER TECHNOLOGY》, vol. 83, 20 January 2024 (2024-01-20), pages 1 - 14, XP087464637, DOI: 10.1016/j.yofte.2024.103685 * |
| 张峻铭: "人工智能在复合材料研究中的应用", 《力学进展》, vol. 51, no. 4, 31 December 2021 (2021-12-31), pages 865 - 900 * |
| 李建乐: "基于深度学习的分布式光纤损伤识别方法", 《机械工程学报》, vol. 58, no. 8, 30 April 2022 (2022-04-30), pages 88 - 95 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117347378B (en) | 2024-08-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wijaya et al. | An automatic methodology to CT-scans of 2D woven textile fabrics to structured finite element and voxel meshes | |
| Rezaie et al. | Comparison of crack segmentation using digital image correlation measurements and deep learning | |
| Liu et al. | Deep autoencoder thermography for defect detection of carbon fiber composites | |
| Ní Annaidh et al. | Automated estimation of collagen fibre dispersion in the dermis and its contribution to the anisotropic behaviour of skin | |
| Meister et al. | Synthetic image data augmentation for fibre layup inspection processes: Techniques to enhance the data set | |
| CN110378252A (en) | A kind of distress in concrete recognition methods based on depth migration study | |
| Marani et al. | Modeling and classification of defects in CFRP laminates by thermal non-destructive testing | |
| Wang et al. | Digital image correlation (DIC) based damage detection for CFRP laminates by using machine learning based image semantic segmentation | |
| CN112837294B (en) | Thermal imaging defect detection method based on convolution self-encoder image amplification | |
| Xu et al. | An infrared thermal image processing framework based on superpixel algorithm to detect cracks on metal surface | |
| Liu et al. | Non-destructive defect evaluation of polymer composites via thermographic data analysis: A manifold learning method | |
| CN111178392B (en) | Aero-engine hole detection image damage segmentation method based on deep neural network | |
| CN109214388B (en) | Tumor segmentation method and device based on personalized fusion network | |
| JP2017156271A (en) | Structure analyzing device, structure analyzing method, and three-dimensional fiber material manufacturing method | |
| Song et al. | Material twins generation of woven polymer composites based on ResL-U-Net convolutional neural networks | |
| Nascimento et al. | Computer vision based quality control for additive manufacturing parts | |
| CN117347378B (en) | Composite material damage identification method based on optical fiber measurement and neural network | |
| Pedrayes et al. | Semantic segmentation for non-destructive testing with step-heating thermography for composite laminates | |
| Wan et al. | Railway tracks defects detection based on deep convolution neural networks | |
| He et al. | Intelligent detection algorithm based on 2D/3D-UNet for internal defects of carbon fiber composites | |
| Peng et al. | Aggregate boundary recognition of asphalt mixture CT images based on convolutional neural networks | |
| Guo et al. | Detection of insulator defects based on YOLO V3 | |
| Perez et al. | Sensitivity thermal analysis in the laser-assisted tape placement process | |
| CN117197085A (en) | Road rapid-inspection image pavement disease detection method based on improved YOLOv8 network | |
| Emerson et al. | Insegt fibre: a user-friendly software for individual fibre segmentation |
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 | ||
| GR01 | Patent grant |