A steel strand stress identification method based on ultrasonic guided waves and convolutional neural network

Abstract

The application discloses a steel strand stress identification method based on ultrasonic guided waves and a convolutional neural network, and comprises the following steps: installing ultrasonic guided wave excitation sensors and receiving sensors on the cross sections of the center wires at the two ends of the steel strand; tensioning the steel strand, exciting modulated sinusoidal wave signals to the excitation sensors under different tension prestresses, and recording the ultrasonic guided wave signals obtained by the receiving sensors; taking the ultrasonic guided wave signals under different prestress states as characteristic values, taking the corresponding prestress values as label values, and constructing a data set of a one-dimensional convolutional neural network; inputting the data of the data set into the convolutional neural network for training and testing, and outputting the prestress prediction values of the ultrasonic guided wave signals under each prestress state; and taking the root mean square error and the normalized root mean square error between the real values and the prediction values of the prestress of the steel strand as indexes to evaluate the prediction results of the one-dimensional convolutional neural network. The application avoids a complicated parameter extraction process and improves the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of structural nondestructive testing and computer artificial intelligence, and particularly relates to a steel strand stress identification method based on ultrasonic guided waves and a convolutional neural network. BACKGROUND

[0002] High-strength and low-relaxation steel strand is a component mainly bearing tension, which has been widely used in prestressed concrete structures and cable-supported bridges. The stress level of steel strand is crucial to the performance of the structure. Previous studies have shown that the effective prestress of steel strand will decrease with the increase of service life and the action of corrosion and other factors, thereby leading to the reduction of the bearing capacity of the structure and the shortening of the service life. Therefore, accurately predicting the effective prestress of steel strand is of great significance to the evaluation of the service performance of the structure.

[0003] In recent years, some scholars have applied ultrasonic guided wave technology to the prestress evaluation of structures. The acoustic-elastic effect method is based on the change of ultrasonic guided wave velocity with prestress. The defect frequency method is based on the "frequency band missing" phenomenon of steel strand under the action of prestress, such as the change of the peak value ratio of the two peaks on the left and right sides of the defect frequency band with prestress. The singular value-based method needs to perform wavelet packet transformation and singular value decomposition on the ultrasonic guided wave signals of steel strand under different prestresses, and then predict the prestress of steel strand based on the linear relationship between singular value and prestress. However, the above methods need to manually extract relevant parameters from the time domain or frequency domain signals of ultrasonic guided waves, which is a tedious and time-consuming process.

[0004] Convolutional neural networks have shown outstanding advantages in the field of damage detection and positioning of civil engineering structures. One-dimensional convolutional neural networks belong to a type of convolutional neural networks and have been applied to time series analysis of sensors. The most prominent advantage is that it can automatically extract parameters related to damage from raw data and make predictions without additional feature extraction operations.

[0005] Therefore, it is necessary to propose a steel strand stress identification method based on ultrasonic guided waves and convolutional neural networks to solve the above problems. SUMMARY

[0006] Therefore, the present application aims to provide a steel strand stress identification method based on ultrasonic guided waves and convolutional neural networks to solve the problem of tedious and time-consuming extraction of relevant parameters from the time domain or frequency domain signals of ultrasonic guided waves in the prior art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] The present application provides a steel strand stress identification method based on ultrasonic guided waves and convolutional neural networks, comprising the following steps:

[0009] S1: install ultrasonic guided wave excitation sensor and receiving sensor on the cross section of the center wire at both ends of the steel strand respectively;

[0010] S2: tension the steel strand, excite modulated sinusoidal wave signal to the excitation sensor under different tension prestress, and record the ultrasonic guided wave signal obtained by the receiving sensor;

[0011] S3: take the ultrasonic guided wave signal under different prestress state as the characteristic value, and take the corresponding prestress value as the label value to construct the data set of one-dimensional convolutional neural network;

[0012] S4: input the data of the data set into the one-dimensional convolutional neural network to train and test, and output the prestress prediction value of the ultrasonic guided wave signal under each prestress state;

[0013] S5: take the root mean square error and normalized root mean square error between the true value and the predicted value of the steel strand prestress as the index to evaluate the prediction result of the one-dimensional convolutional neural network.

[0014] Further, the structure of the one-dimensional convolutional neural network in step S4 comprises:

[0015] The first layer is a convolutional layer for feature extraction of input data;

[0016] The second layer is a batch normalization layer for accelerating the convergence speed of the model and reducing the sensitivity to network initialization;

[0017] The third layer is a maximum pooling layer for compressing the dimension of the feature to reduce the calculation amount;

[0018] The fourth layer is a convolutional layer,

[0019] The fifth layer is a batch normalization layer,

[0020] The sixth layer is a maximum pooling layer;

[0021] The seventh layer is a flattening layer for flattening the input multi-dimensional data into a one-dimensional array to realize the transition from the convolutional layer to the fully connected layer;

[0022] The eighth layer is a random dropout layer for randomly dropping some neurons in each training process to prevent the model from overfitting, and the random dropout rate is 0.2;

[0023] The ninth layer is a fully connected layer for connecting all neurons of the previous layer and outputting the final prestress prediction value.

[0024] Further, the optimizer of the one-dimensional convolutional neural network in step S4 selects Adam function, and the learning rate is 0.005.

[0025] Further, the loss function of the one-dimensional convolutional neural network in step S4 is calculated by the following formula to calculate the mean square error:

[0026]

[0027] In the formula, MSE is the mean square error, n is the number of samples, y true and y pre are the true value and the predicted value of the prestress respectively.

[0028] Further, the root mean square error in step S5 is calculated by the following formula:

[0029]

[0030] In the formula, RMSE is the root mean square error, y max , y min are the maximum value and the minimum value of the sample respectively.

[0031] Further, the normalized root mean square error in step S5 is calculated by the following formula:

[0032]

[0033] In the formula, NRMSE is the normalized root mean square error, y max , y min are the maximum value and the minimum value of the sample respectively.

[0034] Further, the convolutional layer of the first layer includes 8 convolutional kernels, the convolutional layer of the fourth layer includes 4 convolutional kernels, the time domain window length of each convolutional kernel is 10, the step length is 1, and the activation function adopts the ReLU function.

[0035] Further, different signal-to-noise ratios of noise need to be added to each measurement signal, and the calculation formula of the signal-to-noise ratio is:

[0036]

[0037] In the formula, SNR is the signal-to-noise ratio, P S and P N are the effective power of the detection signal and the noise signal respectively.

[0038] Further, the pooling size of the max pooling layer is 2, and the step length is 2.

[0039] Further, the data set includes a training set and a test set, wherein 80% of the data in the data set is used as a training set for training, and the remaining 20% of the data is used as a test set for testing.

[0040] The beneficial effects of the present application are:

[0041] The prestress state of the steel strand can be directly predicted based on the original ultrasonic guided wave signal, so that a cumbersome and time-consuming parameter extraction process is avoided, and high prediction accuracy is achieved.

[0042] Additional advantages, objects, and features of the application will be set forth in the descriptions to follow, and in part will become apparent to those skilled in the art upon examination of the following or can be learned from practice of the application. The objects and other advantages of the application can be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to make the purposes, technical solutions and beneficial effects of the present application clearer, the present application provides the following drawings for illustration:

[0044] Figure 1 The flowchart of the embodiment of the present application;

[0045] Figure 2 The ultrasonic guided wave excitation signal waveform diagram of the embodiment of the present application;

[0046] Figure 3 The ultrasonic guided wave signal waveform diagram of the stress-free state of different signal-to-noise ratios of the embodiment of the present application;

[0047] Figure 4 The structure diagram of the one-dimensional convolutional neural network of the embodiment of the present application;

[0048] Figure 5 The loss change with the training round of the training set and the test set of the embodiment of the present application;

[0049] Figure 6 The prestress prediction result diagram of the training set of the embodiment of the present application;

[0050] Figure 7 The prestress prediction result diagram of the test set of the embodiment of the present application. DETAILED DESCRIPTION

[0051] As shown in Figures 1-7 The present application provides a steel strand stress identification method based on ultrasonic guided wave and convolutional neural network, which comprises the following steps:

[0052] S1: ultrasonic guided wave excitation sensors and receiving sensors are respectively installed at the center wire cross sections of both ends of the steel strand; the steel strand selected is a 7-wire steel strand commonly used in actual engineering, and the ultimate tensile strength thereof is 1860MPa;

[0053] S2: the steel strand is tensioned, a modulated sinusoidal wave signal is excited to the excitation sensor under different tension prestresses, and the ultrasonic guided wave signal obtained by the receiving sensor is recorded; the excitation signal waveform diagram is as shown in Figure 2The steel strand tension limit value is 75% of the ultimate tensile strength in the actual construction process, so the prestressed tension range of the steel strand is taken as 0-1395MPa, and the prestress interval of each loading condition is set as 93MPa, so that 16 groups of ultrasonic guided wave signals under different prestress conditions are obtained;

[0054] S3: The ultrasonic guided wave signals under different prestress states are taken as characteristic values, and the corresponding prestress values are taken as label values, a data set of the one-dimensional convolutional neural network is constructed, and 80% of the data in the data set is taken as a training set and the remaining 20% of the data is taken as a test set;

[0055] S4: The training set and test set data are input into the one-dimensional convolutional neural network to be trained and tested, and the prestress prediction value of the ultrasonic guided wave signal under each prestress state is output; the optimizer of the one-dimensional convolutional neural network to be built is selected as Adam function, and the learning rate is set as 0.005;

[0056] The structure of the one-dimensional convolutional neural network to be built includes:

[0057] The first layer is a convolutional layer, which is used for feature extraction of input data, and there are 8 convolutional kernels, the time domain window length of each convolutional kernel is 10, the step is 1, and the activation function adopts ReLU function;

[0058] The second layer is a batch normalization layer, which is used to speed up the convergence speed of the model and reduce the sensitivity to network initialization;

[0059] The third layer is a maximum pooling layer, which is used to compress the dimension of the feature to reduce the calculation amount, and the pooling size is 2 and the step is 2;

[0060] The fourth layer is a convolutional layer, which has 4 convolutional kernels, the time domain window length of each convolutional kernel is 10, the step is 1, and the activation function adopts ReLU function;

[0061] The fifth layer is a batch normalization layer;

[0062] The sixth layer is a maximum pooling layer, and the pooling size is 2 and the step is 2;

[0063] The seventh layer is a flattening layer, which is used to flatten the multi-dimensional data into a one-dimensional array to realize the transition from the convolutional layer to the fully connected layer;

[0064] The eighth layer is a random dropout layer, which is used to randomly drop some neurons in each training process to prevent the model from overfitting, and the random dropout rate is taken as 0.2;

[0065] The ninth layer is a fully connected layer, which is used to connect all the neurons of the previous layer and output the final prestress prediction value.

[0066] S5: the root mean square error and the normalized root mean square error between the true value and the predicted value of the steel strand prestress are taken as indexes to evaluate the prediction result of the one-dimensional convolutional neural network.

[0067] The working principle of the above technical solution is that.

[0068] The working principle and beneficial effects of the above technical solution are that the proposed steel strand stress identification method based on ultrasonic guided waves and one-dimensional convolutional neural network can directly predict the prestress state of the steel strand based on the original ultrasonic guided wave signal, thereby avoiding the cumbersome and time-consuming parameter extraction process and improving the prediction accuracy.

[0069] In an embodiment of the present application, different signal-to-noise ratios of noise are added to each measurement signal, and the signal-to-noise ratio is calculated by the following formula

[0070]

[0071] In the formula, SNR is the signal-to-noise ratio, P S and P N are the effective powers of the detection signal and the noise signal, respectively.

[0072] The working principle of the above technical solution is that random noise with signal-to-noise ratios of 25dB, 20dB, 15dB, 10dB and 5dB is added to the signal of each working condition, respectively, and 10 groups of signals are generated for each signal-to-noise ratio. For example, the ultrasonic guided wave signal in the unstressed state is shown in Figure 3 , and an ultrasonic guided wave data set with a total number of 800 signals is obtained, of which 640 signals are used as the training set of the one-dimensional convolutional neural network, and the remaining 160 signals are used as the test set.

[0073] The beneficial effects of the above technical solution are that by adding noise with different signal-to-noise ratios to each measurement signal, the sample number of the data set is enhanced, and the influence of environmental noise on the detection signal in the actual situation is considered, thereby ensuring the accuracy of the measurement signal and improving the prediction accuracy.

[0074] In an embodiment of the present application, the loss function of the one-dimensional convolutional neural network built in step S4 is selected as the mean square error, and its expression is

[0075]

[0076] In the formula, MSE is the mean square error, n is the number of samples, y true and y pre are the true value and the predicted value of the prestress, respectively.

[0077] The working principle and beneficial effects of the above technical solution are as follows: the training and test loss of the one-dimensional convolutional neural network changes with the training round as shown in Figure 5 The training set and test set loss functions both show a trend of first rapidly decreasing and then tending to be stable with the increase of the training round, and the numerical values of the training loss curve and the test loss curve are close; this shows that the one-dimensional convolutional neural network built in the application has good working performance and does not produce overfitting phenomenon.

[0078] In an embodiment of the application, the calculation formula of the root mean square error and the normalized root mean square error in step S5 is as follows:

[0079]

[0080]

[0081] In the formula, RMSE and NRMSE are the root mean square error and the normalized root mean square error, respectively, y max , y min are the maximum value and the minimum value of the sample, respectively.

[0082] The working principle of the above technical solution is as follows: the steel strand prestress prediction results of the training set and the test set are shown in Figure 6 and Figure 7 For the training set, the root mean square error between the true value and the predicted value of the prestress is 7.94 MPa, and the normalized root mean square error is 0.57%; for the test set, the root mean square error between the true value and the predicted value of the prestress is 13.01 MPa, and the normalized root mean square error is 0.93%.

[0083] The beneficial effects of the above technical solution are as follows: the error is maintained within 1% by the above method, and compared with the existing prediction method, the prediction accuracy is improved.

[0084] Finally, it should be pointed out that the above preferred embodiments are only used to illustrate the technical solutions of the application and not to limit, although the application has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the application.

Claims

1. A steel strand stress identification method based on ultrasonic guided waves and convolutional neural network, characterized by: The method comprises the following steps: S1: respectively install ultrasonic guided wave excitation sensors and receiving sensors on the cross sections of the center wires at both ends of the steel strand; S2: tension the steel strand, excite modulated sinusoidal wave signals to the excitation sensors under different tension prestresses, and record the ultrasonic guided wave signals obtained by the receiving sensors; S3: take the ultrasonic guided wave signals under different prestress states as characteristic values, and take the corresponding prestress values as label values to construct a data set of one-dimensional convolutional neural network; S4: input the data of the data set into the one-dimensional convolutional neural network to train and test, and output the prestress prediction values of the ultrasonic guided wave signals under each prestress state; S5: take the root mean square error and the normalized root mean square error between the true value and the prediction value of the prestress of the steel strand as indexes to evaluate the prediction results of the one-dimensional convolutional neural network; The structure of the one-dimensional convolutional neural network in step S4 comprises: The first layer is a convolutional layer for feature extraction of input data; The second layer is a batch normalization layer for accelerating the convergence speed of the model and reducing the sensitivity to network initialization; The third layer is a max pooling layer for compressing the dimension of the feature to reduce the calculation amount; The fourth layer is a convolutional layer, The fifth layer is a batch normalization layer, The sixth layer is a max pooling layer; The seventh layer is a flattening layer for flattening the multi-dimensional data into a one-dimensional array to realize the transition from the convolutional layer to the fully connected layer; The eighth layer is a random dropout layer for randomly dropping some neurons in each training process to prevent the model from overfitting, and the random dropout rate is 0.2; The ninth layer is a fully connected layer for connecting all neurons of the previous layer and outputting the final prestress prediction value.

2. The ultrasonic guided wave and convolutional neural network based steel strand stress identification method of claim 1, wherein: The optimizer of the one-dimensional convolutional neural network in step S4 is Adam function, and the learning rate is 0.

005. 3.The ultrasonic guided wave and convolutional neural network based steel strand stress identification method of claim 1, wherein, The loss function of the one-dimensional convolutional neural network in step S4 is calculated by the following formula: wherein, is the mean square error, is the number of samples, and are the true and predicted values of the prestress, respectively.

4. The ultrasonic guided wave and convolutional neural network-based steel strand stress identification method of claim 1, wherein, The root mean square error in step S5 is calculated by the following formula: wherein is the root mean square error, , are the maximum and minimum values of the sample, respectively.

5. The ultrasonic guided wave and convolutional neural network-based steel strand stress identification method of claim 1, wherein, The normalized root mean square error in step S5 is calculated by the following formula: wherein is the normalized root mean square error, , are the maximum and minimum values of the sample, respectively.

6. The ultrasonic guided wave and convolutional neural network-based steel strand stress identification method of claim 2, wherein, The convolution layer of the first layer includes 8 convolution kernels, the convolution layer of the fourth layer includes 4 convolution kernels, the time domain window length of each convolution kernel is 10, the step length is 1, and the activation function adopts ReLU function. ReLU function.

7. The ultrasonic guided wave and convolutional neural network-based steel strand stress identification method of claim 1, wherein, Different signal-to-noise ratios need to be added to each measurement signal, and the formula for calculating the signal-to-noise ratio is: wherein SNR S is the signal-to-noise ratio, and E{S} and E{N} are the effective powers of the detection signal and the noise signal, respectively.

8. The ultrasonic guided wave and convolutional neural network-based steel strand stress identification method of claim 1, wherein, The pooling size of the max pooling layer is 2, and the step is 2.

9. The ultrasonic guided wave and convolutional neural network-based steel strand stress identification method of claim 1, wherein, The data set comprises a training set and a test set, wherein 80% of the data in the data set is used for training as a training set, and the remaining 20% of the data is used for testing as a test set.

Images