CN113657217A - Concrete state recognition model based on improved BP neural network - Google Patents
Concrete state recognition model based on improved BP neural network Download PDFInfo
- Publication number
- CN113657217A CN113657217A CN202110879316.2A CN202110879316A CN113657217A CN 113657217 A CN113657217 A CN 113657217A CN 202110879316 A CN202110879316 A CN 202110879316A CN 113657217 A CN113657217 A CN 113657217A
- Authority
- CN
- China
- Prior art keywords
- neural network
- improved
- acoustic emission
- concrete
- follows
- 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.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
A concrete state recognition model based on an improved BP neural network. Step 1, acoustic emission signal acquisition: collecting concrete signals in different states by using an acoustic emission collection system, wherein a data collection card selects an NI USB-6366 high-speed card, and selects LABVIEW software to build an AE signal collection system; step 2, feature vector composition: decomposing the acoustic emission signal by using an EMD algorithm to obtain an IMF component, then calculating a complex domain energy value corresponding to the IMF component, and forming a feature vector; step 3, improved BP network training: training the improved BP neural network by using the characteristic vector generated in the step 2 until the network converges; and 4, online identification of the model: and (3) installing the improved BP neural network obtained by training in the steps 1-3 in an upper computer, and carrying out state recognition on the acoustic emission signals uploaded by the AE signal acquisition system in real time. The method can accurately and effectively identify the state of the concrete, and has good practical application value.
Description
Technical Field
The invention relates to the field of concrete state recognition, in particular to a concrete state recognition model based on an improved BP neural network.
Background
The concrete structure or the member mainly bears pressure, the concrete member often generates brittle failure in actual engineering, cracks in concrete can work normally under the controllable condition, when the cracks extend to a certain width, a series of problems of bearing capacity reduction, reliability reduction and the like can be caused, the life and property safety of people can be seriously threatened even seriously, if the occurrence and development change conditions of the cracks in the concrete can be captured in time, the position and the severity of concrete damage can be identified in time, the monitoring and diagnosis on a heavy engineering structure or an important structural part can be pertinently carried out, and the early warning on structural failure is carried out, so that the safety of engineering is effectively improved.
The concrete state is usually four, namely a normal state, an initial compaction state, a crack propagation state and a through failure state, but the boundary between the four signals is not so obvious, so that the concrete state is accurately and effectively identified with certain difficulty.
The invention relates to a concrete state recognition method, in particular to a concrete acoustic emission damage recognition system based on a sensing technology (202110097243.1), which comprises a control block, a control panel and other modules, wherein one end of the control panel extends into the control block, a positioning device is arranged on an electric wire and is fixed on a building wall or other building structures, so that the electric wire can be conveniently and flexibly wired repeatedly, an acoustic generator can be rapidly installed on one side of a building to be detected, and a detector can operate on the periphery of the building, so that the safety of the whole detection work is improved, but the invention does not describe in detail how to judge the concrete state from an acoustic emission signal. The invention discloses a concrete crack identification method based on an image processing technology (202110105922.9), which is characterized in that an acquired RGB image of a concrete surface crack is subjected to gray scale ratio, filtering, segmentation and other processing to obtain a binary gray scale image, then crack judgment and segment connection processing are performed, the image is segmented by adopting a threshold segmentation method to distinguish a crack region from a background region, then cracks and residual noise are further distinguished, cracks are accurately identified, and finally crack parameters are calculated.
Disclosure of Invention
In order to solve the problems, the invention provides a concrete state identification model based on an improved BP neural network on the basis of an EMD decomposition algorithm and the BP neural network. In order to highlight the characteristics of IMF obtained after EMD decomposition as much as possible, the patent provides a complex domain energy value index by combining two characteristics of a time domain and a frequency domain; in addition, in order to solve the problems that gradient disappearance easily occurs to the BP neural network activation function and the generalization of the model is insufficient, the patent proposes a Nonlinear fusion unit (NLFU) activation function and an average weighted minu-type distance loss function, thereby improving the speed of model convergence and the classification and recognition accuracy. To achieve the purpose, the invention provides a concrete state identification model based on an improved BP neural network, which comprises the following specific steps:
step 2, feature vector composition: decomposing the acoustic emission signal by using an EMD algorithm to obtain an IMF component, then calculating a complex domain energy value corresponding to the IMF component, and forming a feature vector;
and 4, online identification of the model: and (3) installing the improved BP neural network obtained by training in the steps 1-3 in an upper computer, and carrying out state recognition on the acoustic emission signals uploaded by the AE signal acquisition system in real time.
Further, the specific steps of obtaining the IMF by utilizing the EMD decomposition in the step 2, then solving the corresponding complex domain energy value, and forming the feature vector are as follows:
step 2.1, calculating the mean value of the upper envelope curve and the lower envelope curve of the acoustic emission signal x (t), updating the signal in a difference value iteration mode to continuously obtain an IMF component, finally judging whether iteration is terminated according to a standard deviation sd, stopping iteration when sd is 0.2-0.3, wherein the expression of sd is as follows:
wherein T represents the length of the signal, hk-1(t) and hkAnd (t) IMF components obtained by the k-1 th decomposition and the k-th decomposition respectively.
Step 2.2, calculating complex domain energy values ce of each IMF component, wherein the specific calculation expression of ce is as follows:
where F (i) represents the value of the i-th frequency domain point of x (t), and F (·) represents the fourier transform.
And 2.3, combining the complex domain energy values corresponding to the IMF obtained by calculation in the step 2.2 into a feature vector, setting the dimensionality of the feature vector to be 12, if the total number of the IMFs obtained by decomposition is less than 12, filling zero at the tail of the feature vector to enable the dimensionality to be 12, and if the total number of the IMFs obtained by decomposition is more than 12, removing redundant dimensionalities.
Further, the specific steps of training the improved BP neural network in step 3 are as follows:
step 3.1, building a BP neural network, wherein the structure is as follows: the input layer-hidden layer 1-hidden layer 2-output layer, the specific numerical value is 12-8-6-4;
step 3.2, inputting the feature vector obtained in step 2 into a BP neural network, and processing in the hidden layer 1 and the hidden layer 2 by using a proposed Nonlinear fusion unit (NLFU) activation function, wherein the NLFU has an expression as follows:
step 3.3, solving the loss function of the current iteration according to the output of the output layer and the value of the theoretical label, and the patent provides an average weighted Min-type distance dmawThe expression is as follows:
where S represents the number of weights, set to 5 in this patent, n is the dimensionality of the output vector, 4 in this patent,
and
the k-dimension values of the actual output vector and the theoretical output vector are respectively.
And 3.4, updating the weight coefficient and the bias coefficient in the BP network by using a random gradient descent algorithm, and repeating the steps 3.1-3.3 until a set convergence condition is reached or a set iteration number is reached.
Further, the online identification of the model in step 4 is specifically described as follows:
installing the improved BP neural network obtained by training in the steps 1-3 in an upper computer, and carrying out state recognition on the acoustic emission signals uploaded by the AE signal acquisition system in real time, wherein the four states of the recognized concrete are respectively as follows: normal state, initial compaction state, crack propagation state, and through failure state.
The concrete state recognition model based on the improved BP neural network has the advantages that: the invention has the technical effects that:
1. in order to highlight the characteristics contained in the IMF obtained after EMD decomposition as much as possible, the invention combines two characteristics of a time domain and a frequency domain, provides a complex domain energy value index, enhances the characteristics of BP neural network input data, and improves the identification accuracy of the model;
2. the invention provides the NLFU activation function and the average weighted Min-type distance loss function, solves the problems that the BP neural network activation function is easy to have gradient disappearance and the generalization of the model is insufficient, and improves the speed of model convergence and the classification recognition precision.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a block diagram of an improved BP neural network used in the present invention.
Detailed Description
The invention is described in further detail below with reference to the following detailed description and accompanying drawings:
the invention provides a concrete state identification model based on an improved BP neural network, aiming at realizing accurate identification of different states of concrete and ensuring safety. FIG. 1 is a flow chart of the present invention, and the steps of the present invention will be described in detail in conjunction with the flow chart.
step 2, feature vector composition: decomposing the acoustic emission signal by using an EMD algorithm to obtain an IMF component, then calculating a complex domain energy value corresponding to the IMF component, and forming a feature vector;
in step 2, after obtaining the IMF by EMD decomposition, the corresponding complex domain energy value is solved, and the specific steps of forming the characteristic vector are as follows:
step 2.1, calculating the mean value of the upper envelope curve and the lower envelope curve of the acoustic emission signal x (t), updating the signal in a difference value iteration mode to continuously obtain an IMF component, finally judging whether iteration is terminated according to a standard deviation sd, stopping iteration when sd is 0.2-0.3, wherein the expression of sd is as follows:
wherein T represents the length of the signal, hk-1(t) and hkAnd (t) IMF components obtained by the k-1 th decomposition and the k-th decomposition respectively.
Step 2.2, calculating complex domain energy values ce of each IMF component, wherein the specific calculation expression of ce is as follows:
where F (i) represents the value of the ith frequency domain point of x (t), and F (·) represents the Fourier transform.
And 2.3, combining the complex domain energy values corresponding to the IMF obtained by calculation in the step 2.2 into a feature vector, setting the dimensionality of the feature vector to be 12, if the total number of the IMFs obtained by decomposition is less than 12, filling zero at the tail of the feature vector to enable the dimensionality to be 12, and if the total number of the IMFs obtained by decomposition is more than 12, removing redundant dimensionalities.
the specific steps of training the improved BP neural network in the step 3 are as follows:
step 3.1, building a BP neural network, wherein the structure is as follows: the input layer-hidden layer 1-hidden layer 2-output layer, the specific numerical value is 12-8-6-4;
step 3.2, inputting the feature vector obtained in step 2 into a BP neural network, and processing in the hidden layer 1 and the hidden layer 2 by using a proposed Nonlinear fusion unit (NLFU) activation function, wherein the NLFU has an expression as follows:
step 3.3, solving the loss function of the current iteration according to the output of the output layer and the value of the theoretical label, and the patent provides an average weighted Min-type distance dmawThe expression is as follows:
where S represents the number of weights, set to 5 in this patent, n is the dimensionality of the output vector, 4 in this patent,
and
the k-dimension values of the actual output vector and the theoretical output vector are respectively.
And 3.4, updating the weight coefficient and the bias coefficient in the BP network by using a random gradient descent algorithm, and repeating the steps 3.1-3.3 until a set convergence condition is reached or a set iteration number is reached.
And 4, online identification of the model: and (3) installing the improved BP neural network obtained by training in the steps 1-3 in an upper computer, and carrying out state recognition on the acoustic emission signals uploaded by the AE signal acquisition system in real time.
The online identification of the model in step 4 is specifically described as follows:
installing the improved BP neural network obtained by training in the steps 1-3 in an upper computer, and carrying out state recognition on the acoustic emission signals uploaded by the AE signal acquisition system in real time, wherein the four states of the recognized concrete are respectively as follows: normal state, initial compaction state, crack propagation state, and through failure state.
FIG. 2 is a block diagram of an improved BP neural network used in the present invention. It can be clearly seen from the figure that for the concrete signal acquired by the acoustic emission acquisition system, the IMF component is extracted by using the EMD algorithm, in order to highlight the features contained in the IMF obtained after EMD decomposition as much as possible, the complex domain energy value index fusing the time-frequency domain features is provided, the features of BP neural network input data are enhanced, and the identification accuracy of the model is improved; then, an NLFU activation function is provided in a hidden layer, so that the problem that the BP neural network activation function is easy to have gradient disappearance is solved, and the convergence of the model is accelerated; in addition, an average weighted Min-type distance loss function is provided for quantifying the error between the actual output and the theoretical output of the model, and the classification and identification precision of the model is improved. The model can accurately and effectively identify four different concrete states (dry, air-dry, saturated surface dry and wet) and improve the construction efficiency.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, but any modifications or equivalent variations made according to the technical spirit of the present invention are within the scope of the present invention as claimed.
Claims (4)
1. A concrete state recognition model based on an improved BP neural network comprises the following specific steps:
step 1, acoustic emission signal acquisition: collecting concrete signals in different states by using an acoustic emission collection system, wherein a data collection card selects an NI USB-6366 high-speed card, and simultaneously selects LABVIEW software to build an AE signal collection system;
step 2, feature vector composition: decomposing the acoustic emission signal by using an EMD algorithm to obtain an IMF component, then calculating a complex domain energy value corresponding to the IMF component, and forming a feature vector;
step 3, improved BP network training: training the improved BP neural network by using the characteristic vector generated in the step 2 until the network converges;
and 4, online identification of the model: and (3) installing the improved BP neural network obtained by training in the steps 1-3 in an upper computer, and carrying out state recognition on the acoustic emission signals uploaded by the AE signal acquisition system in real time.
2. The concrete state recognition model based on the improved BP neural network as claimed in claim 1, wherein: in step 2, after obtaining the IMF by EMD decomposition, the corresponding complex domain energy value is solved, and the specific steps of forming the characteristic vector are as follows:
step 2.1, calculating the mean value of the upper envelope curve and the lower envelope curve of the acoustic emission signal x (t), updating the signal in a difference value iteration mode to continuously obtain an IMF component, finally judging whether iteration is terminated according to a standard deviation sd, stopping iteration when sd is 0.2-0.3, wherein the expression of sd is as follows:
wherein T represents the length of the signal, hk-1(t) and hk(t) IMF components obtained by the k-1 st and k-th decompositions respectively;
step 2.2, calculating complex domain energy values ce of each IMF component, wherein the specific calculation expression of ce is as follows:
wherein F (i) represents the value of the i-th frequency domain point of x (t), and F (·) represents the Fourier transform;
and 2.3, combining the complex domain energy values corresponding to the IMF obtained by calculation in the step 2.2 into a feature vector, setting the dimensionality of the feature vector to be 12, if the total number of the IMFs obtained by decomposition is less than 12, filling zero at the tail of the feature vector to enable the dimensionality to be 12, and if the total number of the IMFs obtained by decomposition is more than 12, removing redundant dimensionalities.
3. The concrete state recognition model based on the improved BP neural network as claimed in claim 1, wherein: the specific steps of training the improved BP neural network in the step 3 are as follows:
step 3.1, building a BP neural network, wherein the structure is as follows: the input layer-hidden layer 1-hidden layer 2-output layer, the specific numerical value is 12-8-6-4;
step 3.2, inputting the feature vector obtained in step 2 into a BP neural network, and processing in the hidden layer 1 and the hidden layer 2 by using a proposed Nonlinear fusion unit (NLFU) activation function, wherein the NLFU has an expression as follows:
step 3.3, solving the loss function of the current iteration according to the output of the output layer and the value of the theoretical label, and the patent provides an average weighted Min-type distance dmawThe expression is as follows:
where S represents the number of weights, set to 5 in this patent, n is the dimensionality of the output vector, 4 in this patent,
and
the k-dimension values of the actual output vector and the theoretical output vector are respectively;
and 3.4, updating the weight coefficient and the bias coefficient in the BP network by using a random gradient descent algorithm, and repeating the steps 3.1-3.3 until a set convergence condition is reached or a set iteration number is reached.
4. The concrete state recognition model based on the improved BP neural network as claimed in claim 1, wherein: the online identification of the model in step 4 is specifically described as follows:
installing the improved BP neural network obtained by training in the steps 1-3 in an upper computer, and carrying out state recognition on the acoustic emission signals uploaded by the AE signal acquisition system in real time, wherein the four states of the recognized concrete are respectively as follows: normal state, initial compaction state, crack propagation state, and through failure state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110879316.2A CN113657217A (en) | 2021-08-02 | 2021-08-02 | Concrete state recognition model based on improved BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110879316.2A CN113657217A (en) | 2021-08-02 | 2021-08-02 | Concrete state recognition model based on improved BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113657217A true CN113657217A (en) | 2021-11-16 |
Family
ID=78490200
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110879316.2A Withdrawn CN113657217A (en) | 2021-08-02 | 2021-08-02 | Concrete state recognition model based on improved BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113657217A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114896467A (en) * | 2022-04-24 | 2022-08-12 | 北京月新时代科技股份有限公司 | Neural network-based field matching method and intelligent data entry method |
| CN115255405A (en) * | 2022-09-23 | 2022-11-01 | 相国新材料科技江苏有限公司 | Intelligent control method and system of additive manufacturing equipment |
| CN115655887A (en) * | 2022-11-01 | 2023-01-31 | 广东建设职业技术学院 | Concrete strength prediction method |
-
2021
- 2021-08-02 CN CN202110879316.2A patent/CN113657217A/en not_active Withdrawn
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114896467A (en) * | 2022-04-24 | 2022-08-12 | 北京月新时代科技股份有限公司 | Neural network-based field matching method and intelligent data entry method |
| CN114896467B (en) * | 2022-04-24 | 2024-02-09 | 北京月新时代科技股份有限公司 | Neural network-based field matching method and data intelligent input method |
| CN115255405A (en) * | 2022-09-23 | 2022-11-01 | 相国新材料科技江苏有限公司 | Intelligent control method and system of additive manufacturing equipment |
| CN115655887A (en) * | 2022-11-01 | 2023-01-31 | 广东建设职业技术学院 | Concrete strength prediction method |
| CN115655887B (en) * | 2022-11-01 | 2023-04-21 | 广东建设职业技术学院 | Concrete strength prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113657217A (en) | Concrete state recognition model based on improved BP neural network | |
| CN107941537B (en) | A kind of mechanical equipment health state evaluation method | |
| CN108984893B (en) | Gradient lifting method-based trend prediction method | |
| CN110781592B (en) | Method and system for predicting residual life of equipment | |
| CN110879253B (en) | Steel rail crack acoustic emission signal detection method based on improved long-time and short-time memory network | |
| CN111505643B (en) | Sea surface small target detection method based on time-frequency image deep learning | |
| CN110060368B (en) | Mechanical anomaly detection method based on potential feature codes | |
| CN110209999A (en) | A kind of mobile unit failure trend prediction method | |
| CN110568082A (en) | cable wire breakage distinguishing method based on acoustic emission signals | |
| CN116625683A (en) | Wind turbine generator system bearing fault identification method, system and device and electronic equipment | |
| Afrasiabi et al. | Two-stage deep learning-based wind turbine condition monitoring using SCADA data | |
| CN108592812A (en) | Fan blade optical fiber load strain characteristics extract and crack monitoring method | |
| CN109408998B (en) | Fatigue life evaluation method by rapidly acquiring stress spectrum based on sample continuous increment | |
| CN113112188B (en) | Power dispatching monitoring data anomaly detection method based on pre-screening dynamic integration | |
| CN112986393B (en) | Bridge inhaul cable damage detection method and system | |
| CN113627268A (en) | Model training method, and method and device for detecting fault of speed reducer for mine hoist | |
| CN113987905A (en) | Escalator braking force intelligent diagnosis system based on deep belief network | |
| CN110530631A (en) | A kind of gear list type fault detection method based on hybrid classifer | |
| CN116434258A (en) | Automatic identification method for form data | |
| CN115859148A (en) | Fan blade vibration alarm method and device | |
| Chase et al. | Detection of Damage in OperatingWind Turbines by Signature Distances | |
| CN110598674B (en) | Point switch fault identification method based on pattern identification | |
| CN113919396A (en) | Vibration signal and image characteristic machine tool cutter wear state monitoring method based on semi-supervised learning | |
| Li et al. | Methods for the rapid detection of boundary condition variations in structural systems | |
| Zhang et al. | Structural health monitoring of offshore wind turbine based on online data-driven support vector machine |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20211116 |