CN111062079A - Bridge probability damage detection method based on autoregressive model and Gaussian process - Google Patents
Bridge probability damage detection method based on autoregressive model and Gaussian process Download PDFInfo
- Publication number
- CN111062079A CN111062079A CN201911316968.4A CN201911316968A CN111062079A CN 111062079 A CN111062079 A CN 111062079A CN 201911316968 A CN201911316968 A CN 201911316968A CN 111062079 A CN111062079 A CN 111062079A
- Authority
- CN
- China
- Prior art keywords
- damage
- gaussian process
- bridge
- autoregressive model
- training sample
- 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
Abstract
The invention provides a bridge probability damage detection method based on an autoregressive model and a Gaussian process, which can process response data of an acceleration sensor through a Gaussian process classifier and a Gaussian process regression machine to obtain damage position and damage degree information; the beneficial technical effects of the invention are as follows: the scheme is realized without depending on the undamaged state information and the external excitation information of the bridge, is low in realization difficulty, can identify the damage position and the damage degree at the same time, and can remove unreliable results.
Description
Technical Field
The invention relates to a bridge damage detection method, in particular to a bridge probability damage detection method based on an autoregressive model and a Gaussian process.
Background
By 2018, the number of highway bridges and railway bridges in China exceeds one million, and with the increase of service life, the internal condition of a bridge structure is continuously deteriorated, so that the bearing capacity of the structure is continuously reduced, and the normal use performance of the bridge structure during operation is seriously damaged. The method realizes the damage positioning and damage degree measurement of the bridge structure by corresponding means, and is an important basis for bridge management and maintenance.
Bridge impairment identification includes four levels of content: firstly, damage is identified; secondly, positioning the damage; thirdly, identifying the damage degree; and fourthly, predicting the residual life of the structure. Due to the influence of environmental factors, the problems that the multiple damage states are difficult to identify and the reliability of the prediction result is difficult to judge exist in the field of bridge damage identification at present.
Disclosure of Invention
Aiming at the problems in the background art, the invention provides a bridge probability damage detection method based on an autoregressive model and a Gaussian process, which is innovative in that: the bridge probability damage detection method comprises the following steps:
1) establishing a simulation model of the actual bridge by adopting finite element software, wherein the simulation model is provided with a plurality of simulation acceleration sensors, and the number, the serial number and the positions of the simulation acceleration sensors correspond to the number, the serial number and the positions of the acceleration sensors on the actual bridge one by one;
2) designing a plurality of damage schemes; the individual injury protocols included: arranging a single or a plurality of damaged parts on the simulation model, and adjusting the structural elastic modulus of the damaged parts on the simulation model; recording the position of the damaged part on the simulation model as damaged position information, and recording the structural elastic modulus as damaged degree information;
3) under the condition of a single damage scheme, response data of all simulated acceleration sensors are obtained to obtain a sensor data set; each damage scheme corresponds to a sensor data set;
4) carrying out standardization processing on the plurality of sensor data sets, and establishing an autoregressive model according to a standardization processing result;
5) extracting an autoregressive model residual error standard deviation of each simulated acceleration sensor according to the autoregressive model; constructing damage characteristic parameters DSF for corresponding simulation acceleration sensors according to autoregressive model residual error standard deviationiI is 1,2,3, …, d, d is the number of simulated acceleration sensors;
DSFi=σεi
wherein, DSFiThe damage characteristic parameters corresponding to the ith simulation acceleration sensor are obtained; sigmaεiThe standard deviation of the autoregressive model residual error of the ith simulation acceleration sensor is obtained;
according to DSFiAnd constructing structural damage position information L1 and structural damage state information L2:
according to DSFiL1 and L2, and constructing a classification algorithm feature vector RSD1 and a regression algorithm feature vector RSD 2:
RSD1=(σε1,σε2,…,σεi,L1,L2)
RSD2=(σε1,σε2,…,σεi)
establishing a first training sample set; a plurality of RSDs 1 corresponding to different damage schemes are used as a plurality of input samples in a first training sample set, and a plurality of damage position information in one-to-one correspondence with the plurality of RSDs 1 are used as a plurality of output samples in the first training sample set; establishing a second training sample set; a plurality of RSDs 2 corresponding to different damage schemes are used as a plurality of input samples in a second training sample set, and a plurality of damage degree information in one-to-one correspondence with the plurality of RSDs 2 are used as a plurality of output samples in the second training sample set;
6) establishing a Gaussian process classifier according to a Gaussian process classification algorithm, and establishing a Gaussian process regression machine according to a Gaussian process regression algorithm; using the RSD1 as an input vector, using the damage position information as output, and training a Gaussian process classifier by using a first training sample set; using the RSD2 as an input vector, using the damage degree information as output, and training a Gaussian process regression machine by using a second training sample set;
7) putting the trained Gaussian process classifier and the Gaussian process regression into use; response data of all acceleration sensors on an actual bridge are periodically acquired, and an actual sensor data set is obtained; constructing a first eigenvector RSD11 and a second eigenvector RSD22 according to the autoregressive model obtained in the step 4) and the actual sensor data set;
RSD11=(σ′ε1,σ′ε2,…,σ′εi,L1′,L2′)
RSD22=(σ′ε1,σ′ε2,…,σ′εi)
wherein the content of the first and second substances,
wherein, DSF'iThe damage characteristic parameters corresponding to the ith acceleration sensor on the actual bridge are obtained; DSF'i=σ′εi;σ′εiThe standard deviation of the autoregressive model residual error of the ith acceleration sensor on the actual bridge is obtained;
inputting the RSD11 as an input vector into a Gaussian process classifier, and identifying a damage position according to an output result of the Gaussian process classifier; inputting the RSD22 as an input vector into a Gaussian process regression machine, and then judging whether the output result of the Gaussian process regression machine is reliable according to the following formula:
wherein u is the mean value of the Gaussian distribution obeyed by the output result of the Gaussian process regression machine; sigma is the standard deviation of Gaussian distribution obeyed by the output result of the Gaussian process regression machine;
and if the output result of the Gaussian process regression machine is reliable, identifying the damage degree information according to the output result of the Gaussian process regression machine.
The principle of the invention is as follows: firstly, the scheme of the invention does not need bridge undamaged state information and external excitation information and is based on the damage characteristic parameter DSF 'extracted from the time domain'iDamage identification can be carried out, and the realization difficulty is small; secondly, L1 and L2 (and corresponding L1 'and L2') were introduced in the present invention, wherein L1 has the meaning: regarding each number i as a discrete random variable, and taking the DSF corresponding to each number i as a discrete random variableiIf the numerical value of (a) is the corresponding probability, then L1 is equivalent to the mean value of a plurality of numbers i, and thus L1 can reflect the information related to the damage position; the significance of L2 is: the L2 plays a clustering role, the probability is similar to the normalization, under different damage conditions, the numerical values after the probability normalization have differences, and thus information related to the damage degree can be embodied through the L2; originally, the inventor also introduces L1 and L2 into RSD2, and then experiments prove that the treatment effect of a Gaussian process regression machine is better when L1 and L2 are not introduced, so that L1 and L2 are not introduced into RSD2 in the final scheme; in addition, the invention also introduces a limit lambda for judging the reliability of the output result, and the significance of the limit lambda is as follows: when the condition of ≦ λ is established, the output result has a 95% possibility of falling in the interval [ u-1.96 σ, u +1.96 σ ≦]And when u is the output result, the deviation will not exceed 5%, which is acceptable in engineering and is considered reliable, otherwise it is considered unreliable, such asTherefore, unreliable output results can be eliminated.
The beneficial technical effects of the invention are as follows: the scheme is realized without depending on the undamaged state information and the external excitation information of the bridge, is low in realization difficulty, can identify the damage position and the damage degree at the same time, and can remove unreliable results.
Detailed Description
A bridge probability damage detection method based on an autoregressive model and a Gaussian process is innovative in that: the bridge probability damage detection method comprises the following steps:
1) establishing a simulation model of the actual bridge by adopting finite element software, wherein the simulation model is provided with a plurality of simulation acceleration sensors, and the number, the serial number and the positions of the simulation acceleration sensors correspond to the number, the serial number and the positions of the acceleration sensors on the actual bridge one by one;
2) designing a plurality of damage schemes; the individual injury protocols included: arranging a single or a plurality of damaged parts on the simulation model, and adjusting the structural elastic modulus of the damaged parts on the simulation model; recording the position of the damaged part on the simulation model as damaged position information, and recording the structural elastic modulus as damaged degree information;
3) under the condition of a single damage scheme, response data of all simulated acceleration sensors are obtained to obtain a sensor data set; each damage scheme corresponds to a sensor data set;
4) carrying out standardization processing on the plurality of sensor data sets, and establishing an autoregressive model according to a standardization processing result;
5) extracting an autoregressive model residual error standard deviation of each simulated acceleration sensor according to the autoregressive model; constructing damage characteristic parameters DSF for corresponding simulation acceleration sensors according to autoregressive model residual error standard deviationiI is 1,2,3, …, d, d is the number of simulated acceleration sensors;
DSFi=σεi
wherein, DSFiIs an ith simulationDamage characteristic parameters corresponding to the true acceleration sensor; sigmaεiThe standard deviation of the autoregressive model residual error of the ith simulation acceleration sensor is obtained;
according to DSFiAnd constructing structural damage position information L1 and structural damage state information L2:
according to DSFiL1 and L2, and constructing a classification algorithm feature vector RSD1 and a regression algorithm feature vector RSD 2:
RSD1=(σε1,σε2,…,σεi,L1,L2)
RSD2=(σε1,σε2,…,σεi)
establishing a first training sample set; a plurality of RSDs 1 corresponding to different damage schemes are used as a plurality of input samples in a first training sample set, and a plurality of damage position information in one-to-one correspondence with the plurality of RSDs 1 are used as a plurality of output samples in the first training sample set; establishing a second training sample set; a plurality of RSDs 2 corresponding to different damage schemes are used as a plurality of input samples in a second training sample set, and a plurality of damage degree information in one-to-one correspondence with the plurality of RSDs 2 are used as a plurality of output samples in the second training sample set;
6) establishing a Gaussian process classifier according to a Gaussian process classification algorithm, and establishing a Gaussian process regression machine according to a Gaussian process regression algorithm; using the RSD1 as an input vector, using the damage position information as output, and training a Gaussian process classifier by using a first training sample set; using the RSD2 as an input vector, using the damage degree information as output, and training a Gaussian process regression machine by using a second training sample set;
7) putting the trained Gaussian process classifier and the Gaussian process regression into use; response data of all acceleration sensors on an actual bridge are periodically acquired, and an actual sensor data set is obtained; constructing a first eigenvector RSD11 and a second eigenvector RSD22 according to the autoregressive model obtained in the step 4) and the actual sensor data set;
RSD11=(σ′ε1,σ′ε2,…,σ′εi,L1′,L2′)
RSD22=(σ′ε1,σ′ε2,…,σ′εi)
wherein the content of the first and second substances,
wherein, DSFi ′The damage characteristic parameters corresponding to the ith acceleration sensor on the actual bridge are obtained; DSF'i=σ′εi;σ′εiThe standard deviation of the autoregressive model residual error of the ith acceleration sensor on the actual bridge is obtained;
inputting the RSD11 as an input vector into a Gaussian process classifier, and identifying a damage position according to an output result of the Gaussian process classifier; inputting the RSD22 as an input vector into a Gaussian process regression machine, and then judging whether the output result of the Gaussian process regression machine is reliable according to the following formula:
wherein u is the mean value of the Gaussian distribution obeyed by the output result of the Gaussian process regression machine; sigma is the standard deviation of Gaussian distribution obeyed by the output result of the Gaussian process regression machine;
and if the output result of the Gaussian process regression machine is reliable, identifying the damage degree information according to the output result of the Gaussian process regression machine.
Claims (1)
1. A bridge probability damage detection method based on an autoregressive model and a Gaussian process is characterized by comprising the following steps: the bridge probability damage detection method comprises the following steps:
1) establishing a simulation model of the actual bridge by adopting finite element software, wherein the simulation model is provided with a plurality of simulation acceleration sensors, and the number, the serial number and the positions of the simulation acceleration sensors correspond to the number, the serial number and the positions of the acceleration sensors on the actual bridge one by one;
2) designing a plurality of damage schemes; the individual injury protocols included: arranging a single or a plurality of damaged parts on the simulation model, and adjusting the structural elastic modulus of the damaged parts on the simulation model; recording the position of the damaged part on the simulation model as damaged position information, and recording the structural elastic modulus as damaged degree information;
3) under the condition of a single damage scheme, response data of all simulated acceleration sensors are obtained to obtain a sensor data set; each damage scheme corresponds to a sensor data set;
4) carrying out standardization processing on the plurality of sensor data sets, and establishing an autoregressive model according to a standardization processing result;
5) extracting an autoregressive model residual error standard deviation of each simulated acceleration sensor according to the autoregressive model; constructing damage characteristic parameters DSF for corresponding simulation acceleration sensors according to autoregressive model residual error standard deviationiI is 1,2,3, …, d, d is the number of simulated acceleration sensors;
DSFi=σεi
wherein, DSFiThe damage characteristic parameters corresponding to the ith simulation acceleration sensor are obtained; sigmaεiThe standard deviation of the autoregressive model residual error of the ith simulation acceleration sensor is obtained;
according to DSFiAnd constructing structural damage position information L1 and structural damage state information L2:
according to DSFiL1 and L2, and constructing a classification algorithm feature vector RSD1 and a regression algorithm feature vector RSD 2:
RSD1=(σεσ,σε2,…,σεi,L1,L2)
RSD2=(σε1,σε2,…,σεi)
establishing a first training sample set; a plurality of RSDs 1 corresponding to different damage schemes are used as a plurality of input samples in a first training sample set, and a plurality of damage position information in one-to-one correspondence with the plurality of RSDs 1 are used as a plurality of output samples in the first training sample set; establishing a second training sample set; a plurality of RSDs 2 corresponding to different damage schemes are used as a plurality of input samples in a second training sample set, and a plurality of damage degree information in one-to-one correspondence with the plurality of RSDs 2 are used as a plurality of output samples in the second training sample set;
6) establishing a Gaussian process classifier according to a Gaussian process classification algorithm, and establishing a Gaussian process regression machine according to a Gaussian process regression algorithm; using the RSD1 as an input vector, using the damage position information as output, and training a Gaussian process classifier by using a first training sample set; using the RSD2 as an input vector, using the damage degree information as output, and training a Gaussian process regression machine by using a second training sample set;
7) putting the trained Gaussian process classifier and the Gaussian process regression into use; response data of all acceleration sensors on an actual bridge are periodically acquired, and an actual sensor data set is obtained; constructing a first eigenvector RSD11 and a second eigenvector RSD22 according to the autoregressive model obtained in the step 4) and the actual sensor data set;
RSD11=(σ′ε1,σ′ε2,…,σ′εi,L1′,L2′)
RSD22=(σ′ε1,σ′ε2,…,σ′εi)
wherein the content of the first and second substances,
wherein, DSF'iThe damage characteristic parameters corresponding to the ith acceleration sensor on the actual bridge are obtained; DSF'i=σ′εi;σ′εiThe standard deviation of the autoregressive model residual error of the ith acceleration sensor on the actual bridge is obtained;
inputting the RSD11 as an input vector into a Gaussian process classifier, and identifying a damage position according to an output result of the Gaussian process classifier; inputting the RSD22 as an input vector into a Gaussian process regression machine, and then judging whether the output result of the Gaussian process regression machine is reliable according to the following formula:
wherein u is the mean value of the Gaussian distribution obeyed by the output result of the Gaussian process regression machine; sigma is the standard deviation of Gaussian distribution obeyed by the output result of the Gaussian process regression machine;
and if the output result of the Gaussian process regression machine is reliable, identifying the damage degree information according to the output result of the Gaussian process regression machine.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911316968.4A CN111062079B (en) | 2019-12-19 | 2019-12-19 | Bridge probability damage detection method based on autoregressive model and Gaussian process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911316968.4A CN111062079B (en) | 2019-12-19 | 2019-12-19 | Bridge probability damage detection method based on autoregressive model and Gaussian process |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111062079A true CN111062079A (en) | 2020-04-24 |
CN111062079B CN111062079B (en) | 2022-03-08 |
Family
ID=70302321
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911316968.4A Active CN111062079B (en) | 2019-12-19 | 2019-12-19 | Bridge probability damage detection method based on autoregressive model and Gaussian process |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111062079B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112966435A (en) * | 2021-03-01 | 2021-06-15 | 重庆交通大学 | Bridge deformation real-time prediction method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104677666A (en) * | 2015-03-18 | 2015-06-03 | 西安公路研究院 | Continuous rigid frame bridge prestress damage identification method based on deflection monitoring |
CN104881567A (en) * | 2015-04-24 | 2015-09-02 | 浙江工业大学 | Statistical model based bridge health monitoring data wavelet denoising method |
US20150247940A1 (en) * | 2014-03-03 | 2015-09-03 | King Abdullah University Of Science And Technology | History matching of time-lapse crosswell data using ensemble kalman filtering |
CN106556498A (en) * | 2016-10-31 | 2017-04-05 | 石家庄铁道大学 | Damage Identification Methods for Bridge Structures and system |
CN106910186A (en) * | 2017-01-13 | 2017-06-30 | 陕西师范大学 | A kind of Bridge Crack detection localization method based on CNN deep learnings |
CN110147781A (en) * | 2019-05-29 | 2019-08-20 | 重庆交通大学 | Bridge vibration mode based on machine learning visualizes damnification recognition method |
-
2019
- 2019-12-19 CN CN201911316968.4A patent/CN111062079B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150247940A1 (en) * | 2014-03-03 | 2015-09-03 | King Abdullah University Of Science And Technology | History matching of time-lapse crosswell data using ensemble kalman filtering |
CN104677666A (en) * | 2015-03-18 | 2015-06-03 | 西安公路研究院 | Continuous rigid frame bridge prestress damage identification method based on deflection monitoring |
CN104881567A (en) * | 2015-04-24 | 2015-09-02 | 浙江工业大学 | Statistical model based bridge health monitoring data wavelet denoising method |
CN106556498A (en) * | 2016-10-31 | 2017-04-05 | 石家庄铁道大学 | Damage Identification Methods for Bridge Structures and system |
CN106910186A (en) * | 2017-01-13 | 2017-06-30 | 陕西师范大学 | A kind of Bridge Crack detection localization method based on CNN deep learnings |
CN110147781A (en) * | 2019-05-29 | 2019-08-20 | 重庆交通大学 | Bridge vibration mode based on machine learning visualizes damnification recognition method |
Non-Patent Citations (3)
Title |
---|
SHILEI ZHANG: "Damage Detection in Structures Using Artificial Neural Networks", 《2010 INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND COMPUTATIONAL INTELLIGENCE》 * |
付雷: "复杂系统在桥梁挠度变化异常检测中的应用探究", 《四川建材》 * |
李冬雪: "桥梁高墩及高塔表观损伤检测技术研究", 《CNKI优秀硕士学位论文全文库 工程科技Ⅱ辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112966435A (en) * | 2021-03-01 | 2021-06-15 | 重庆交通大学 | Bridge deformation real-time prediction method |
CN112966435B (en) * | 2021-03-01 | 2022-04-01 | 重庆交通大学 | Bridge deformation real-time prediction method |
Also Published As
Publication number | Publication date |
---|---|
CN111062079B (en) | 2022-03-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112254964A (en) | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network | |
CN110365647B (en) | False data injection attack detection method based on PCA and BP neural network | |
CN111191740B (en) | Fault diagnosis method for rolling bearing | |
CN111436944B (en) | Falling detection method based on intelligent mobile terminal | |
CN113567130A (en) | Bearing fault diagnosis method based on multiple working conditions of equipment | |
CN107577945A (en) | URL attack detection methods, device and electronic equipment | |
CN111505424A (en) | Large experimental device power equipment fault diagnosis method based on deep convolutional neural network | |
CN108363902B (en) | Accurate prediction method for pathogenic genetic variation | |
CN112193959A (en) | Method and system for detecting abnormal sound of elevator | |
CN111062079B (en) | Bridge probability damage detection method based on autoregressive model and Gaussian process | |
CN105675038A (en) | Device for predicting faults of instruments | |
CN107682317B (en) | method for establishing data detection model, data detection method and equipment | |
CN110706213A (en) | Bridge cluster structure damage judgment method based on strain response cumulative distribution function difference | |
CN113179263A (en) | Network intrusion detection method, device and equipment | |
CN111678699B (en) | Early fault monitoring and diagnosing method and system for rolling bearing | |
CN110568082A (en) | cable wire breakage distinguishing method based on acoustic emission signals | |
CN114563150A (en) | Bridge health online detection module generation method, detection method, tool box and device | |
CN114545234A (en) | Method for monitoring battery state based on battery temperature gradient change | |
CN115327417A (en) | Early warning method and system for abnormity of power battery monomer and electronic equipment | |
CN111832442B (en) | Method for automatically separating temperature strain components from massive bridge dynamic strain data | |
CN109840386B (en) | Damage identification method based on factor analysis | |
CN112966345B (en) | Rotary machine residual life prediction hybrid shrinkage method based on countertraining and transfer learning | |
CN115508706A (en) | Lithium battery residual life prediction method considering capacity regeneration phenomenon | |
CN115618263A (en) | Feature-level SMOTE-based complex equipment unbalance fault diagnosis method | |
CN110532512B (en) | Equipment failure mode diagnosis characteristic parameter analysis method based on big data analysis |
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 | ||
GR01 | Patent grant |