阅读说明：本技术 一种基于声发射信号的缆索断丝的判别方法 (cable wire breakage distinguishing method based on acoustic emission signals ) 是由 王涛 周文茜 姚超 任贝宁 于 2019-09-02 设计创作，主要内容包括：本发明公开的一种基于声发射信号的缆索断丝的判别方法,属于无损检测技术领域。本发明将多个传感器分别固定于缆索之上,同时获取各个传感器接收的声发射信号的波形；对各组断丝声发射信号进行时频分析,得到各组断丝声发射信号的时频图；对声发射信号做连续小波变换,根据预设强度阈值选取尺度和时间范围,得到尺寸为k×p的声发射信号的时频图；使用局部均值法对时频图进行降采样,用所有均值构成尺寸为m×m的图像；将m×m的图像作为输入数据,输入到自动编码器模型中,得到对应输出数据；计算输入数据与自动编码器的输出数据之间的重构误差；通过将重构误差与判别阈值作比较,判断所获取的声发射信号是否为断丝声发射信号,实现对缆索断丝现象的判别。(The invention discloses a method for judging broken wires of a cable based on an acoustic emission signal, and belongs to the technical field of nondestructive testing. The invention respectively fixes a plurality of sensors on a cable, and simultaneously obtains the waveform of an acoustic emission signal received by each sensor; carrying out time-frequency analysis on each group of broken wire acoustic emission signals to obtain a time-frequency graph of each group of broken wire acoustic emission signals; performing continuous wavelet transformation on the acoustic emission signals, and selecting a scale and a time range according to a preset intensity threshold to obtain a time-frequency graph of the acoustic emission signals with the size of k multiplied by p; using a local mean value method to perform down-sampling on the time-frequency image, and forming an image with the size of m multiplied by m by using all mean values; inputting the mxm image as input data into an automatic encoder model to obtain corresponding output data; calculating a reconstruction error between the input data and the output data of the auto-encoder; and comparing the reconstruction error with a discrimination threshold value to judge whether the obtained acoustic emission signal is a broken wire acoustic emission signal or not, so as to realize discrimination of the cable wire breakage phenomenon.)

1. a method for discriminating broken wires of a cable based on acoustic emission signals is characterized in that: comprises the following steps of (a) carrying out,

Firstly, respectively fixing two or more acoustic emission sensors on a cable, and when a broken wire acoustic emission phenomenon occurs in the cable, simultaneously acquiring the waveform of an acoustic emission signal received by each sensor, wherein the broken wire acoustic emission signals received by every two adjacent sensors are used as a group of broken wire acoustic emission signals;

Secondly, performing time-frequency analysis on each group of broken wire acoustic emission signals acquired in the first step to respectively obtain a time-frequency graph of each group of broken wire acoustic emission signals, wherein for each data point in the time-frequency graph, the abscissa corresponds to the frequency of the point, the ordinate corresponds to the time of the point, and the value of the point corresponds to the frequency and the magnitude of the energy density of the signal component at the time; performing continuous wavelet transformation on the acoustic emission signals acquired in the step one, and selecting a scale and a time range according to a preset acoustic emission signal intensity threshold to obtain a time-frequency graph of the acoustic emission signals with the size of k multiplied by p; using a local mean value method to carry out down-sampling on the obtained time-frequency image of the acoustic emission signal, wherein the down-sampling method is to equally divide the time-frequency image into a plurality of sub-blocks, calculate the mean value of all points in each sub-block, and form an image with the size of m multiplied by m by using all the mean values; calculating a reconstruction error between the input data constructed in the second step and the output data of the automatic encoder obtained in the third step;

step three, inputting the time-frequency diagram with the size of m multiplied by m constructed in the step two into the constructed automatic encoder model as input data to obtain corresponding output data;

Step four, calculating the reconstruction error between the input data constructed in the step two and the output data of the automatic encoder obtained in the step three;

And step five, comparing the reconstruction error obtained in the step four with a judgment threshold value, and judging whether the obtained acoustic emission signal is a broken wire acoustic emission signal or not, so as to realize the judgment of the cable wire breakage phenomenon.

2. The method for discriminating the broken wire of the cable based on the acoustic emission signal as claimed in claim 1, wherein: step three, the automatic encoder is a neural network model comprising an input layer, a single-layer hidden layer and an output layer, the output layer and the input layer of the network have the same dimension, the neural network model is constructed by the following method,

Step 3.1: acquiring a large number of acoustic emission signals generated under the condition of no wire breakage in the cable according to the waveform acquisition method of the acoustic emission signals in the first step, processing the signals by adopting the time-frequency analysis method in the second step, taking the obtained data as training data, and establishing a training data set; the acoustic emission signals generated under the condition of non-broken wires comprise friction among steel wires in the cable and impact on the cable;

step 3.2: determining the structure of an automatic encoder model and constructing a corresponding objective function for the automatic encoder model;

the dimension of an input layer of the automatic encoder is m, the dimension of a hidden layer is n, and the dimension of an output layer of the automatic encoder is m; defining the input vector as x, the hidden layer output as h, and the output vector as y, i.e.

x＝[x₁,x₂,…,x_m]

h＝[h₁,h₂,…,h_n]

y＝[y₁,y₂,…,y_m]

let W be the weight of the input layer to the hidden layer, W 'be the weight of the hidden layer to the output layer, b be the bias on the hidden layer, and b' be the bias on the output layer, i.e.

b＝[b₁,b₂,…,b_n]

b′＝[b′₁,b′₂,…,b′_m]

Then there is

F (x) in the formula is an activation function, and a sigmoid function is usually used, and the expression is

The formula for the objective function of the auto-encoder is:

Wherein the content of the first and second substances,

Wherein:

n is the number of training samples in each batch during training,

x^(k)for the input vector corresponding to the kth sample in the current training batch,

y^(k)the output vector corresponding to the kth sample in the current training batch,

is the average liveness of the jth neuron over the current training batch,

β_rAnd rho are constants;

step 3.3: and (3) training the automatic encoder model constructed in the step (3.2) by using a training data set, and updating the weight and the bias parameters in the automatic encoder model until the value of the target function is reduced to be within an allowable range.

3. A method for discriminating cable broken wire based on acoustic emission signal as claimed in claim 1 or 2, characterized in that: the calculation formula of the reconstruction error in the step four is as follows,

Wherein:

x is the input vector corresponding to the input data constructed in the step two,

y is the output vector corresponding to the output data of the automatic encoder obtained in the step three,

and m is the dimension of the input and output layer.

4. the method as claimed in claim 3, wherein the method comprises determining whether the cable is broken based on the acoustic emission signalthe method is characterized in that: step five, the method for obtaining the discrimination threshold value comprises the steps of calculating the reconstruction errors between the input and the output of all the training samples in the training set according to the step four, and defining the average value of the obtained reconstruction errors as L_fif the discrimination threshold is 3L_f(ii) a Defining the loss function value corresponding to the current input as L, and then determining the rule as:

L<3L_fthe current signal is a friction signal; l is more than or equal to 3L_fthe current signal is a wire break signal.

Technical Field

the invention relates to a method for judging broken wires of a cable based on the construction and classification of an automatic encoder of an acoustic emission signal, which is used for monitoring whether the cable of a cable-stayed bridge has the phenomenon of broken wires and belongs to the technical field of nondestructive testing.

Background

the cable is used as a core bearing part of a cable-stayed bridge, the quality of the cable is directly related to the safety of the bridge, and the quality of the cable is mainly influenced by the breakage degree of the steel wires in the cable. Therefore, in order to ensure the safety and reliability of the cable-stayed bridge, the cable is subjected to factors such as dynamic load, environmental corrosion, stress corrosion and fatigue for a long time during the operation of the bridge, and all the factors can cause the steel wires in the cable to break during the use process. And the acoustic emission technology is adopted to effectively judge the cable wire breakage phenomenon.

The current acoustic emission damage judgment method is mainly based on a characteristic parameter method or a modal analysis method. The characteristic parameter method has limited analysis capability on acoustic emission signals due to less used information amount, and the change of external conditions easily influences the judgment result. Although the accuracy of the modal analysis method is high, different models must be established manually for different materials and structures. Therefore, it is necessary to provide a discrimination algorithm with higher adaptability while ensuring higher accuracy.

Disclosure of Invention

in order to solve the problems of low algorithm recognition rate, low adaptability and the like in the conventional cable wire breakage judgment method, the invention discloses a cable wire breakage judgment method based on an acoustic emission signal, which mainly solves the technical problems that: the automatic encoder based on the acoustic emission signal time-frequency diagram realizes the discrimination of the cable wire breakage phenomenon, can identify the hidden characteristics of the acoustic emission signal without manual characteristic extraction in the discrimination process, and has the advantages of accurate discrimination, strong adaptability and the like.

the purpose of the invention is realized by the following technical scheme.

The invention discloses a method for judging broken wires of a cable based on acoustic emission signals. And carrying out time-frequency analysis on each group of broken wire acoustic emission signals to obtain a time-frequency graph of each group of broken wire acoustic emission signals. Performing continuous wavelet transformation on the acoustic emission signals, and selecting a scale and a time range according to a preset intensity threshold to obtain a time-frequency graph of the acoustic emission signals with the size of k multiplied by p; and (3) performing down-sampling on the time-frequency image by using a local mean value method, and forming an image with the size of m multiplied by m by using all mean values. The m × m image is input as input data to the automatic encoder model, and corresponding output data is obtained. A reconstruction error between the input data and the output data of the auto-encoder is calculated. And comparing the reconstruction error with a discrimination threshold value to judge whether the obtained acoustic emission signal is a broken wire acoustic emission signal or not, so as to realize discrimination of the cable wire breakage phenomenon.

the invention discloses a method for judging broken wires of a cable based on acoustic emission signals, which comprises the following steps:

the method comprises the following steps that firstly, two or more acoustic emission sensors are respectively fixed on a cable, when a broken wire acoustic emission phenomenon occurs in the cable, the waveforms of acoustic emission signals received by the sensors are simultaneously obtained, and broken wire acoustic emission signals received by every two adjacent sensors serve as a group of broken wire acoustic emission signals.

And secondly, performing time-frequency analysis on the groups of broken wire acoustic emission signals acquired in the first step to respectively obtain a time-frequency graph of the groups of broken wire acoustic emission signals, wherein for each data point in the time-frequency graph, the abscissa corresponds to the frequency of the point, the ordinate corresponds to the time of the point, and the value of the point corresponds to the frequency and the energy density of the signal component at the time. Performing continuous wavelet transformation on the acoustic emission signals acquired in the step one, and selecting a scale and a time range according to a preset acoustic emission signal intensity threshold to obtain a time-frequency graph of the acoustic emission signals with the size of k multiplied by p; and (3) performing down-sampling on the obtained time-frequency image of the acoustic emission signal by using a local mean value method, wherein the down-sampling method is to equally divide the time-frequency image into a plurality of sub-blocks, calculate the mean value of all points in each sub-block, and form an image with the size of m multiplied by m by using all the mean values. And calculating the reconstruction error between the input data constructed in the step two and the output data of the automatic encoder obtained in the step three.

And step three, inputting the time-frequency diagram with the size of m multiplied by m constructed in the step two into the constructed automatic encoder model as input data to obtain corresponding output data.

Step three, the automatic encoder is a neural network model comprising an input layer, a single-layer hidden layer and an output layer, the output layer and the input layer of the network have the same dimensionality, and the neural network model construction method comprises the following steps:

step 3.1: acquiring a large number of acoustic emission signals generated under the condition of no wire breakage in the cable according to the waveform acquisition method of the acoustic emission signals in the first step, processing the signals by adopting the time-frequency analysis method in the second step, taking the obtained data as training data, and establishing a training data set; the acoustic emission signals generated under the condition of non-broken wires comprise friction among steel wires in the cable and impact on the cable.

Step 3.2: determining the structure of an automatic encoder model and constructing a corresponding objective function for the automatic encoder model;

The dimension of the input layer of the automatic encoder is m, the dimension of the hidden layer is n, and the dimension of the output layer is m. Defining the input vector as x, the hidden layer output as h, and the output vector as y, i.e.

x＝[x₁,x₂,…,x_m]

h＝[h₁,h₂,…,h_n]

y＝[y₁,y₂,…,y_m]

Let W be the weight of the input layer to the hidden layer, W 'be the weight of the hidden layer to the output layer, b be the bias on the hidden layer, and b' be the bias on the output layer, i.e.

b＝[b₁,b₂,…,b_n]

b′＝[b′₁,b′₂,…,b′_m]

Then there is

f (x) in the formula is an activation function, and a sigmoid function is usually used, and the expression is

The formula for the objective function of the auto-encoder is:

wherein the content of the first and second substances,

Wherein:

n is the number of training samples in each batch during training,

x^(k)for the input vector corresponding to the kth sample in the current training batch,

y^(k)output direction corresponding to the kth sample in the current training batchThe amount of the compound (A) is,

is the average liveness of the jth neuron over the current training batch,

β_rAnd rho are constants;

step 3.3: and (3) training the automatic encoder model constructed in the step (3.2) by using a training data set, and updating the weight and the bias parameters in the automatic encoder model until the value of the target function is reduced to be within an allowable range.

And step four, calculating the reconstruction error between the input data constructed in the step two and the output data of the automatic encoder obtained in the step three.

the calculation formula of the reconstruction error in the step four is as follows:

wherein:

x is the input vector corresponding to the input data constructed in the step two,

y is the output vector corresponding to the output data of the automatic encoder obtained in the step three,

m is the dimension of the input and output layer;

and step five, comparing the reconstruction error obtained in the step four with a judgment threshold value, and judging whether the obtained acoustic emission signal is a broken wire acoustic emission signal or not, so as to realize the judgment of the cable wire breakage phenomenon.

The method for acquiring the discrimination threshold comprises the following steps: calculating the reconstruction errors between the input and the output of all training samples in the training set according to the step four, and defining the average value of the obtained reconstruction errors as L_fif the discrimination threshold is 3L_f. Defining the loss function value corresponding to the current input as L, and then determining the rule as:

L<3L_fthe current signal is a friction signal; l is more than or equal to 3L_fthe current signal is a wire break signal.

has the advantages that:

1. the invention discloses a method for judging broken cable wires based on acoustic emission signals, which is a method for judging broken cable wires based on the construction and classification of an automatic encoder of the acoustic emission signals.

2. Because the automatic encoder can learn the characteristics of the input sample automatically, the time-frequency diagram of the acoustic emission signal generated by friction can be used as a training sample to train the automatic encoder, and the automatic encoder can find the internal rule of the friction signal. For a trained automatic encoder, a friction signal is used as an input, the obtained output can better reproduce the input, and the target function of the automatic encoder is a smaller value, which indicates that the value of the loss function is smaller. However, in the case of a wire break signal or a crack propagation signal as input, the value of the loss function will be relatively large because it does not have an intrinsic law similar to the friction signal. According to the method for judging the cable wire breakage based on the acoustic emission signal, the cable wire breakage phenomenon is judged based on the acoustic emission signal time-frequency diagram automatic encoder, the hidden characteristics of the acoustic emission signal can be identified without manual characteristic extraction in the judging process, the identification accuracy is improved compared with the existing characteristic parameter method, and the adaptability of the algorithm is enhanced compared with the existing modal analysis method.

Drawings

FIG. 1 is a flow chart of a method for discriminating broken wire of a cable based on the construction and classification of an automatic encoder of an acoustic emission signal, which is disclosed by the invention;

fig. 2 is a network configuration diagram of the automatic encoder.

Detailed Description

The invention is further illustrated with reference to the following figures and examples.