CN112986393A - Bridge inhaul cable damage detection method and system - Google Patents

Bridge inhaul cable damage detection method and system Download PDF

Info

Publication number
CN112986393A
CN112986393A CN202110196642.3A CN202110196642A CN112986393A CN 112986393 A CN112986393 A CN 112986393A CN 202110196642 A CN202110196642 A CN 202110196642A CN 112986393 A CN112986393 A CN 112986393A
Authority
CN
China
Prior art keywords
acoustic emission
emission signal
model
damage
trained
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
Application number
CN202110196642.3A
Other languages
Chinese (zh)
Other versions
CN112986393B (en
Inventor
吴启明
姜瑞娟
郭宗明
乐颖
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Municipal Design and Research Institute Co Ltd
Original Assignee
Shenzhen Municipal Design and Research Institute Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Municipal Design and Research Institute Co Ltd filed Critical Shenzhen Municipal Design and Research Institute Co Ltd
Priority to CN202110196642.3A priority Critical patent/CN112986393B/en
Publication of CN112986393A publication Critical patent/CN112986393A/en
Application granted granted Critical
Publication of CN112986393B publication Critical patent/CN112986393B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/14—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object using acoustic emission techniques
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4463—Signal correction, e.g. distance amplitude correction [DAC], distance gain size [DGS], noise filtering
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4481—Neural networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/023—Solids
    • G01N2291/0234—Metals, e.g. steel

Abstract

The invention relates to a method and a system for detecting damage of a bridge inhaul cable, wherein the method comprises the following steps: acquiring an acoustic emission signal set of a bridge cable to be detected, and inputting the acoustic emission signal set into a first screening model to obtain a first judgment result; and when the first judgment result shows that the damage is generated, acquiring an acoustic emission signal of the bridge cable to be detected in the current time period, and inputting the acoustic emission signal of the bridge cable to be detected in the current time period into a second screening model to obtain the detection damage type of the bridge cable to be detected in the current time period. The invention realizes the automatic and intelligent all-weather real-time online detection of the damage of the inhaul cable and improves the detection precision.

Description

Bridge inhaul cable damage detection method and system
Technical Field
The invention relates to the field of bridge cable damage detection, in particular to a method and a system for detecting bridge cable damage.
Background
In a cable weighing system bridge, the function of the stay cable is to transmit constant and live loads from a bridge deck to a main arch rib, a cable tower and a main cable, and belongs to one of the most critical stressed components. The inhaul cable is limited by the structural specificity of the member and the current anchor head anticorrosion technology and construction process level, has short actual service life and larger discreteness, and can not reach the design service life of 100 years required by the standard of the bridge engineering industry. And the damaged part of the inhaul cable is positioned in the lower anchoring area, and the position is hidden, so that the inhaul cable belongs to a detection blind area, and most of the existing detection and monitoring technologies have no substantial effect.
The existing inhaul cable broken wire monitoring technology based on acoustic emission judges through stress waves emitted when the inhaul cable is damaged. The method has the advantages that even if the stay cable is not damaged, sound can be generated under the action of dynamic load or environmental noise, so that errors exist in detection results, actually monitored acoustic emission signals are continuous time sequences, the generated data are also massive, it is not practical to obviously judge whether damage occurs or not by means of manual work, and even if the judgment is carried out by means of manual work, the difficulty of feature extraction can be increased under the action of environmental excitation by means of currently and generally adopted parameter analysis methods based on features such as energy counting, ringing counting and the like, so that wrong judgment is generated. On the other hand, the mode recognition technology based on deep learning such as RNN and CNN may have better performance, but has the disadvantages that model training requires a large number of signal data samples, the acquisition and labeling of the samples are very large engineering quantities, the cost in practical application is very high, the manual labeling is not accurate due to the influence of the field environment, and the final detection precision is also inaccurate.
Disclosure of Invention
The invention aims to provide a method and a system for detecting damage of a bridge inhaul cable, which can realize automatic and intelligent all-weather real-time online monitoring and detection of the damage of the inhaul cable and improve the detection precision.
In order to achieve the purpose, the invention provides the following scheme:
a detection method for damage of a bridge inhaul cable comprises the following steps:
acquiring an acoustic emission signal set of a bridge cable to be detected, wherein the acoustic emission signal set comprises an acoustic emission signal at the time t, an acoustic emission signal at the time t-1 and an acoustic emission signal at the time t-2; wherein t > 2;
inputting the acoustic emission signal set of the bridge cable to be detected into a first screening model to obtain a first judgment result; the first screening model is determined by a model parameter and an acoustic emission signal set of the bridge cable to be detected; the model parameters are obtained by inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a second-order Markov chain model and solving by adopting a maximum likelihood method, and the model parameters are parameters in the second-order Markov chain model;
when the first judgment result is that damage is generated, acquiring an acoustic emission signal of the bridge cable to be detected in the current time period, and inputting the acoustic emission signal of the bridge cable to be detected in the current time period into a second screening model to obtain a detection probability set of the bridge cable to be detected in the current time period; the current time period is a continuous time period comprising a first time period, a time t and a second time period, wherein the first time period is before the time t, and the second time period is after the time t; the second screening model is obtained by training the sequentially connected encoder and the multi-type logistic regression model by taking the acoustic emission signal of the bridge cable to be trained as input and taking the real damage type corresponding to the acoustic emission signal of the bridge cable to be trained as a label; the detection probability set comprises shearing damage probability, tension damage probability and damage-free probability;
processing the detection probability set by adopting a cost sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period; the detection damage types comprise shearing damage, tension damage and no damage.
Optionally, the determining method of the first screening model includes:
acquiring an acoustic emission signal set of a bridge cable to be trained;
inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a first formula, and solving by adopting a maximum likelihood method to obtain model parameters; the first formula is
p(x(t)|x(t-1)，x(t-2))＝N(x(t)|c1x(t-1)+c2x(t-2)+c3，σ2)t≥3
p(x(t)|x(t-1))＝N(x(t)|c1x(t-1)+c3，σ2)t＝2
Determining a first screening model according to the model parameters and the acoustic emission signal set of the bridge cable to be detected; the first screening model is
Figure BDA0002947033960000021
Wherein, p (x)(t)|x(t-1)，x(t-2)) The transition probability distribution of a second order Markov chain when t is greater than or equal to 3, p (x)(t)|t(t-1)) Transition probability distribution of the second order Markov chain when t is 2, x(t)Is the acoustic emission signal at time t, x(t-1)Is the acoustic emission signal at time t-1, x(t-2)Is the acoustic emission signal at time t-2, c1First linear coefficient being the mean value in the transition probability distribution, c2Second linear coefficient being the mean value in the transition probability distribution, c3Third linear coefficient, σ, being the mean in the transition probability distribution2α is a control detection standard coefficient for variance of transition probability distribution.
Optionally, the determining method of the second screening model includes:
acquiring an acoustic emission signal of a bridge cable to be trained, and dividing the acoustic emission signal of the bridge cable to be trained into a plurality of acoustic emission signal segments according to a set time period;
dividing all acoustic emission signal segments into a training set, a verification set and a test set according to a set proportion;
taking the training set as the input of an encoder, taking a feature vector set output by the encoder as the input of a decoder, performing iterative training on the encoder and the decoder, and determining the encoder with the smallest loss function as the trained encoder, wherein the loss function is determined according to the training set, the feature vector set output by the encoder and a reconstruction vector output by the decoder;
inputting the training set into the trained encoder to obtain a trained feature vector set;
training a plurality of types of logistic regression models by taking the trained feature vector set as input and the real damage types corresponding to the training set as labels to obtain the trained plurality of types of logistic regression models;
sequentially connecting the trained encoder and the trained multi-type logistic regression models to obtain a second model;
determining the hyper-parameters of the second model by using the verification set to obtain a trained second model;
and evaluating the generalization ability of the trained second model by using the test set, and if the generalization ability reaches a set threshold, determining the trained second model as a second screening model.
Optionally, the loss function is:
Figure BDA0002947033960000031
wherein x is an acoustic emission signal of the bridge cable to be trained in the training set, and hnFor the nth element in the feature vector set output by the encoder, λ is a regularization parameter greater than zero, g (h) is a reconstruction vector output by the decoder, and L (x, g (h)) is a reconstruction error.
Optionally, the processing the detection probability set by using a cost sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period specifically includes: according to the formula
Figure BDA0002947033960000032
Obtaining the detection damage type of the bridge guy cable to be detected in the current time period, wherein p iskFor the probability of detecting the damage type k in the probability set, CkiAnd detecting the economic loss when the actual damage type of the stay is i and the damage type is k.
A bridge cable damage detection system comprises:
the acquisition module is used for acquiring an acoustic emission signal set of the bridge cable to be detected, wherein the acoustic emission signal set comprises an acoustic emission signal at the time t, an acoustic emission signal at the time t-1 and an acoustic emission signal at the time t-2; wherein t > 2;
the first judgment module is used for inputting the acoustic emission signal set of the bridge cable to be detected into a first screening model to obtain a first judgment result; the first screening model is determined by a model parameter and an acoustic emission signal set of the bridge cable to be detected; the model parameters are obtained by inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a second-order Markov chain model and solving by adopting a maximum likelihood method, and the model parameters are parameters in the second-order Markov chain model;
the probability set determining module is used for acquiring an acoustic emission signal of the bridge cable to be detected in the current time period when the first judgment result indicates that the damage is generated, and inputting the acoustic emission signal of the bridge cable to be detected in the current time period into a second screening model to obtain a detection probability set of the bridge cable to be detected in the current time period; the current time period is a continuous time period comprising a first time period, a time t and a second time period, wherein the first time period is before the time t, and the second time period is after the time t; the second screening model is obtained by training the sequentially connected encoder and the multi-type logistic regression model by taking the acoustic emission signal of the bridge cable to be trained as input and taking the real damage type corresponding to the acoustic emission signal of the bridge cable to be trained as a label; the detection probability set comprises shearing damage probability, tension damage probability and damage-free probability;
the second judgment module is used for processing the detection probability set by adopting a cost sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period; the detection damage types comprise shearing damage, tension damage and no damage.
Optionally, the system for detecting damage to the bridge cable further includes: a first screening model determination module; the first screening model determining module specifically includes:
the first acquisition unit is used for acquiring an acoustic emission signal set of a bridge cable to be trained;
the parameter determining unit is used for inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a first formula and solving by adopting a maximum likelihood method to obtain model parameters; the first formula is
p(x(t)|x(t-1)，x(t-2))＝N(x(t)|c1x(t-1)+c2x(t-2)+c3，σ2)t≥3
p(x(t)|x(t-1))＝N(x(t)|c1x(t-1)+c3，σ2)t＝2
The first screening model determining unit is used for determining a first screening model according to the model parameters and the acoustic emission signal set of the bridge cable to be detected; the first screening model is
Figure BDA0002947033960000051
Wherein, p (x)(t)|x(t-1)，x(t-2)) The transition probability distribution of a second order Markov chain when t is greater than or equal to 3, p (x)(t)|x(t-1)) Transition probability distribution of the second order Markov chain when t is 2, x(t)Is the acoustic emission signal at time t, x(t-1)Is the acoustic emission signal at time t-1, x(t-2)Is the acoustic emission signal at time t-2, c1First linear coefficient being the mean value in the transition probability distribution, c2Second linear coefficient being the mean value in the transition probability distribution, c3Third linear coefficient, σ, being the mean in the transition probability distribution2α is a control detection standard coefficient for variance of transition probability distribution.
Optionally, the system for detecting damage to the bridge cable further includes: a second screening model determination module; the second screening model determining module specifically includes:
the second acquisition unit is used for acquiring the acoustic emission signal of the bridge cable to be trained and dividing the acoustic emission signal of the bridge cable to be trained into a plurality of acoustic emission signal segments according to a set time period;
the set determining unit is used for dividing all the acoustic emission signal fragments into a training set, a verification set and a test set according to a set proportion;
an encoder determining unit, configured to use the training set as an input of an encoder, use a feature vector set output by the encoder as an input of a decoder, perform iterative training on the encoder and the decoder, and determine an encoder with a smallest loss function as a trained encoder, where the loss function is determined according to the training set, a feature vector set output by the encoder, and a reconstruction vector output by the decoder;
a vector set determining unit, configured to input the training set into the trained encoder to obtain a trained feature vector set;
the regression model determining unit is used for training the multi-class logistic regression model by taking the trained feature vector set as input and the real damage type corresponding to the training set as a label to obtain the trained multi-class logistic regression model;
the second model determining unit is used for sequentially connecting the trained encoder and the trained multi-class logistic regression models to obtain a second model;
and the optimized second model determining unit is used for determining the hyper-parameters of the second model by using the verification set to obtain the trained second model.
And the second screening model determining unit is used for evaluating the generalization ability of the trained second model by using the test set, and if the generalization ability reaches a set threshold, determining the trained second model as a second screening model.
Optionally, the loss function is:
Figure BDA0002947033960000061
wherein x is an acoustic emission signal of the bridge cable to be trained in the training set, and hnFor the nth element in the feature vector set output by the encoder, λ is a regularization parameter greater than zero, g (h) is a reconstruction vector output by the decoder, and L (x, g (h)) is a reconstruction error.
Optionally, the second determining module includes: a second judgment unit for judging whether the first and second judgment units are in accordance with the formula
Figure BDA0002947033960000062
Obtaining the detection damage type of the bridge guy cable to be detected in the current time period, wherein p iskFor the probability of detecting the damage type k in the probability set, CkiAnd detecting the economic loss when the actual damage type of the stay is i and the damage type is k.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the invention uses the first screening model to carry out pre-screening, and then carries out secondary detection through the second screening model consisting of the encoder and the multi-type logistic regression model, thereby realizing the automatic and intelligent all-weather real-time online detection of the damage of the inhaul cable and improving the detection precision.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of a method for detecting damage to a bridge cable according to embodiment 1 of the present invention;
fig. 2 is a flowchart of a more specific method for detecting damage to a bridge cable according to embodiment 1 of the present invention;
FIG. 3 is a schematic diagram of a second-order Markov chain pre-screening model discrimination process provided in embodiment 1 of the present invention;
FIG. 4 is a connection diagram of guy cable acoustic emission signal data streams provided in embodiment 1 of the present invention;
FIG. 5 is a schematic diagram of a network architecture of a sparse self-encoder and a sparse self-decoder according to embodiment 1 of the present invention;
FIG. 6 is a flowchart illustrating a process of self-encoder and multi-class logistic regression model training according to embodiment 1 of the present invention;
FIG. 7 is a signal diagram of an acoustic emission signal sample when a tension damage is generated in the guy cable provided in embodiment 1 of the present invention;
fig. 8 is a structural block diagram of a system for detecting damage to a bridge cable according to embodiment 2 of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Example 1
The present embodiments relate to artificial intelligence and machine learning techniques, designed based on neural networks, representation learning, and transfer learning.
Machine learning is a multi-field cross discipline and relates to a plurality of disciplines such as probability theory, matrix theory, statistics, approximation theory, convex optimization, functional analysis, game theory, differential manifold and the like. The method aims to research how to improve the performance of the system by an algorithm and experience, so that the system has the capability of acquiring knowledge by itself, and is the most promising approach for realizing artificial intelligence at present.
Representation learning exploits features by machine learning and based on tasks, rather than determining appropriate features by human beings. Features learned through machine learning tend to perform better than manual determinations. Most conventional machine learning algorithms require manually determining features and then mapping the features to an output. But for many tasks it is difficult to know which features should be extracted. Using representation learning in this case can solve this problem. The encoder provided by the embodiment of the invention belongs to a typical representation learning algorithm.
The transfer learning utilizes what has been learned in one scenario to improve generalization in another scenario. In migration learning, the learner performs two or more different tasks, but assumes that these tasks share some low-level concept or factor pool in the data, so that by learning a good feature space for data distribution in one task, for other scenarios, only a small number of labeled samples are needed to train and generalize well.
For better understanding, this embodiment provides a detection method of bridge cable damage. As shown in fig. 1, the method includes:
101: acquiring an acoustic emission signal set of a bridge cable to be detected, wherein the acoustic emission signal set comprises an acoustic emission signal at the time t, an acoustic emission signal at the time t-1 and an acoustic emission signal at the time t-2; wherein t > 2.
102: inputting the acoustic emission signal set of the bridge cable to be detected into a first screening model to obtain a first judgment result; the first screening model is determined by a model parameter and an acoustic emission signal set of the bridge cable to be detected; the model parameters are obtained by inputting the acoustic emission signal set of the bridge cable to be trained into a second-order Markov chain model and solving by adopting a maximum likelihood method, and the model parameters are parameters in the second-order Markov chain model.
103: when the first judgment result is that damage is generated, acquiring an acoustic emission signal of the bridge cable to be detected in the current time period, and inputting the acoustic emission signal of the bridge cable to be detected in the current time period into a second screening model to obtain a detection probability set of the bridge cable to be detected in the current time period; the current time period is a continuous time period comprising a first time period, a time t and a second time period, wherein the first time period is before the time t, and the second time period is after the time t; the second screening model is obtained by training the sequentially connected encoder and the multi-type logistic regression model by taking the acoustic emission signal of the bridge cable to be trained as input and taking the real damage type corresponding to the acoustic emission signal of the bridge cable to be trained as a label; the detection probability set comprises shearing damage probability, tension damage probability and damage-free probability;
104: processing the detection probability set by adopting a cost sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period; the detection damage types comprise shearing damage, tension damage and no damage.
In practical application, the method for determining the first screening model comprises the following steps:
and acquiring an acoustic emission signal set of the bridge cable to be trained.
Inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a first formula, and solving by adopting a maximum likelihood method to obtain model parameters; the first formula is
p(x(t)|x(t-1)，x(t-2))＝N(x(t)|c1x(t-1)+c2x(t-2)+c3，σ2)t≥3
p(x(t)|x(t-1))＝N(x(t)|c1x(t-1)+c3，σ2)t＝2
Determining a first screening model according to the model parameters and the acoustic emission signal set of the bridge cable to be detected; the first screening model is
Figure BDA0002947033960000091
Wherein, p (x)(t)|x(t-1)，x(t-2)) The transition probability distribution of a second order Markov chain when t is greater than or equal to 3, p (x)(t)|x(t-1)) When t is 2, the second-order Mark mayTransition probability distribution of the Fugu chain, x(t)Is the acoustic emission signal at time t, x(t-1)Is the acoustic emission signal at time t-1, x(t-2)Is the acoustic emission signal at time t-2, c1First linear coefficient being the mean value in the transition probability distribution, c2Second linear coefficient being the mean value in the transition probability distribution, c3Third linear coefficient, σ, being the mean in the transition probability distribution2α is a control detection standard coefficient for variance of transition probability distribution.
In practical application, the second screening model is determined by the following method:
the method comprises the steps of obtaining an acoustic emission signal of a bridge cable to be trained, and dividing the acoustic emission signal of the bridge cable to be trained into a plurality of acoustic emission signal segments according to a set time period.
And dividing all the acoustic emission signal segments into a training set, a verification set and a test set according to a set proportion.
And taking the training set as the input of an encoder, taking the feature vector set output by the encoder as the input of a decoder, performing iterative training on the encoder and the decoder, and determining the encoder with the smallest loss function as the trained encoder, wherein the loss function is determined according to the training set, the feature vector set output by the encoder and the reconstructed vector output by the decoder.
And inputting the training set into the trained coder to obtain a trained feature vector set.
And training the multi-class logistic regression model by taking the trained feature vector set as input and the real damage type corresponding to the training set as a label to obtain the trained multi-class logistic regression model.
And connecting the trained encoder and the trained multi-class logistic regression models in sequence to obtain a second model.
And determining the hyper-parameters of the second model by using the verification set to obtain the trained second model. Specifically, the verification set is used as input, and the real damage type corresponding to the verification set is used as output to optimize the hyper-parameters of the second model to obtain the optimized second model.
And evaluating the generalization ability of the trained second model by using the test set, and if the generalization ability reaches a set threshold, determining the trained second model as a second screening model.
In practical applications, the loss function is:
Figure BDA0002947033960000092
wherein x is an acoustic emission signal of the bridge cable to be trained in the training set, and hnFor the nth element in the feature vector set output by the encoder, λ is a regularization parameter greater than zero, g (h) is a reconstruction vector output by the decoder, and L (x, g (h)) is a reconstruction error.
In practical application, the detection probability set is processed by a cost sensitive learning method to obtain the detection damage types of the bridge cable to be detected in the current time period as follows: according to the formula
Figure BDA0002947033960000101
Obtaining the detection damage type of the bridge guy cable to be detected in the current time period, wherein p iskFor the probability of detecting the damage type k in the probability set, CkiAnd detecting the economic loss when the actual damage type of the stay is i and the damage type is k.
In this embodiment, as shown in fig. 2, a first screening model obtained by using a second-order markov chain probability model is used for pre-screening, and then a second detection is performed by using an encoder-multi-class logistic regression model, where the specific process is as follows:
(1) a second-order markov chain pre-screening model (a first screening model) monitors each data in the acoustic emission signal data stream collected by the inhaul cable acoustic emission sensor and the collecting device in real time on line, and preliminarily judges whether the data at the current moment belong to the damage abnormal signal or not, as shown in fig. 3, according to the data value x of the previous two moments(t-1),x(t-2)Judging the data value x of the current time according to the following formula(t)Whether the abnormality is detected:
Figure BDA0002947033960000102
in the formula, α is a coefficient greater than 0 for controlling the detection standard, and according to the performance metric principle, the parameter α is preferably selected to have a smaller value, and meanwhile, in practical application, the performance metric can be controlled by adjusting the value of α: increasing alpha, improving precision ratio and reducing recall ratio; reducing alpha reduces precision ratio and improves recall ratio. The default value 1 can be selected at the initial stage of model operation, and the abnormity judgment standard of the second-order Markov chain pre-screening model is based on the performance measurement principle of high recall ratio and low recall ratio, namely, the missed diagnosis ratio is reduced at the cost of higher misdiagnosis ratio.
(2) And if the acoustic emission signal data flow is abnormal, intercepting data in a period of time before and after the current abnormal data in the acoustic emission signal data flow of the inhaul cable. Specifically, all data between the ring count start point and end point may be taken.
(3) Inputting the intercepted data into a trained encoder, encoding abnormal data by the encoder, and outputting a feature vector h with a fixed length by the encoder. The feature vector is equivalent to encoding and dimensionality reduction of a data stream (from the mathematical point of view, the feature vector belongs to nonlinear mapping from a vector to a vector; from the machine learning point of view, the feature vector is extracted by an encoder and can represent the key features of the internal structure of data).
(4) And inputting the feature vector h into the trained multi-class logistic regression model, and carrying out secondary detection judgment on the multi-class logistic regression model according to the feature vector h. The multiclass logistic regression model outputs the final decision ("no damage" or "shear damage" or "tension damage").
And rescaling the softmax function output of the multi-class logistic regression model based on cost-sensitive learning and a loss matrix C during prediction, wherein elements in the loss matrix C represent equivalent economic loss caused by the fact that the output class of the classifier does not accord with the actual class. The output categories of the classifier are:
Figure BDA0002947033960000111
pkprobability of class k being output as softmax function, CkiFor the economic loss caused when the actual result is the category i but the classifier output is the category k, the subscripts i and k take values of 1, 2 and 3, which respectively represent "no damage", "shear damage", "tension damage": and (4) calculating the equivalent economic loss according to the cost of opening the stay cable anchor head for detection, the cost of replacing the stay cable, the influence of the broken stay cable on the structure safety and the indirect economic loss caused by the broken stay cable.
The loss matrix is specifically defined as: c21: the acoustic emission signal of the stay cable is actually caused by 'environmental excitation', but the classifier judges that the equivalent economic loss is caused by 'generation of shearing damage'. The cost of manual inspection by opening the anchor head is 10 ten thousand due to misjudgment, so C 2110 ten thousand.
C31: the acoustic emission signal of the stay cable is actually caused by 'environmental excitation', but the classifier judges that the equivalent economy is caused by 'generation of tension damage'. Because of misdiagnosis, the cost for manual detection by opening the anchor head is 10 ten thousand, so C 3110 ten thousand.
C12: the acoustic emission signal of the stay cable is actually caused by 'shear damage', but the classifier judges that the equivalent economic loss is caused by 'environmental excitation'. No early warning is given after the stay cable is broken due to 'missed diagnosis', the influence of the stay cable on the structural safety and the subsequent maintenance and reinforcement cost are considered, C1240 ten thousand.
C13: the acoustic emission signal of the stay cable is actually caused by 'generation of tension damage', but the classifier judges that the equivalent economic loss is caused by 'no damage'. No early warning is given after the stay cable is broken due to 'missed diagnosis', the influence of the stay cable on the structural safety and the subsequent maintenance and reinforcement cost are considered, C1350 ten thousand.
C32: the acoustic emission signal of the stay cable is actually caused by 'shear damage', but the classifier judges that the equivalent economic loss is caused by 'tension damage'.The loss caused by this type of misdiagnosis is low, C320.5 ten thousand.
C23: the acoustic emission signal of the stay cable is actually caused by 'generation of tensile damage', but the classifier judges that the equivalent economic loss is caused by 'generation of shear damage'. The loss caused by this type of misdiagnosis is low, C 231 ten thousand.
The following describes the training process of the model applied in the above method:
1, establishing and training a second-order Markov chain pre-screening model by adopting the following method:
(1) and (3) mounting an acoustic emission sensor on a new cable on the bridge site for training data of the model, and acquiring continuous data for 1-2 months. The method comprises the steps of taking a stay cable acoustic emission signal data stream acquired at a certain frequency as a random variable sequence { x }(1)，x(2)，…，x(t)… }. As shown in FIG. 4, each variable x in the sequence(t)The variable x to which t is 1 and 2 … are connected in a directional connection manner(t+1)、x(t+2)The variable at a certain time is dependent on the variables at the previous two times or the previous time, and the dependence is expressed by a linear Gaussian distribution: the linear gaussian respective model includes:
p(x(t)|x(t-1)，x(t-2))＝N(x(t)|c1x(t-1)+c2x(t-2)+c3，σ2)t≥3
p(x(t)|x(t-1))＝N(x(t)|c1x(t-1)+c3，σ2)t＝2
the initial state probability adopts a Gaussian model:
Figure BDA0002947033960000121
p(x(1)) For an initial state distribution of a second order Markov chain, μ0Is the mean value of the distribution of the initial states,
Figure BDA0002947033960000122
is the variance of the initial state distribution.
(2) Inputting the collected training data into a linear Gaussian distribution model, and solving by adopting a maximum likelihood method to obtain parameters
Figure BDA0002947033960000123
(3) The resulting parameters are substituted into the following equation:
Figure BDA0002947033960000124
a second order markov chain pre-screening model is obtained, where α is pre-set and adjustable.
The encoder 2 is a feedforward neural network and has the following characteristics:
as shown in fig. 5, the encoder portion of the sparse encoder employs local concatenation and parameter sharing to learn local statistical features of the data and adapt to variable dimensional input data, where the first layer employs an architecture of width 3 and step 1, and the concatenation weights at the same position between each layer and the width are equal. The last layer of the encoder part is a pooling layer with the width of 3 and the step of 1, a feature vector h with the dimension of 5 is output by adopting a maximum pooling function, input data with different dimensions are adapted by adopting down sampling and changing a sampling range, and meanwhile, a feature vector with the fixed length of a cable acoustic emission signal is output. All detectors of the hidden layer use a rectifying linear activation function. The decoder part of the sparse encoder is a fully connected network, which is used to reconstruct the eigenvector h as the vector g (h), which should be close to x when the eigenvector h is the main characteristic of the input data x.
The 3 types of logistic regression models are linear classifiers, and have the following characteristics after cost sensitive learning:
(1) using the feature vector output by the encoder as input, and outputting the probabilities (p) of the categories { 'no damage', 'shear damage', 'tension damage' } through a softmax function1，p2，p3}。
(2) After rescaling softmax output based on the cost sensitive learning and the loss matrix C, the output categories of the classifier are as follows:
Figure BDA0002947033960000125
the classifier is used for solving the classification problem, and the judgment process comprises two steps: 1) outputting a result by adopting a softmax function, wherein the result is a vector, and elements in the vector are the probability of each category; 2) the existing method directly takes the category with the maximum probability value as a judgment result. However, this model is particularly useful in that cost sensitive learning is used, and particularly useful in that the second step differs from the above: the loss matrix C is used for rescaling the judgment result, and the final output category is
Figure BDA0002947033960000131
4, a joint training method of an encoder and multiple types of logistic regression models is shown in fig. 6, and training data of the encoder and the multiple types of logistic regression models are obtained by the following method:
(1) the training data are cable acoustic emission signals, including acoustic emission signals of the cable under environmental excitation and abnormal acoustic emission signals when the cable is damaged. The acoustic emission signal of the inhaul cable under the environmental excitation is acquired after an acoustic emission sensor is installed on the inhaul cable of a newly-built bridge or a newly-replaced inhaul cable, then the part with signal response in the acquired data is extracted through an existing matched software system or a manual mode, and the part is divided into a section of independent data samples, and the samples are marked as 'no damage'. And obtaining an abnormal acoustic emission signal data sample when the cable is damaged through a cable stretching experiment. Specifically, before a tensile test, the cable damage is artificially manufactured, during the tensile test, the cable damage gradually develops until the cable damage is broken, the acoustic emission signals are continuously acquired in the process, and after the acquisition is finished, the data sample is marked as 'shear damage generation' or 'tension damage generation' according to the type of the damage. Fig. 7 is a sample of an acoustic emission signal labeled "produce tension damage" obtained by a guy cable tension experiment. In order to obtain a larger data set, an acoustic emission sensor is installed on a guy cable operated for more than ten years on the existing bridge and effective data of 1-2 months are continuously collected, the data of the part have no label and are used for unsupervised training of an encoder, and in order to avoid the problem of category imbalance, an undersampling method is adopted, so that the number of the data of three categories is close to that of the data of the three categories.
(2) All of the above including labeled and unlabeled data are used as the data set for training the encoder. All the labeled data are used as a data set for training a multi-class logistic regression model, and the data are mapped to a feature space through an encoder during training.
(3) And preprocessing samples in the data set before training of the sparse encoder. Specifically, all samples are adjusted to the same dimension, wherein the dimension of the samples with smaller dimension is expanded by zero padding, and the dimension of the samples with larger dimension is reduced by resampling.
(4) The sparse encoder adopts an unsupervised layer-by-layer pre-training method. The training process includes two parts of an encoder and a decoder, and only the encoder is needed in prediction and test. Specifically, the first layer is trained, then the output of the trained first layer network is used as the input of the second layer network and the second layer network is trained, and so on until all layers are trained (the loss function value is minimum). The sparse encoder is trained by adopting a random gradient descent method, and a loss function consists of a reconstruction error and a sparse penalty: l (x, g (h)) + lambda sigmam|hmL, wherein x is the stay cable acoustic emission signal data stream, h is the feature vector output by the encoder, hmThe m element of h, lambda is a regularization parameter larger than zero, which can be adjusted between 0.001 and 0.1, and g (h) is a reconstruction vector output by a decoder. And finally, taking the trained parameters as initial values, and then training the whole network. After the model is trained, the decoder portion is removed and the encoder portion is left for encoding the input data.
(5) A data set formed by labeled data is divided into a training set, a verification set and a test set according to the ratio of 6:2:2, then the training set, the verification set and the test set are mapped to a feature space through an encoder, training, verification and testing of a multi-type logistic regression model are all carried out on the feature space, when the deep framework of the encoder is trained, adjustment and optimization are carried out on the basis of the performance of the coded feature vector and the multi-type logistic regression model in the verification set, and the encoder and the multi-type logistic regression model framework with the optimal performance are selected.
(6) The generalization ability of the entire encoder-multi-class logistic regression model was evaluated through the test set.
(7) In practical application, the inhaul cable acoustic emission signal data structure learned by the encoder can generalize a model to other bridge projects similar to inhaul cable bearing systems in a transfer learning mode, so that the marking cost and the experiment cost of data are obviously reduced.
Example 2
The embodiment also provides a computer program product comprising a bridge cable damage detection system, wherein the product comprises a computer storage medium, and computer executable instructions are stored in the storage medium and used for realizing the bridge cable damage detection method. The method for detecting damage to a bridge cable provided in this embodiment may also be implemented in the form of a program product, where the program product includes program codes, and when the program product runs on a computer device, the program codes are used to make the computer device execute the method for detecting damage to a bridge cable described above, for example, the computer device may execute the flow of the method for detecting damage to a bridge cable shown in fig. 2. The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. Readable storage media include, but are not limited to: a U disk, a removable hard disk, a magnetic disk, an optical disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The bridge cable damage detection system that this embodiment provided specifically includes as shown in fig. 8:
the acquisition module A1 is used for acquiring an acoustic emission signal set of the bridge cable to be detected, wherein the acoustic emission signal set comprises an acoustic emission signal at the time t, an acoustic emission signal at the time t-1 and an acoustic emission signal at the time t-2; wherein t > 2.
The first judgment module A2 is used for inputting the acoustic emission signal set of the bridge cable to be detected into a first screening model to obtain a first judgment result; the first screening model is determined by a model parameter and an acoustic emission signal set of the bridge cable to be detected; the model parameters are obtained by inputting the acoustic emission signal set of the bridge cable to be trained into a second-order Markov chain model and solving by adopting a maximum likelihood method, and the model parameters are parameters in the second-order Markov chain model.
A probability set determining module A3, configured to, when the first determination result is that a damage is generated, obtain an acoustic emission signal of the bridge cable to be detected in the current time period, and input the acoustic emission signal of the bridge cable to be detected in the current time period into a second screening model to obtain a detection probability set of the bridge cable to be detected in the current time period; the current time period is a continuous time period comprising a first time period, a time t and a second time period, wherein the first time period is before the time t, and the second time period is after the time t; the second screening model is obtained by training the sequentially connected encoder and the multi-type logistic regression model by taking the acoustic emission signal of the bridge cable to be trained as input and taking the real damage type corresponding to the acoustic emission signal of the bridge cable to be trained as a label; the detection probability set comprises shearing damage probability, tension damage probability and damage-free probability;
the second judgment module A4 is used for processing the detection probability set by adopting a cost sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period; the detection damage types comprise shearing damage, tension damage and no damage.
As an optional implementation, the method further includes: a first screening model determination module; the first screening model determining module specifically includes:
the first acquisition unit is used for acquiring an acoustic emission signal set of the bridge inhaul cable to be trained.
The parameter determining unit is used for inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a first formula and solving by adopting a maximum likelihood method to obtain model parameters; the first formula is
p(x(t)|x(t-1)，x(t-2))＝N(x(t)|c1x(t-1)+c2x(t-2)+c3，σ2)t≥3
p(x(t)|x(t-1))＝N(x(t)|c1x(t-1)+c3，σ2)t＝2
The first screening model determining unit is used for determining a first screening model according to the model parameters and the acoustic emission signal set of the bridge cable to be detected; the first screening model is
Figure BDA0002947033960000161
Wherein, p (x)(t)|x(t-1)，x(t-2)) The transition probability distribution of a second order Markov chain when t is greater than or equal to 3, p (x)(t)|x(t-1)) Transition probability distribution of the second order Markov chain when t is 2, x(t)Is the acoustic emission signal at time t, x(t-1)Is the acoustic emission signal at time t-1, x(t-2)Is the acoustic emission signal at time t-2, c1First linear coefficient being the mean value in the transition probability distribution, c2Second linear coefficient being the mean value in the transition probability distribution, c3Third linear coefficient, σ, being the mean in the transition probability distribution2α is a control detection standard coefficient for variance of transition probability distribution.
As an optional implementation, the method further includes: a second screening model determination module; the second screening model determining module specifically includes:
and the second acquisition unit is used for acquiring the acoustic emission signal of the bridge cable to be trained and dividing the acoustic emission signal of the bridge cable to be trained into a plurality of acoustic emission signal segments according to a set time period.
And the set determining unit is used for dividing all the acoustic emission signal segments into a training set, a verification set and a test set according to a set proportion.
And the encoder determining unit is used for taking the training set as the input of an encoder, taking the feature vector set output by the encoder as the input of a decoder, performing iterative training on the encoder and the decoder, and determining the encoder with the smallest loss function as the trained encoder, wherein the loss function is determined according to the training set, the feature vector set output by the encoder and the reconstructed vector output by the decoder.
And the vector set determining unit is used for inputting the training set into the trained encoder to obtain a trained characteristic vector set.
And the regression model determining unit is used for training the multi-class logistic regression model by taking the trained feature vector set as input and the real damage type corresponding to the training set as a label to obtain the trained multi-class logistic regression model.
And the second model determining unit is used for sequentially connecting the trained encoder and the trained multi-class logistic regression models to obtain a second model.
And the optimized second model determining unit is used for determining the hyper-parameters of the second model by using the verification set to obtain the trained second model.
And the second screening model determining unit is used for evaluating the generalization ability of the trained second model by using the test set, and if the generalization ability reaches a set threshold, determining the trained second model as a second screening model.
As an alternative embodiment, the loss function is:
Figure BDA0002947033960000171
wherein x is an acoustic emission signal of the bridge cable to be trained in the training set, and hnFor the nth element in the feature vector set output by the encoder, λ is a regularization parameter greater than zero, g (h) is a reconstruction vector output by the decoder, and L (x, g (h)) is a reconstruction error.
As an optional implementation, the second determining module includes: first, theA second judging unit for judging the formula
Figure BDA0002947033960000172
Obtaining the detection damage type of the bridge guy cable to be detected in the current time period, wherein p iskFor the probability of detecting the damage type k in the probability set, CkiAnd detecting the economic loss when the actual damage type of the stay is i and the damage type is k.
The invention has the following beneficial effects:
(1) automatic and intelligent all-weather real-time online monitoring and detection of the stay cable are realized through machine learning and artificial intelligence, the detection precision is improved, and the problem that massive stay cable data generated in the bridge operation process are difficult to judge through manpower is solved.
(2) The characteristics of the acoustic emission signals of the inhaul cable are learned through a machine learning algorithm, different conditions of the inhaul cable can be effectively and accurately identified, and therefore whether the inhaul cable is damaged or not and the type of the damage are judged. The problem that a traditional parameter-based analysis method is easily interfered by environmental noise is solved.
(3) The intrinsic data structure and characteristics of the guy cable learned by the encoder can be applied to other bridge scenes of a cable weighing system in a transfer learning mode, so that the data labeling and experiment cost are obviously reduced.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (10)

1. A method for detecting damage of a bridge inhaul cable is characterized by comprising the following steps:
acquiring an acoustic emission signal set of a bridge cable to be detected, wherein the acoustic emission signal set comprises an acoustic emission signal at the time t, an acoustic emission signal at the time t-1 and an acoustic emission signal at the time t-2; wherein t is greater than 2;
inputting the acoustic emission signal set of the bridge cable to be detected into a first screening model to obtain a first judgment result; the first screening model is determined by a model parameter and an acoustic emission signal set of the bridge cable to be detected; the model parameters are obtained by inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a second-order Markov chain model and solving by adopting a maximum likelihood method, and the model parameters are parameters in the second-order Markov chain model;
when the first judgment result is that damage is generated, acquiring an acoustic emission signal of the bridge cable to be detected in the current time period, and inputting the acoustic emission signal of the bridge cable to be detected in the current time period into a second screening model to obtain a detection probability set of the bridge cable to be detected in the current time period; the current time period is a continuous time period comprising a first time period, a time t and a second time period, wherein the first time period is before the time t, and the second time period is after the time t; the second screening model is obtained by training the sequentially connected encoder and the multi-type logistic regression model by taking the acoustic emission signal of the bridge cable to be trained as input and taking the real damage type corresponding to the acoustic emission signal of the bridge cable to be trained as a label; the detection probability set comprises shearing damage probability, tension damage probability and damage-free probability;
processing the detection probability set by adopting a cost sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period; the detection damage types comprise shearing damage, tension damage and no damage.
2. The method for detecting the damage to the bridge cable according to claim 1, wherein the first screening model is determined by:
acquiring an acoustic emission signal set of a bridge cable to be trained;
inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a first formula, and solving by adopting a maximum likelihood method to obtain model parameters; the first formula is
p(x(t)|x(t-1)，x(t-2))＝N(x(t)|c1x(t-1)+c2x(t-2)+c3，σ2)t≥3
p(x(t)|x(t-1))＝N(x(t)|c1x(t-1)+c3，σ2)t＝2
Determining a first screening model according to the model parameters and the acoustic emission signal set of the bridge cable to be detected; the first screening model is
Figure FDA0002947033950000021
Wherein, p (x)(t)|x(t-1)，x(t-2)) The transition probability distribution of a second order Markov chain when t is greater than or equal to 3, p (x)(t)|x(t-1)) Transition probability distribution of the second order Markov chain when t is 2, x(t)Is the acoustic emission signal at time t, x(t-1)Is the acoustic emission signal at time t-1, x(t-2)Is the acoustic emission signal at time t-2, c1First linear coefficient being the mean value in the transition probability distribution, c2Second linear coefficient being the mean value in the transition probability distribution, c3Third linear coefficient, σ, being the mean in the transition probability distribution2α is a control detection standard coefficient for variance of transition probability distribution.
3. The method for detecting the damage to the bridge cable according to claim 1, wherein the second screening model is determined by:
acquiring an acoustic emission signal of a bridge cable to be trained, and dividing the acoustic emission signal of the bridge cable to be trained into a plurality of acoustic emission signal segments according to a set time period;
dividing all acoustic emission signal segments into a training set, a verification set and a test set according to a set proportion;
taking the training set as the input of an encoder, taking a feature vector set output by the encoder as the input of a decoder, performing iterative training on the encoder and the decoder, and determining the encoder with the smallest loss function as the trained encoder, wherein the loss function is determined according to the training set, the feature vector set output by the encoder and a reconstruction vector output by the decoder;
inputting the training set into the trained encoder to obtain a trained feature vector set;
training a plurality of types of logistic regression models by taking the trained feature vector set as input and the real damage types corresponding to the training set as labels to obtain the trained plurality of types of logistic regression models;
sequentially connecting the trained encoder and the trained multi-type logistic regression models to obtain a second model;
determining the hyper-parameters of the second model by using the verification set to obtain a trained second model;
and evaluating the generalization ability of the trained second model by using the test set, and if the generalization ability reaches a set threshold, determining the trained second model as a second screening model.
4. The method for detecting damage to a bridge cable according to claim 3, wherein the loss function is:
Figure FDA0002947033950000022
wherein x is the training set to be trainedAcoustic emission signal of bridge cable, hnFor the nth element in the feature vector set output by the encoder, λ is a regularization parameter greater than zero, g (h) is a reconstruction vector output by the decoder, and L (x, g (h)) is a reconstruction error.
5. The method for detecting damage to the bridge cable according to claim 1, wherein the step of processing the detection probability set by using a cost-sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period specifically comprises the steps of: according to the formula
Figure FDA0002947033950000031
Obtaining the detection damage type of the bridge guy cable to be detected in the current time period, wherein p iskFor the probability of detecting the damage type k in the probability set, CkiAnd detecting the economic loss when the actual damage type of the stay is i and the damage type is k.
6. The utility model provides a detection system of bridge cable damage which characterized in that includes:
the acquisition module is used for acquiring an acoustic emission signal set of the bridge cable to be detected, wherein the acoustic emission signal set comprises an acoustic emission signal at the time t, an acoustic emission signal at the time t-1 and an acoustic emission signal at the time t-2; wherein t > 2;
the first judgment module is used for inputting the acoustic emission signal set of the bridge cable to be detected into a first screening model to obtain a first judgment result; the first screening model is determined by a model parameter and an acoustic emission signal set of the bridge cable to be detected; the model parameters are obtained by inputting the acoustic emission signal set of the bridge cable to be trained into a second-order Markov chain model and solving by adopting a maximum likelihood method, and the model parameters are parameters in the second-order Markov chain model.
The probability set determining module is used for acquiring an acoustic emission signal of the bridge cable to be detected in the current time period when the first judgment result indicates that the damage is generated, and inputting the acoustic emission signal of the bridge cable to be detected in the current time period into a second screening model to obtain a detection probability set of the bridge cable to be detected in the current time period; the current time period is a continuous time period comprising a first time period, a time t and a second time period, wherein the first time period is before the time t, and the second time period is after the time t; the second screening model is obtained by training the sequentially connected encoder and the multi-type logistic regression model by taking the acoustic emission signal of the bridge cable to be trained as input and taking the real damage type corresponding to the acoustic emission signal of the bridge cable to be trained as a label; the detection probability set comprises shearing damage probability, tension damage probability and damage-free probability;
the second judgment module is used for processing the detection probability set by adopting a cost sensitive learning method to obtain the detection damage type of the bridge cable to be detected in the current time period; the detection damage types comprise shearing damage, tension damage and no damage.
7. The system for detecting damage to a bridge cable of claim 6, further comprising: a first screening model determination module; the first screening model determining module specifically includes:
the first acquisition unit is used for acquiring an acoustic emission signal set of a bridge cable to be trained;
the parameter determining unit is used for inputting the acoustic emission signal set of the bridge inhaul cable to be trained into a first formula and solving by adopting a maximum likelihood method to obtain model parameters; the first formula is
p(x(t)|x(t-1)，x(t-2))＝N(x(t)|c1x(t-1)+c2x(t-2)+c3，σ2)t≥3
p(x(t)|x(t-1))＝N(x(t)|c1x(t-1)+c3，σ2)t＝2
The first screening model determining unit is used for determining a first screening model according to the model parameters and the acoustic emission signal set of the bridge cable to be detected; the first screening model is
Figure FDA0002947033950000041
Wherein, p (x)(t)|x(t-1)，x(t-2)) The transition probability distribution of a second order Markov chain when t is greater than or equal to 3, p (x)(t)|x(t-1)) Transition probability distribution of the second order Markov chain when t is 2, x(t)Is the acoustic emission signal at time t, x(t-1)Is the acoustic emission signal at time t-1, x(t-2)Is the acoustic emission signal at time t-2, c1First linear coefficient being the mean value in the transition probability distribution, c2Second linear coefficient being the mean value in the transition probability distribution, c3Third linear coefficient, σ, being the mean in the transition probability distribution2α is a control detection standard coefficient for variance of transition probability distribution.
8. The system for detecting damage to a bridge cable of claim 6, further comprising: a second screening model determination module; the second screening model determining module specifically includes:
the second acquisition unit is used for acquiring the acoustic emission signal of the bridge cable to be trained and dividing the acoustic emission signal of the bridge cable to be trained into a plurality of acoustic emission signal segments according to a set time period;
the set determining unit is used for dividing all the acoustic emission signal fragments into a training set, a verification set and a test set according to a set proportion;
an encoder determining unit, configured to use the training set as an input of an encoder, use a feature vector set output by the encoder as an input of a decoder, perform iterative training on the encoder and the decoder, and determine an encoder with a smallest loss function as a trained encoder, where the loss function is determined according to the training set, a feature vector set output by the encoder, and a reconstruction vector output by the decoder;
a vector set determining unit, configured to input the training set into the trained encoder to obtain a trained feature vector set;
the regression model determining unit is used for training the multi-class logistic regression model by taking the trained feature vector set as input and the real damage type corresponding to the training set as a label to obtain the trained multi-class logistic regression model;
the second model determining unit is used for sequentially connecting the trained encoder and the trained multi-class logistic regression models to obtain a second model;
the optimized second model determining unit is used for determining the hyper-parameters of the second model by using the verification set to obtain a trained second model;
and the second screening model determining unit is used for evaluating the generalization ability of the trained second model by using the test set, and if the generalization ability reaches a set threshold, determining the trained second model as a second screening model.
9. The bridge cable damage detection system of claim 8, wherein the loss function is:
Figure FDA0002947033950000051
wherein x is an acoustic emission signal of the bridge cable to be trained in the training set, and hnFor the nth element in the feature vector set output by the encoder, λ is a regularization parameter greater than zero, g (h) is a reconstruction vector output by the decoder, and L (x, g (h)) is a reconstruction error.
10. The system for detecting damage to a bridge cable according to claim 6, wherein the second determination module comprises: a second judgment unit for judging whether the first and second judgment units are in accordance with the formula
Figure FDA0002947033950000052
To obtain a bridge to be detectedDetecting damage type of the beam stay cable in the current time period, wherein pkFor the probability of detecting the damage type k in the probability set, CkiAnd detecting the economic loss when the actual damage type of the stay is i and the damage type is k.
CN202110196642.3A 2021-02-22 2021-02-22 Bridge inhaul cable damage detection method and system Active CN112986393B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110196642.3A CN112986393B (en) 2021-02-22 2021-02-22 Bridge inhaul cable damage detection method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110196642.3A CN112986393B (en) 2021-02-22 2021-02-22 Bridge inhaul cable damage detection method and system

Publications (2)

Publication Number Publication Date
CN112986393A true CN112986393A (en) 2021-06-18
CN112986393B CN112986393B (en) 2023-09-29

Family

ID=76349299

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110196642.3A Active CN112986393B (en) 2021-02-22 2021-02-22 Bridge inhaul cable damage detection method and system

Country Status (1)

Country Link
CN (1) CN112986393B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113654979A (en) * 2021-09-03 2021-11-16 郑州大学 Acoustic emission monitoring device and evaluation method for bridge prestressed steel wire in rainfall simulation
CN115393396A (en) * 2022-08-18 2022-11-25 西安电子科技大学 Unmanned aerial vehicle target tracking method based on mask pre-training

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5457994A (en) * 1992-11-06 1995-10-17 Southwest Research Institute Nondestructive evaluation of non-ferromagnetic materials using magnetostrictively induced acoustic/ultrasonic waves and magnetostrictively detected acoustic emissions
WO2011016857A2 (en) * 2009-08-05 2011-02-10 Elliott James C Equipment and system for structure inspection and monitoring
US20150323413A1 (en) * 2012-12-28 2015-11-12 Tsinghua University Tap-Scan Bridge Damage Detection System
CN110044554A (en) * 2019-04-16 2019-07-23 重庆大学 A kind of online test method of the metal pressure container leakage based on acoustic emission signal
CN110263945A (en) * 2019-05-22 2019-09-20 中国电力科学研究院有限公司 Based on markovian Line Loss of Distribution Network System processing system
CN111027256A (en) * 2020-03-09 2020-04-17 杭州鲁尔物联科技有限公司 Bridge risk prediction method and system based on vehicle load spatial distribution
CN111912907A (en) * 2020-07-22 2020-11-10 柳州欧维姆机械股份有限公司 Intelligent inhaul cable system for evaluating corrosion broken wire and analysis method thereof

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5457994A (en) * 1992-11-06 1995-10-17 Southwest Research Institute Nondestructive evaluation of non-ferromagnetic materials using magnetostrictively induced acoustic/ultrasonic waves and magnetostrictively detected acoustic emissions
WO2011016857A2 (en) * 2009-08-05 2011-02-10 Elliott James C Equipment and system for structure inspection and monitoring
US20150323413A1 (en) * 2012-12-28 2015-11-12 Tsinghua University Tap-Scan Bridge Damage Detection System
CN110044554A (en) * 2019-04-16 2019-07-23 重庆大学 A kind of online test method of the metal pressure container leakage based on acoustic emission signal
CN110263945A (en) * 2019-05-22 2019-09-20 中国电力科学研究院有限公司 Based on markovian Line Loss of Distribution Network System processing system
CN111027256A (en) * 2020-03-09 2020-04-17 杭州鲁尔物联科技有限公司 Bridge risk prediction method and system based on vehicle load spatial distribution
CN111912907A (en) * 2020-07-22 2020-11-10 柳州欧维姆机械股份有限公司 Intelligent inhaul cable system for evaluating corrosion broken wire and analysis method thereof

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
EMMANUEL RAMASSO 等: "Unsupervised Consensus Clustering of Acoustic Emission Time-Series for Robust Damage Sequence Estimation in Composites", IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, vol. 64, no. 12, pages 3297 - 3307, XP011589145, DOI: 10.1109/TIM.2015.2450354 *
朱三凡 等: "桥梁技术状况退化模型研究现状与展望", 市政技术, no. 02, pages 64 - 68 *
钱骥 等: "桥梁拉索损伤声发射监测研究进展", 应用声学, vol. 35, no. 04, pages 369 - 376 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113654979A (en) * 2021-09-03 2021-11-16 郑州大学 Acoustic emission monitoring device and evaluation method for bridge prestressed steel wire in rainfall simulation
CN115393396A (en) * 2022-08-18 2022-11-25 西安电子科技大学 Unmanned aerial vehicle target tracking method based on mask pre-training
CN115393396B (en) * 2022-08-18 2024-02-02 西安电子科技大学 Unmanned aerial vehicle target tracking method based on mask pre-training

Also Published As

Publication number Publication date
CN112986393B (en) 2023-09-29

Similar Documents

Publication Publication Date Title
CN108984893B (en) Gradient lifting method-based trend prediction method
CN110728654B (en) Automatic pipeline detection and classification method based on deep residual error neural network
Xia et al. Multi-stage fault diagnosis framework for rolling bearing based on OHF Elman AdaBoost-Bagging algorithm
CN110751108B (en) Subway distributed vibration signal similarity determination method
CN112036075A (en) Abnormal data judgment method based on environmental monitoring data association relation
CN112986393B (en) Bridge inhaul cable damage detection method and system
CN112200237B (en) Time sequence monitoring data abnormality diagnosis method for structural health monitoring system
CN116625438A (en) Gas pipe network safety on-line monitoring system and method thereof
CN115673596B (en) Welding abnormity real-time diagnosis method based on Actor-Critic reinforcement learning model
CN112507479B (en) Oil drilling machine health state assessment method based on manifold learning and softmax
Son et al. Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge
CN112763967A (en) BiGRU-based intelligent electric meter metering module fault prediction and diagnosis method
CN111044617B (en) Inhaul cable damage identification method based on deep learning and acoustic emission technology
CN114118219A (en) Data-driven real-time abnormal detection method for health state of long-term power-on equipment
CN114861778A (en) Method for rapidly classifying rolling bearing states under different loads by improving width transfer learning
Chou et al. SHM data anomaly classification using machine learning strategies: A comparative study
CN117056678B (en) Machine pump equipment operation fault diagnosis method and device based on small sample
CN117371608A (en) Pig house multi-point temperature and humidity prediction method and system based on deep learning
CN116819423A (en) Method and system for detecting abnormal running state of gateway electric energy metering device
CN111523557A (en) Wind power intelligent fault diagnosis method based on big data
CN111160419B (en) Deep learning-based electronic transformer data classification prediction method and device
CN112731567B (en) Time-space collaborative dry-wet enhancement discrimination method for ultrahigh frequency microwave
CN115204362A (en) Method for predicting residual life interval of mechanical equipment
CN113988210A (en) Method and device for restoring distorted data of structure monitoring sensor network and storage medium
CN116400675B (en) Fault diagnosis system and method based on improved CNN-LSTM model

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