A method for calculating freeze-thaw crack depth of expansive soil slope based on deep learning

Abstract

The present invention discloses a method for calculating the depth of freeze-thaw cracks in an expansive soil slope based on deep learning, the method comprising the following steps: step S1: collecting an electrical signal data set; step S2: processing the electrical signal data; step S3: training a model; step S4: verifying and tuning the model; step S5: applying the model. The present invention can detect and identify the depth of cracks in an expansive soil slope under the action of freeze-thaw cycles, and once an abnormal situation is found, it can be discovered and remedied in time. The present invention utilizes a deep learning method, combined with the alternating current induced polarization (SIP) technology to measure the depth of cracks in expansive soil under the action of freeze-thaw cycles, thereby improving the efficiency and accuracy of identifying the depth of cracks in expansive soil.

Description

Technical Field

[0001] The present invention belongs to the technical field of crack identification and optimization, and relates to a method for calculating the depth of freeze-thaw cracks of an expansive soil slope based on deep learning. Background Art

[0002] Expansive soil is widely distributed in my country. Generally speaking, expansive soil has a high bearing capacity, but it is highly sensitive to water, has the characteristics of water absorption expansion, water loss shrinkage, water bearing capacity attenuation, shrinkage crack development, etc., and its properties are extremely unstable. When the seasonal freezing depth exceeds 0.5m, the engineering buildings will show obvious freeze-thaw disasters. In particular, when the expansive soil is in the alternating freeze-thaw cycle, it will cause the cracks in the soil to develop rapidly and the strength to decrease. A large number of engineering practices have shown that freeze-thaw cycles are one of the important factors leading to deformation and landslide damage of expansive soil slopes. As the distribution range of seasonal frozen areas gradually increases, the expansive soil within its range is repeatedly frozen and thawed, affecting the structure and mechanical indicators of the soil. In seasonal frozen soil areas, due to the large temperature difference between winter and summer, the expansive soil slope engineering undergoes repeated freeze-thaw cycles, which has a significant impact on the stability of the slope.

[0003] Unlike clay, expansive soil is mainly composed of hydrophilic minerals such as illite and montmorillonite, and is prone to swelling due to water absorption. Expansive soil slopes suffer from a large amount of slope instability and expansion due to repeated expansion and contraction under freeze-thaw cycles, which is difficult to prevent and control.

[0004] Generally speaking, the freeze-thaw cycle changes the matrix suction of the soil in the frost heave area of ​​the slope. When the tensile stress exceeds its own tensile capacity, the expansive soil shrinks unevenly and cracks increase. The increase in cracks not only reduces the integrity of the soil, but also increases the permeability of the expansive soil, increases the degree of water infiltration, and greatly reduces the shear strength and safety factor of the expansive soil slope, significantly deteriorates the engineering properties, and is more likely to cause landslide accidents in the expansive soil slope.

[0005] A large number of studies have shown that cracks play a controlling role in the strength of expansive soils. The final cracking depth is closely related to the instability and sliding depth of the expansive soil slope, which is the key to predicting and preventing expansive soil engineering disasters. At the same time, timely discovery and timely treatment of expansive soil slope problems will greatly reduce the losses caused by their destruction.

[0006] At present, most methods for detecting soil crack depth use manual detection methods. The detection process is usually that on-site detection personnel go to the vicinity of the structure to be detected, use relevant detection equipment to measure the crack depth, and then manually record relevant information such as the crack location and corresponding parameters. This type of method requires large manpower investment, high cost, long time consumption, and poor accuracy. Summary of the invention

[0007] In order to solve the above problems existing in the background technology, the present invention provides a method for calculating the depth of freeze-thaw cracks of expansive soil slopes based on deep learning.

[0008] The objective of the present invention is achieved through the following technical solutions:

[0009] A method for calculating the depth of freeze-thaw cracks in an expansive soil slope based on deep learning comprises the following steps:

[0010] Step S1: Collecting electrical signal data sets

[0011] Step S101: Preparation of expansive soil samples: drying the expansive soils of different environments, temperatures and slopes, crushing them, sieving them, adding water and mixing them to prepare expansive soil samples, and then subjecting the expansive soil samples to freeze-thaw cycle experiments, and recording the corresponding freeze-thaw crack depths;

[0012] Step S102: Electrochemical test: Graphite electrodes are buried at both ends of the expansive soil sample and connected with wires, which are then connected to an electrochemical workstation. First, the natural potential of the expansive soil sample, i.e., the static potential, is tested, and then the electrochemical workstation is used for zeroing. Finally, a disturbance voltage of 50 mV is used for testing, and the measurement frequency range is 10 5 Hz to 0.05Hz, and obtain the electrical signal data set of expansive soil samples with different crack depths;

[0013] Step S2: Processing electrical signal data

[0014] Step S201: Divide the electrical signal data set collected in step S1 into two groups, one group is marked as sample set A, and the other group is marked as sample set B;

[0015] Step S202: marking sample set A and sample set B respectively, so that each set of electrical signals corresponds to the actual crack depth;

[0016] Step S3: Training the model

[0017] Step S301: using the sample set A obtained in step S201 as model training data based on deep learning technology; constructing a VGG-16 convolutional neural network, inputting an expansive soil electrical signal data set A with different crack depths, and extracting training set features;

[0018] Step S302: using deep learning technology CNN for supervised learning and data training to obtain a calculation model for identifying the depth of cracks in expansive soil;

[0019] Step S4: Validate and tune the model

[0020] Step S401: taking the sample set B obtained in step S201 as an input parameter and inputting it into the calculation model for identifying the depth of cracks in expansive soil obtained in step S302;

[0021] Step S402: comparing the output result of the calculation model for identifying the depth of cracks in expansive soil with the corresponding depth data in sample set B;

[0022] Step S403: Based on the comparison results, the training weight parameters and network topology parameters of the calculation model for identifying the depth of cracks in expansive soil are optimized;

[0023] Step S5: Apply the model

[0024] Two graphite electrodes are arranged at two points on the surface of the earth, and an electrical signal similar to a sine wave is released to the input end through the measuring instrument. The disturbance voltage is 50mV and the measurement frequency range is 10 5 Hz to 0.05Hz, and then extract the disturbed feedback information at the receiving end, and use the optimized expansive soil crack depth identification calculation model to identify the collected electrical signal feedback information, so as to observe the development of the cracks.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] 1. The present invention can detect and identify the depth of cracks in expansive soil slopes under the action of freeze-thaw cycles, and once an abnormal situation is found, it can be discovered and remedied in time.

[0027] 2. The present invention utilizes a deep learning method and combines it with the alternating current induced polarization (SIP) technology to measure the depth of cracks in expansive soil under freeze-thaw cycles, thereby improving the efficiency and accuracy of identifying the depth of cracks in expansive soil.

[0028] 3. The present invention can not only automatically extract crack depth features, but also achieve good results in detection accuracy.

[0029] 4. The present invention has fast detection speed and high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The present invention is a flow chart for calculating the depth of freeze-thaw cracks in an expansive soil slope based on deep learning;

[0031] Figure 2 This is the CNN structure diagram. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is further described below in conjunction with the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.

[0033] Soil is generally a three-phase system consisting of soil particles, water, and gas. In unsaturated soil mechanics, soil is also described as a four-phase system, and the fourth phase is the interface between water and gas or the so-called shrinkage membrane. The mineral components of expansive soil mainly include viscous particles such as montmorillonite, illite, and kaolinite. The microstructure of expansive soil is a face-to-face aggregate. The surface of clay particles carries a certain negative charge through isotropic replacement, lattice defects and bond breaking, selective adsorption, dissociative adsorption, etc. Under the action of the electrostatic attraction generated, the cations in the aqueous solution are adsorbed to the surface of the clay particles. The clay particles are particularly small and have a large specific surface area that can adsorb a lot of cations. The electrical properties of the soil are related to the number of cations adsorbed by the clay particles, and the ability of the soil to adsorb cations is related to the clay content of the soil. In addition, the amount of pore water in the soil affects the size of the conductive path in the soil, which is reflected in the water content of the soil. The number of ions adsorbed on the surface of soil particles that dissolve in water varies with the pore water content, and the conductivity is also different. At the same time, the salt content of pore water (or mineralization in the case of groundwater) is different, and the conductivity is also different. Therefore, the conductivity of soil mainly involves two different processes: conduction through pore water and conduction on the surface of particles (cation conduction). Regarding the conductivity of soil, many researchers reflect the composition and structural characteristics of soil by studying the resistivity of soil. The resistivity of soil is the resistance presented when current passes vertically through a cubic soil with a side length of 1m. Resistance R is a parameter that characterizes the conductivity of an object, that is:

[0034]

[0035] Where: L is the length of the conductor; A is the cross-sectional area of ​​the conductor; ρ is the resistivity of the conductor.

[0036] As a mature technical means, high-frequency dielectric reflection testing has been widely practiced in the fields of geophysics and geotechnical engineering, and is widely used in the exploration of mineral veins, detection of rock and soil water conservancy characteristics, etc. In recent years, the alternating current induced polarization method (SIP) has emerged in the field of geotechnical engineering as a high-frequency dielectric reflection test.

[0037] High-frequency dielectric reflection testing is to observe the induced polarization effect of the electric field changing with frequency (frequency characteristics) under the excitation of alternating current (also called "frequency domain"). The surface lattices of clay particles usually carry negatively charged ions, and then an ionosphere (i.e., double ionosphere) with positively charged ions is formed on the surface of the particles. When a variable frequency power supply is applied to both ends of the soil sample, the distribution of cations in the double electric layer on the surface of the particles changes, thereby generating a polarized electric field. After removing the variable frequency power supply at both ends, the polarized electric field cannot be immediately returned to zero, but will dissipate over time, and the surface double electric layer ion distribution returns to the original state, forming a spectral polarization potential. However, after the external electric field is generated, the movement of the double electric layer ions is not instantaneous, but there is a certain phase delay. By measuring the phase delay, soil parameters such as surface attachments, porosity, and permeability coefficient can be obtained, and then the soil mechanical properties such as shear strength can be inferred, providing important information for site investigation.

[0038] By releasing an electrical signal similar to a sinusoidal wave at the input end, and then extracting the feedback information after the disturbance at the receiving end. The physical properties of expansive soil are a black box to us. To know its internal structure, composition and mechanical properties, a series of measures need to be taken. Establish an input and an output end on the expansive soil, input a disturbance signal at the input end, and you can get a signal output from the output end of the rock. The resistivity of the cracks in the expansive soil can be regarded as infinite, and there are no conductive ions in the cracks, and they will not be polarized. The cracks generally show high resistivity and low polarizability. Therefore, the depth of the cracks can be judged by the feedback information of the electrical signal.

[0039] According to the above principles, the present invention provides a method for calculating the depth of freeze-thaw cracks in expansive soil slopes based on deep learning, such as Figure 1 As shown, the method comprises the following steps:

[0040] Step S1: Collecting electrical signal data sets

[0041] Step S101: Preparation of expansive soil samples: The expansive soils of different environments, temperatures and slopes are dried, crushed, sieved, mixed with water, and compacted in a cylindrical container with a bottom diameter of 15 cm and a height of 15 cm to prepare expansive soil samples. Then, freeze-thaw cycle experiments are performed on the expansive soil samples, and the corresponding freeze-thaw crack depths are recorded.

[0042] Step S102: Electrochemical test: Graphite electrodes are buried at both ends of the expansive soil sample. The size of the graphite electrodes is uniformly 2cm×1.5cm×0.5cm (length×width×height), and the wires are connected to the electrochemical workstation. First, the natural potential of the expansive soil sample, that is, the static potential, is tested, and then the electrochemical workstation is used for zeroing; finally, a disturbance voltage of 50mV is used for testing, and the measurement frequency range is 10 5Hz to 0.05Hz, and obtain the electrical signal data set of expansive soil samples with different crack depths.

[0043] Step S2: Processing electrical signal data

[0044] Step S201: Divide the electrical signal data set collected in step S1 into two groups according to a ratio of 4:1, one group is marked as sample set A, and the other group is marked as sample set B;

[0045] Step S202: Mark the sample set A and the sample set B respectively, so that each set of electrical signals corresponds to the actual crack depth.

[0046] Step S3: Training the model

[0047] Step S301: using the sample set A obtained in step S201 as model training data based on deep learning technology; constructing a VGG-16 convolutional neural network, inputting an expansive soil electrical signal data set A with different crack depths, and extracting training set features;

[0048] Step S302: using deep learning technology CNN for supervised learning and data training to obtain a calculation model for identifying the depth of cracks in expansive soil;

[0049] Step S30201: Train the CNN model, such as Figure 2 As shown, the CNN model includes an RPN network and a RoI Pooling network connected in sequence;

[0050] Step S30202: Based on deep learning technology, use the CNN model to train the expansive soil crack depth dataset.

[0051] Step S4: Validate and tune the model

[0052] Step S401: taking the sample set B obtained in step S201 as an input parameter and inputting it into the calculation model for identifying the depth of cracks in expansive soil obtained in step S302;

[0053] Step S402: comparing the output result of the calculation model for identifying the depth of cracks in expansive soil with the corresponding depth data in sample set B;

[0054] Step S403: Based on the comparison results, the training weight parameters and network topology parameters of the calculation model for identifying the depth of cracks in expansive soil are optimized.

[0055] The sample set A of the training data is input into the CNN model, and the fit of the CNN model is verified using the sample set B of the test data. The expansive soil crack depth electrical signal dataset is prepared according to the PascalVOC dataset format used by CNN, and an XML format file is generated for subsequent CNN training. The XML format file contains the name of the expansive soil crack depth electrical signal dataset.

[0056] Verify and optimize the CNN model. Based on sample set B as input parameter, input the calculation model for crack depth recognition in expansive soil to obtain the crack depth result. Statistically view the corresponding accuracy and recall rate. By adjusting the weights of the calculation model for crack depth recognition in expansive soil and network topology and other parameters, the calculation model for crack depth recognition in expansive soil is optimized so that the accuracy and recall rate meet the requirements of the sample verification model effect.

[0057] Step S5: Apply the model

[0058] Two graphite electrodes are arranged at two points on the surface of the earth, and an electrical signal similar to a sine wave is released to the input end through the measuring instrument. The disturbance voltage is 50mV and the measurement frequency range is 10 5 Hz to 0.05Hz, and then extract the disturbed feedback information at the receiving end, and use the optimized expansive soil crack depth identification calculation model to identify the collected electrical signal feedback information, so as to observe the development of the cracks.

Claims

1. A method for calculating the depth of freeze-thaw cracks in expansive soil slopes based on deep learning, characterized in that The method comprises the following steps: Step S1: Collecting electrical signal data sets Step S101: Preparation of expansive soil samples: drying the expansive soils of different environments, temperatures and slopes, crushing them, sieving them, adding water and mixing them to prepare expansive soil samples, and then subjecting the expansive soil samples to freeze-thaw cycle experiments, and recording the corresponding freeze-thaw crack depths; Step S102: Electrochemical test: Graphite electrodes are buried at both ends of the expansive soil sample and connected with wires, which are then connected to an electrochemical workstation. First, the natural potential of the expansive soil sample, i.e., the static potential, is tested, and then the electrochemical workstation is used for zeroing. Finally, a disturbance voltage of 50 mV is used for testing, and the measurement frequency range is 10 5 Hz to 0.05Hz, and obtain the electrical signal data set of expansive soil samples with different crack depths; Step S2: Processing electrical signal data Step S201: Divide the electrical signal data set collected in step S1 into two groups, one group is marked as sample set A, and the other group is marked as sample set B; Step S202: marking sample set A and sample set B respectively, so that each set of electrical signals corresponds to the actual crack depth; Step S3: Training the model Step S301: using the sample set A obtained in step S201 as model training data based on deep learning technology; constructing a VGG-16 convolutional neural network, inputting an expansive soil electrical signal data set A with different crack depths, and extracting training set features; Step S302: using deep learning technology CNN for supervised learning and data training to obtain a calculation model for identifying the depth of cracks in expansive soil; Step S4: Validate and tune the model Step S401: taking the sample set B obtained in step S201 as an input parameter and inputting it into the calculation model for identifying the depth of cracks in expansive soil obtained in step S302; Step S402: comparing the output result of the calculation model for identifying the depth of cracks in expansive soil with the corresponding depth data in sample set B; Step S403: Based on the comparison results, the training weight parameters and network topology parameters of the calculation model for identifying the depth of cracks in expansive soil are optimized; Step S5: Apply the model Two graphite electrodes are arranged at two points on the surface of the earth. An electrical signal similar to a sine wave is released to the input end through a measuring instrument. The disturbance voltage of the electrical signal is 50mV and the measurement frequency range is 10 5 Hz to 0.05Hz, and then extract the disturbed feedback information at the receiving end, and use the optimized expansive soil crack depth identification calculation model to identify the collected electrical signal feedback information, so as to observe the development of the cracks.

2. The method for calculating the depth of freeze-thaw cracks of expansive soil slopes based on deep learning according to claim 1 is characterized in that The specific steps of S302 are as follows: Step S30201: training a CNN model, wherein the CNN model includes an RPN network and a RoI Pooling network connected in sequence; Step S30202: Based on deep learning technology, use the CNN model to train the expansive soil crack depth dataset.

Images