CN110276743A - A kind of structural damage degree recognition methods based on convolutional neural networks - Google Patents
A kind of structural damage degree recognition methods based on convolutional neural networks Download PDFInfo
- Publication number
- CN110276743A CN110276743A CN201910402301.XA CN201910402301A CN110276743A CN 110276743 A CN110276743 A CN 110276743A CN 201910402301 A CN201910402301 A CN 201910402301A CN 110276743 A CN110276743 A CN 110276743A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- layer
- convolutional
- unit
- 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
- 230000006378 damage Effects 0.000 title claims abstract description 58
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 31
- 239000011159 matrix material Substances 0.000 claims abstract description 19
- 208000027418 Wounds and injury Diseases 0.000 claims description 9
- 208000014674 injury Diseases 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 239000000470 constituent Substances 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000004321 preservation Methods 0.000 claims 1
- 230000001066 destructive effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006735 deficit Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The structural damage degree recognition methods based on convolutional neural networks that the invention discloses a kind of, the following steps are included: S1: building simple beam structure simultaneously acquires training sample data according to simple beam structure, the simple beam structure includes N number of unit, each unit has different modal strain energies, and the training sample data include the modal strain energy of each unit;S2: input matrix is constructed according to the modal strain energy of each unit;S3: according to input matrix, the model of convolutional neural networks is constructed;S4: using include each unit modal strain energy training sample data training convolutional neural networks and save trained convolutional neural networks;S5: unknown simple beam structure damage is predicted using trained convolutional neural networks.The method that the present invention utilizes convolutional neural networks weight to share, reduces calculating parameter, and calculating speed has positive meaning quickly, for the application of large scale structure.
Description
Technical field
The present invention relates to field of neural networks, more particularly, to a kind of structural damage journey based on convolutional neural networks
Spend recognition methods.
Background technique
Damage Assessment Method has very great significance in damage check field tool.Structural damage not only threatens the people's
Life security also causes huge national property loss.Structure is the important support in people's life, only ensures that structure is pacified
Entirely, just daily life can be made smooth, then the non-destructive tests of structure are particularly important, and how accurately to be known
Other degree of injury is the major issue currently faced.
The damage characteristic of structure usually shows the variation of structure build-in attribute, but the complexity due to structure, environment,
The influence of data processing method etc. can have many interference informations, to cause the effect of non-destructive tests and bad.Knot at present
Structure non-destructive tests are in the lower stage, can only judge whether structure has damage and rough damage position, and for damage
Degree can not make accurate judgement, and with being constantly progressive for computerized algorithm, convolutional neural networks are gradually applied to every field,
Convolutional neural networks have excellent nonlinear fitting ability, non-linear change tendencies when can learn to structural damage, because
Convolutional neural networks, can be applied to the identification field of degree of injury by this.
Summary of the invention
The present invention is to overcome that accurate judgement can not be carried out to degree of injury described in the above-mentioned prior art, provides one kind and is based on
The structural damage degree recognition methods of volume and neural network.
In order to solve the above technical problems, technical scheme is as follows:
A kind of structural damage degree recognition methods based on convolutional neural networks, which comprises the following steps:
S1: building simple beam structure simultaneously acquires training sample data according to simple beam structure, and the simple beam structure includes N
A unit, each unit have different modal strain energies, and the training sample data include the modal strain energy of each unit;
S2: input matrix is constructed according to the modal strain energy of each unit;
S3: according to input matrix, the model of convolutional neural networks is constructed, the input matrix is the defeated of convolutional neural networks
Enter, the output of convolutional neural networks is the degree of injury of the simple beam structure of prediction;
S4: using include each unit modal strain energy training sample data training convolutional neural networks and save instruction
The convolutional neural networks perfected;
S5: unknown simple beam structure damage is predicted using trained convolutional neural networks.
Preferably, the simple beam structure includes 36 units.
Preferably, the training sample data are to acquire each unit to damage 20%, 30%, 40%, 50%, 60% respectively
Damage data, totally 36 × 5=180 kind situation, the modal strain energy of available 36 units in each case, by constituent parts
The damage data and corresponding modal strain energy of damage 20%, 30%, 40%, 50% are as training set, constituent parts damage 60%
Damage data and corresponding modal strain energy as test set.
Preferably, the input matrix is 6 × 6 matrixes, and the element of the input matrix is the modal strain of each unit
Energy.
Preferably, the model of convolutional neural networks includes input layer, the first convolutional layer, pond layer, volume Two in step S3
Lamination, full articulamentum and recurrence layer, the first convolutional layer include 50 convolution kernels, and the second convolutional layer includes 100 convolution kernels.
Preferably, specific step is as follows by step S4:
S4.1: training set enters the first convolutional layer by input layer, and obtains 50 by 50 convolution kernels of the first convolutional layer
A 3 × 3 eigenmatrix;
S4.2: 50 3 × 3 eigenmatrixes that step S4.1 is obtained are input to pond layer, obtain 50 2 of Chi Huahou
× 2 eigenmatrix;
S4.3: 50 2 × 2 eigenmatrixes of Chi Huahou are input to the second convolutional layer, by 100 in the second convolutional layer
A convolution kernel obtains 100 characteristic values;
S4.4: the vector of degree of injury is represented using 100 characteristic values as the input of full articulamentum and from recurrence layer output;
S4.5: the vector for returning layer output is compared with corresponding actual damage degree, obtains the mean square error of the two
Difference;
S4.6: the weight parameter of volume and convolutional neural networks is adjusted according to mean square error, repeats S4.1 to S4.5, Zhi Daojun
Square error completes training after no longer changing, and saves the convolutional neural networks for completing training.
Preferably, stochastic gradient descent method adjustment volume and the weight parameter of neural network are used in step S4.6.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
The present invention utilizes the Nonlinear Learning ability of convolutional neural networks, obtains the nonlinear change under Injured level
Rule saves obtained weight parameter, and convolutional neural networks can regard a nonlinear function as at this time, when the new damage of input
It is available close to true prediction result when hurting the corresponding modal strain energy data of degree;Convolutional neural networks utilize simultaneously
The shared method of weight, reduces calculating parameter, and calculating speed has positive meaning quickly, for the application of large scale structure.
Detailed description of the invention
Fig. 1 is a kind of structural damage degree recognition methods flow chart based on convolutional neural networks.
Fig. 2 is simple beam structure schematic diagram.
Fig. 3 is the building schematic diagram of input matrix;
In figure, E1~E36 is the modal strain energy of 36 units;Input of the A matrix as convolutional neural networks.
Fig. 4 is convolutional neural networks model schematic.
Fig. 5 is convolution operation schematic illustration.
Fig. 6 is pond operating principle schematic diagram.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent;
In order to better illustrate this embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent actual product
Size;
To those skilled in the art, it is to be understood that certain known features and its explanation, which may be omitted, in attached drawing
's.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
A kind of structural damage degree recognition methods based on convolutional neural networks, such as Fig. 1, comprising the following steps:
S1: building simple beam structure simultaneously acquires training sample data according to simple beam structure, and the simple beam structure includes
36 units, such as Fig. 2, each unit have different modal strain energies, and the training sample data include the mould of each unit
State strain energy, the training sample data are the damage that each unit of acquisition damages 20%, 30%, 40%, 50%, 60% respectively
Data, totally 36 × 5=180 kind situation, the modal strain energy of available 36 units, constituent parts are damaged in each case
20%, 30%, 40%, 50% damage data and corresponding modal strain energy are as training set, the damage of constituent parts damage 60%
Hurt data and corresponding modal strain energy as test set;
S2: according to the modal strain energy of each unit construct input matrix, the input matrix be 6 × 6 matrixes, such as Fig. 3,
The element of the input matrix is the modal strain energy of each unit;
S3: according to input matrix, the model of convolutional neural networks is constructed, the input matrix is the defeated of convolutional neural networks
Enter, the output of convolutional neural networks is the degree of injury of the simple beam structure of prediction, and the model of convolutional neural networks includes input
Layer, the first convolutional layer, pond layer, the second convolutional layer, full articulamentum and recurrence layer, the first convolutional layer include 50 convolution kernels, the
Two convolutional layers include 100 convolution kernels, such as Fig. 4, by it is different by different degree of impairment be different vectors, for example, No.1 list
Member damage 20% is set as vector [0.2,0,0 ..., 0], and No. two unit damages 20% are set as vector [0,0.2,0 ..., 0] ...,
And so on;
S4: using include each unit modal strain energy training sample data training convolutional neural networks and save instruction
The convolutional neural networks perfected;Wherein, convolution sum pond operating principle such as Fig. 5,6, the specific steps are as follows:
S4.1: training set enters the first convolutional layer by input layer, and obtains 50 by 50 convolution kernels of the first convolutional layer
A 3 × 3 eigenmatrix;
S4.2: 50 3 × 3 eigenmatrixes that step S4.1 is obtained are input to pond layer, obtain 50 2 of Chi Huahou
× 2 eigenmatrix;
S4.3: 50 2 × 2 eigenmatrixes of Chi Huahou are input to the second convolutional layer, by 100 in the second convolutional layer
A convolution kernel obtains 100 characteristic values;
S4.4: the vector of degree of injury is represented using 100 characteristic values as the input of full articulamentum and from recurrence layer output;
S4.5: the vector for returning layer output is compared with corresponding actual damage degree, obtains the mean square error of the two
Difference;
S4.6: the weight parameter of stochastic gradient descent method convolutional neural networks is used according to mean square error adjustment, is repeated
S4.1 to S4.5 completes training after mean square error no longer changes, and saves the convolutional neural networks for completing training.
S5: unknown simple beam structure damage is predicted using trained convolutional neural networks.
The same or similar label correspond to the same or similar components;
The terms describing the positional relationship in the drawings are only for illustration, should not be understood as the limitation to this patent;
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (7)
1. a kind of structural damage degree recognition methods based on convolutional neural networks, which comprises the following steps:
S1: building simple beam structure simultaneously acquires training sample data according to simple beam structure, and the simple beam structure includes N number of list
Member, each unit have different modal strain energies, and the training sample data include the modal strain energy of each unit;
S2: input matrix is constructed according to the modal strain energy of each unit;
S3: according to input matrix, constructing the model of convolutional neural networks, and the input matrix is the input of convolutional neural networks,
The output of convolutional neural networks is the degree of injury of the simple beam structure of prediction;
S4: it is trained using the training sample data training convolutional neural networks and preservation for the modal strain energy for including each unit
Convolutional neural networks;
S5: unknown simple beam structure damage is predicted using trained convolutional neural networks.
2. the structural damage degree recognition methods according to claim 1 based on convolutional neural networks, which is characterized in that institute
Stating simple beam structure includes 36 units.
3. the structural damage degree recognition methods according to claim 2 based on convolutional neural networks, which is characterized in that institute
Stating training sample data is the damage data that each unit of acquisition damages 20%, 30%, 40%, 50%, 60% respectively, totally 36 ×
5=180 kind situation, the modal strain energy of available 36 units in each case, constituent parts are damaged 20%, 30%,
40%, 50% damage data and corresponding modal strain energy be as training set, the damage data of constituent parts damage 60% and right
The modal strain energy answered is as test set.
4. the structural damage degree recognition methods according to claim 3 based on convolutional neural networks, which is characterized in that institute
Stating input matrix is 6 × 6 matrixes, and the element of the input matrix is the modal strain energy of each unit.
5. the structural damage degree recognition methods according to claim 4 based on convolutional neural networks, which is characterized in that step
The model of convolutional neural networks includes input layer, the first convolutional layer, pond layer, the second convolutional layer, full articulamentum and returns in rapid S3
Return layer, the first convolutional layer includes 50 convolution kernels, and the second convolutional layer includes 100 convolution kernels.
6. the structural damage degree recognition methods according to claim 5 based on convolutional neural networks, which is characterized in that step
Specific step is as follows by rapid S4:
S4.1: training set by input layer enter the first convolutional layer, and by 50 convolution kernels of the first convolutional layer obtain 50 3 ×
3 eigenmatrix;
S4.2: 50 3 × 3 eigenmatrixes that step S4.1 is obtained are input to pond layer, obtain 50 2 × 2 of Chi Huahou
Eigenmatrix;
S4.3: being input to the second convolutional layer for 50 2 × 2 eigenmatrixes of Chi Huahou, by 100 volumes in the second convolutional layer
Product core obtains 100 characteristic values;
S4.4: the vector of degree of injury is represented using 100 characteristic values as the input of full articulamentum and from recurrence layer output;
S4.5: the vector for returning layer output is compared with corresponding actual damage degree, obtains the mean square error of the two;
S4.6: according to mean square error adjust convolutional neural networks weight parameter, repeat S4.1 to S4.5, until mean square error not
Training is completed after changing again, saves the convolutional neural networks for completing training.
7. the structural damage degree recognition methods according to claim 6 based on convolutional neural networks, which is characterized in that step
Stochastic gradient descent method adjustment volume and the weight parameter of neural network are used in rapid S4.6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910402301.XA CN110276743A (en) | 2019-05-14 | 2019-05-14 | A kind of structural damage degree recognition methods based on convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910402301.XA CN110276743A (en) | 2019-05-14 | 2019-05-14 | A kind of structural damage degree recognition methods based on convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110276743A true CN110276743A (en) | 2019-09-24 |
Family
ID=67959968
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910402301.XA Pending CN110276743A (en) | 2019-05-14 | 2019-05-14 | A kind of structural damage degree recognition methods based on convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110276743A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111815510A (en) * | 2020-09-11 | 2020-10-23 | 平安国际智慧城市科技股份有限公司 | Image processing method based on improved convolutional neural network model and related equipment |
CN113139691A (en) * | 2021-04-30 | 2021-07-20 | 北华大学 | High-speed rail bridge health monitoring system and method based on piezoelectric sensor |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109145446A (en) * | 2018-08-22 | 2019-01-04 | 广东工业大学 | A kind of Structural Damage Identification based on modal strain energy and convolutional neural networks |
-
2019
- 2019-05-14 CN CN201910402301.XA patent/CN110276743A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109145446A (en) * | 2018-08-22 | 2019-01-04 | 广东工业大学 | A kind of Structural Damage Identification based on modal strain energy and convolutional neural networks |
Non-Patent Citations (2)
Title |
---|
吴春志等: "基于一维卷积神经网络的齿轮箱故障诊断", 《振动与冲击》 * |
陈贡发 等: "基于曲率模态差和模态置信的钢架损伤识别研究", 《广东建材》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111815510A (en) * | 2020-09-11 | 2020-10-23 | 平安国际智慧城市科技股份有限公司 | Image processing method based on improved convolutional neural network model and related equipment |
CN111815510B (en) * | 2020-09-11 | 2020-12-22 | 平安国际智慧城市科技股份有限公司 | Image processing method based on improved convolutional neural network model and related equipment |
CN113139691A (en) * | 2021-04-30 | 2021-07-20 | 北华大学 | High-speed rail bridge health monitoring system and method based on piezoelectric sensor |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109886358B (en) | Human behavior recognition method based on multi-time-space information fusion convolutional neural network | |
CN110472779A (en) | A kind of power-system short-term load forecasting method based on time convolutional network | |
CN112001110B (en) | Structural damage identification monitoring method based on vibration signal space real-time recurrent graph convolutional neural network | |
CN109523021B (en) | Dynamic network structure prediction method based on long-time and short-time memory network | |
KR102275126B1 (en) | Electrical facility safety evaluation system using artificial intelligence technique | |
Hwang et al. | Artificial intelligent analyzer for mechanical properties of rolled steel bar by using neural networks | |
CN103714382B (en) | A kind of multi-index comprehensive evaluation method for reliability of urban rail train security detection sensor network | |
CN105469376A (en) | Method and device for determining picture similarity | |
CN110276743A (en) | A kind of structural damage degree recognition methods based on convolutional neural networks | |
CN105373830A (en) | Prediction method and system for error back propagation neural network and server | |
CN103955714A (en) | Navy detection model construction method and system and navy detection method | |
CN106022287A (en) | Over-age face verification method based on deep learning and dictionary representation | |
CN111159863A (en) | Neural network algorithm-based material deformation and failure prediction method and system | |
CN112836937A (en) | Flood disaster loss evaluation method based on entropy weight and BP neural network technology | |
CN117059269A (en) | Juvenile myopia prediction method and model based on deep learning | |
CN112949944B (en) | Intelligent groundwater level prediction method and system based on space-time characteristics | |
CN110781968B (en) | Extensible class image identification method based on plastic convolution neural network | |
CN110145695B (en) | Heating pipeline leakage detection method based on deep confidence network information fusion | |
CN111310964B (en) | Load prediction method and device | |
CN112818455A (en) | Bridge structure response monitoring method and system | |
Wang et al. | Convolution neural network based load model parameter selection considering short-term voltage stability | |
Jovanović et al. | Application of аrtificial neural networks for prediction of heating energy consumption in university buildings | |
Cao et al. | The urban arterial traffic flow forecasting based on BP neural network | |
Jun | Structural damage detection using parameters combined with changes in flexibility based on BP neural networks | |
Song et al. | Predicting human locations with big five personality and neural network |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190924 |