Three-dimensional convolution prediction method for concrete pavement life based on microvibration signals

Abstract

The present invention discloses a three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals, including: obtaining a number of pavement samples to be monitored, performing a loading test on each pavement sample to be monitored, and obtaining a test data set; obtaining micro-vibration characteristic graphs at different times according to the test data set; obtaining a training data set based on the micro-vibration characteristic graph; constructing a Transformer model and a convolutional neural network model, and performing training respectively through the training data set to obtain a pavement life prediction model; collecting time-series micro-vibration signals of the pavement to be monitored, and predicting a concrete pavement life map based on the pavement life prediction model and the time-series micro-vibration signals. The present invention obtains a pavement life prediction model through the joint training of a Transformer model and a convolutional neural network model, and inputting the real-time collected micro-vibration signals into the model can accurately predict the life of the concrete pavement, which helps to achieve more accurate road maintenance and repair decisions.

Description

Technical Field

[0001] The invention belongs to the field of road engineering, and in particular relates to a three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals. Background Art

[0002] Before the softening point problem of asphalt pavement was solved, cement pavement was widely used on highways in my country. There are more than 3 million kilometers of existing cement pavement in my country. Cement pavement is a rigid pavement. Compared with asphalt pavement, which has a higher cost and a design life of up to 15 years, cement pavement not only has many advantages such as lower cost and longer design life, but also has higher strength, stronger diffusion load capacity and better stability. At present, there is a lack of means to monitor the life of cement pavement in the maintenance of cement pavement. During the maintenance process, FWD, ground penetrating radar and other methods are mainly used to detect the stress state of the board, thereby indirectly reflecting the life of the concrete pavement. The current means cannot directly reflect the life state of the board. The complex stress environment cannot represent the true life state of the board itself. It can only reflect the development direction of the force size of the concrete board from the side, which is difficult to meet the current needs of road pre-maintenance in my country. Concrete is a material that works with cracks. The cracks will cause the concrete slab to break after penetration. Cracks are caused by the rapid release of strain energy and elastic waves caused by the initiation of new cracks or the expansion of primary microcracks inside the material after being subjected to load. Therefore, there is an urgent need for a concrete pavement life prediction method based on micro-vibration detection technology to provide a reference for pavement maintenance. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above object, the present invention provides a three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals, comprising:

[0005] Obtain a number of road surface samples to be monitored, perform a loading test on each road surface sample to be monitored, and obtain a test data set;

[0006] Obtaining micro-vibration characteristic graphs at different times according to the test data set; obtaining a training data set based on the micro-vibration characteristic graphs;

[0007] Constructing a Transformer model and a convolutional neural network model, and training the Transformer model and the convolutional neural network model using the training data set to obtain a pavement life prediction model;

[0008] The time-series micro-vibration signal of the road surface to be monitored is collected, and based on the road surface life prediction model and the time-series micro-vibration signal, a concrete road surface life diagram is predicted.

[0009] Optionally, the process of performing a loading test on each road surface sample to be monitored includes:

[0010] A loading test plan is designed, micro-vibration signal sensors are evenly arranged at the upper and lower ends of the road surface sample to be monitored, micro-vibration signal receiving sensors are arranged around the sample, the road surface sample to be monitored is placed in a triaxial testing machine, and a loading test is performed on each sample based on the loading test plan to obtain a test data set, wherein the road surface sample to be monitored is obtained by core sampling the road surface to be monitored and processing it into a standard cylinder.

[0011] Optionally, the test data set includes the amplitude of the micro-vibration signal, the frequency of the micro-vibration signal, and the time and load when the sample is destroyed.

[0012] Optionally, the Transformer model includes an input layer, a feature extraction module, a feature conversion layer, a position encoding layer, a Transformer encoder, and a fully connected layer; the feature extraction module includes a three-dimensional convolution layer and a pooling layer.

[0013] Optionally, the process of obtaining a training data set includes:

[0014] Clean the test data set, check the integrity of the data, remove invalid or abnormal samples, and then perform normalization to extract time domain and frequency domain features from the normalized data;

[0015] Divide the time into several time periods of length t, where the value of t is between 0 and the moment Tmax when the sample is destroyed;

[0016] The feature set of each time point in the time period is arranged in chronological order to obtain a micro-vibration feature map, which is a three-dimensional feature map. According to the moment when the sample in the test data set is destroyed, the micro-vibration feature map is annotated with a life label to obtain a training data set.

[0017] Optionally, the process of obtaining a three-dimensional convolution prediction model includes:

[0018] Forward propagating the training data set through a convolutional neural network model to obtain an output of the convolutional neural network, and using the output of the convolutional neural network as a target label;

[0019] Construct a loss function, which is composed of the cross entropy loss function and the KL divergence loss. Train the Transformer model according to the target label and the real label of the training data set. Update the weight of the Transformer model through back propagation to obtain the pavement life prediction model.

[0020] Optionally, the process of collecting the time-series micro-vibration signal of the road surface to be monitored includes:

[0021] Delineate the equipment layout area, drill holes at equal intervals in the area, place the micro-vibration sensors in the holes, seal and fix the hole openings with clay, and collect the time-series micro-vibration signals of the road surface to be monitored through the micro-vibration sensors.

[0022] Optionally, the cross entropy loss function is as follows:

[0023]

[0024] In the formula, A train is the training loss; l is the lth training sample; c is the number of batch samples; y is the true label of the training sample; Predict results for the training samples.

[0025] Optionally, the three-dimensional convolution prediction method of concrete pavement life based on micro-vibration signals also includes: defining the pavement life at the moment of destruction, dividing the concrete pavement life map into four equal parts, corresponding to low-risk area, medium-risk area, relatively high-risk area, and high-risk area respectively.

[0026] Compared with the prior art, the present invention has the following advantages and technical effects:

[0027] The present invention discloses a three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals. A plurality of pavement samples to be monitored are obtained, and a loading test is performed on each pavement sample to be monitored to obtain a test data set; micro-vibration characteristic graphs at different times are obtained according to the test data set; a training data set is obtained based on the micro-vibration characteristic graph; a Transformer model and a convolutional neural network model are constructed, and the Transformer model and the convolutional neural network model are trained through the training data set to obtain a pavement life prediction model; time-series micro-vibration signals of the pavement to be monitored are collected, and a concrete pavement life graph is predicted based on the pavement life prediction model and the time-series micro-vibration signals.

[0028] The present invention obtains a pavement life prediction model through the joint training of the Transformer model and the convolutional neural network model. The real-time collected micro-vibration signal is input into the model to accurately predict the life of the concrete pavement, which helps to achieve more accurate road maintenance and repair decisions. The Transformer model lacks an inherent understanding of the spatial structure of the data. The combination of the Transformer model and the convolutional neural network model improves the model's generalization ability for new data. By dividing the pavement life map into areas with different risk levels, managers can more intuitively identify and prioritize high-risk areas. The automated data collection, processing and prediction process reduces manual intervention and improves work efficiency. The method of the present invention is applicable to a variety of different pavement conditions and environments and has strong adaptability and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0030] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0031] Figure 2 Schematic diagram of the layout of the acoustic emission sensor according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0033] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] Embodiment 1

[0035] like Figure 1-Figure 2 As shown, this embodiment provides a three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals, including:

[0036] Obtain a number of road surface samples to be monitored, perform a loading test on each road surface sample to be monitored, and obtain a test data set;

[0037] A loading test plan was designed. Micro-vibration signal sensors were evenly arranged at the upper and lower ends of the road surface sample to be monitored, and micro-vibration signal receiving sensors were arranged around the sample. The road surface sample to be monitored was placed in a triaxial testing machine. A loading test was performed on each sample based on the loading test plan to obtain a test data set. The road surface sample to be monitored was obtained by drilling core samples from the road surface to be monitored and processing them into standard cylinders.

[0038] The test data set includes the amplitude of the micro-vibration signal, the frequency of the micro-vibration signal, the time when the sample was destroyed, and the load.

[0039] Specifically, core samples are drilled from the road surface to be monitored and processed into a standard cylinder of Φ50mm×100mm. Micro-vibration signal receiving sensors are arranged around the cylinder, and three micro-vibration signal sensors are arranged at the upper and lower ends of the sample. The sample is placed in a triaxial tester for loading.

[0040] The tests with different loading rates are designed as follows, the amplitude and frequency of the micro-vibration signals at different positions of the sample cross section under different loading speeds are obtained, and the load at failure is recorded. The loading rate examples are shown in Table 1 below.

[0041] Table 1

[0042]

[0043] Model construction and training

[0044] The micro-vibration feature maps at different times are obtained according to the experimental data set; the Transformer model and the convolutional neural network model are constructed, and the Transformer model and the convolutional neural network model are trained through the micro-vibration feature maps at different times to obtain a three-dimensional convolution prediction model; a certain number of micro-vibration feature maps are used to construct a micro-vibration feature map combination and input it into the model training.

[0045] Clean the test data set, check the integrity of the data, remove invalid or abnormal samples, and then perform normalization to extract time domain and frequency domain features from the normalized data;

[0046] The time is divided into several time periods of length t, where the value of t is between 0 and the moment Tmax when the sample is destroyed. The finer the division, the shorter the time interval.

[0047] The feature set of each time point in the time period is arranged in chronological order to obtain a micro-vibration feature map, which is a three-dimensional feature map. According to the moment when the sample in the test data set is destroyed, the micro-vibration feature map is annotated with a life label to obtain a training data set.

[0048] The Transformer model includes an input layer, a feature extraction module, a feature conversion layer, a position encoding layer, a Transformer encoder, and a fully connected layer; the feature extraction module includes a three-dimensional convolution layer and a pooling layer;

[0049] Residual connections and layer normalization are introduced into the Transformer model to solve the problems of gradient vanishing and gradient exploding in model training and improve the training efficiency and performance of the model.

[0050] The training data set is forward propagated through a convolutional neural network model. In addition to the input of the training data set, sample sampling location information is also introduced during training, and the influence of different road sections is considered to obtain the output of the convolutional neural network, and the output of the convolutional neural network is used as the target label;

[0051] Construct a loss function, which is composed of the cross entropy loss function and the KL divergence loss. Train the Transformer model according to the target label and the real label of the training data set. Update the weight of the Transformer model through back propagation to obtain the pavement life prediction model.

[0052] Transformer can effectively capture long-distance dependencies in sequences, but lacks an inherent understanding of the spatial structure of data. The combination of the Transformer model and the convolutional neural network model improves the model's ability to generalize new data.

[0053] Calculate the cross entropy loss between the Transformer model's predicted label and the true label, and define the training set cross entropy loss as follows:

[0054]

[0055] Where: A train is the training loss; l is the lth training sample; since all data cannot be input into the model at once when training a neural network, it needs to be imported in batches, c is the number of samples in each batch; y is the true label of the training sample; Predict results for the training samples.

[0056] Repeat the above steps, using the minimization of the loss function as the training node during training to complete the model training.

[0057] The time-series micro-vibration signals of the road surface to be monitored are collected, and the concrete pavement life map is predicted based on the three-dimensional convolution prediction model and the time-series micro-vibration signals.

[0058] The process of collecting the time-series micro-vibration signals of the road surface to be monitored includes: demarcating the equipment layout area, drilling holes at equal intervals in the area, placing micro-vibration sensors in the holes, and sealing and fixing the hole mouths with clay, and collecting the time-series micro-vibration signals of the road surface to be monitored through the micro-vibration sensors.

[0059] The method of the three-dimensional convolution prediction method of concrete pavement life based on micro-vibration signals also includes: defining the pavement life at the moment of failure, dividing the concrete pavement life map into four equal parts, corresponding to low-risk area, medium-risk area, relatively high-risk area, and high-risk area respectively.

[0060] Specifically, the microseismic sensor is put back into the hole, and the edge of the hole is sealed with clay to prevent it from loosening during the collection process.

[0061] The micro-vibration signal of the concrete road slab is collected in real time by the micro-vibration sensor, and the time-series three-dimensional micro-vibration signal emitted by the micro-cracks inside the concrete pavement is obtained. The time-series three-dimensional micro-vibration signal is input into the three-dimensional convolutional neural network model to predict the complete concrete slab life map as shown in the figure. Figure 2 shown.

[0062] Define the life span at the moment of destruction as T max , the predicted life graph is obtained according to T max The percentage is divided into four equal parts, T 0～25％ 、T 25％～50％ 、T 50％～75％ 、T 75％～100％ . They correspond to low-risk areas, medium-risk areas, relatively high-risk areas, and high-risk areas respectively.

[0063] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals, characterized in that: The following steps are involved: Obtain a number of road surface samples to be monitored, perform a loading test on each road surface sample to be monitored, and obtain a test data set; Obtaining micro-vibration characteristic graphs at different times according to the test data set; Obtaining a training data set based on the micro-vibration feature map; The process of obtaining a training data set includes: Clean the test data set, check the integrity of the data, remove invalid or abnormal samples, and then perform normalization to extract time domain and frequency domain features from the normalized data; Divide the time into several time periods of length t, where the value of t is between 0 and the moment Tmax when the sample is destroyed; The feature set of each time point in the time period is arranged in chronological order to obtain a micro-vibration feature map, which is a three-dimensional feature map; according to the moment when the sample in the test data set is destroyed, the micro-vibration feature map is annotated with a life label to obtain a training data set; Constructing a Transformer model and a convolutional neural network model, and training the Transformer model and the convolutional neural network model using the training data set to obtain a pavement life prediction model; The Transformer model includes an input layer, a feature extraction module, a feature conversion layer, a position encoding layer, a Transformer encoder, and a fully connected layer; the feature extraction module includes a three-dimensional convolution layer and a pooling layer; The process of obtaining a pavement life prediction model includes: Forward propagating the training data set through a convolutional neural network model to obtain an output of the convolutional neural network, and using the output of the convolutional neural network as a target label; Construct a loss function, which is composed of a cross entropy loss function and a KL divergence loss. Train the Transformer model based on the target label and the real label of the training data set. Update the weight of the Transformer model through back propagation to obtain a pavement life prediction model. The time-series micro-vibration signal of the road surface to be monitored is collected, and based on the road surface life prediction model and the time-series micro-vibration signal, a concrete road surface life diagram is predicted.

2. The three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals according to claim 1 is characterized in that: The process of loading test for each pavement sample to be monitored includes: A loading test plan is designed, micro-vibration signal sensors are evenly arranged at the upper and lower ends of the road surface sample to be monitored, micro-vibration signal receiving sensors are arranged around the sample, the road surface sample to be monitored is placed in a triaxial testing machine, and a loading test is performed on each sample based on the loading test plan to obtain a test data set, wherein the road surface sample to be monitored is obtained by core sampling the road surface to be monitored and processing it into a standard cylinder.

3. The three-dimensional convolution prediction method for concrete pavement life based on microvibration signals according to claim 1 is characterized in that: The test data set includes the amplitude of the micro-vibration signal, the frequency of the micro-vibration signal, the time when the sample is destroyed, and the load.

4. The three-dimensional convolution prediction method for concrete pavement life based on microvibration signals according to claim 1 is characterized in that: The process of collecting the time-series micro-vibration signal of the road surface to be monitored includes: Delineate the equipment layout area, drill holes at equal intervals in the area, place the micro-vibration sensors in the holes, seal and fix the hole openings with clay, and collect the time-series micro-vibration signals of the road surface to be monitored through the micro-vibration sensors.

5. The three-dimensional convolution prediction method for concrete pavement life based on micro-vibration signals according to claim 1 is characterized in that: The cross entropy loss function is as follows: In the formula, A train is the training loss; l is the lth training sample; c is the number of batch samples; y is the true label of the training sample; Predict results for the training samples.

6. The three-dimensional convolution prediction method for concrete pavement life based on microvibration signals according to claim 1 is characterized in that: The method of the three-dimensional convolution prediction method of concrete pavement life based on micro-vibration signals also includes: defining the pavement life at the moment of failure, dividing the concrete pavement life map into four equal parts, corresponding to low-risk area, medium-risk area, relatively high-risk area, and high-risk area respectively.

Images