Ground stress measurement method based on indentation technology and machine learning

Abstract

The invention belongs to the field of crustal stress measurement, and discloses a crustal stress measurement method based on an indentation technology and machine learning, which comprises the following steps: firstly drilling a rock core, and processing the rock core into a sample; performing a shallow indentation test to obtain an indentation load-displacement curve, and obtaining a rock core equivalent elastic model; carrying out a deep indentation test, and taking test data as an actual training sample of machine learning; inputting the actual training sample training set into a neural network for network training to obtain a ground stress inverse problem model; carrying out an in-situ indentation test, and calculating to obtain a minimum horizontal principal stress and a maximum horizontal principal stress according to the rock core equivalent elastic model; measuring a drilling well bottom indentation main crack direction by means of an imaging logging instrument, wherein the drilling well bottom indentation main crack direction is the maximum horizontal main stress direction; the earth deep rock crustal stress measurement method is constructed based on an indentation technology and machine learning; compared with an existing conventional rock mechanical property measurement technology, the indentation technology is convenient to operate, and in-situ testing can be carried out on mechanical behaviors of rocks in the deep part of the earth and in a large deformation area of soft rocks.

Description

Technical Field

[0001] The present invention belongs to the field of ground stress measurement, and in particular relates to a ground stress measurement method based on indentation technology and machine learning. Background Art

[0002] With the increasing demand for energy and mineral resources and the continuous increase in mining intensity, shallow mineral resources are decreasing. The proportion of broken and weak surrounding rock tunnels under high ground stress conditions in deep mines has increased sharply, and large deformation disasters of soft rocks occur from time to time, usually showing typical characteristics such as high stress, large deformation, strong rheology, low strength, and difficult support. It has become the focus and difficult problem of rock mechanics research in deep mining.

[0003] The commonly used geostress measurement methods nowadays usually simplify the rock mass into mechanical models such as elasticity, viscoelasticity, poroelasticity or ideal elastoplasticity, resulting in various commonly used geostress testing methods having their own scope of application and certain limitations.

[0004] The hydraulic fracturing method is based on linear elastic mechanics and proposes the following assumptions: the rock mass at the measuring point is a continuous, uniform, isotropic linear elastic body; the rock mass at the measuring point is non-permeable; the borehole must be parallel to one of the principal stresses. Although this method can measure the minimum horizontal principal stress well, the error in measuring the maximum horizontal principal stress is large, and it cannot be used to measure the ground stress value in the rock crushing area or plastic deformation area. The stress relief method (Ground Stress Relief Method) assumes that the rock mass is a continuous, homogeneous and isotropic linear elastic body. The stress and strain of the rock mass have the same functional relationship during loading and unloading. The solution of elastic theory is used to calculate the stress magnitude and direction of the in-situ stress of the rock unit. This method has high requirements for measuring the rock mass and is difficult to measure deep ground stress. The Anelastic Strain Recovery Method assumes that the rock is a homogeneous, isotropic viscoelastic material. When the core is separated from the surrounding rock mass, part of it will immediately undergo elastic recovery, and the other part will undergo slow hysteresis elastic recovery over time, and the strain recovery in each direction is positively correlated with the previous pressure. For rocks with high clay mineral content (such as shale), the shrinkage deformation caused by mineral dehydration during the measurement is opposite to the expansion deformation caused by stress release. Therefore, the Anelastic Strain Recovery Method is not suitable for rock samples with high shale content. The Borehole Breakout Method uses the shape of the borehole wall collapse and the rock strength parameters to determine the magnitude of the horizontal principal stress, and estimates the stress value based on the depth and width of the borehole wall collapse. When there is no collapse in the borehole, the relevant ground stress information cannot be obtained; if the anisotropy or heterogeneity of the rock is prominent, it will also cause a large error in the determination of the ground stress value and orientation. The Acoustic Emission Method is a method for measuring ground stress based on elastic theory using the Kaiser effect of rock acoustic emission. However, since acoustic emission is related to elastic wave propagation, high-strength brittle rocks usually have a more obvious Kaiser effect, while the Kaiser effect of porous, low-strength and plastic rock masses is often not obvious. Therefore, it is generally not recommended to use the acoustic emission method to measure the stress in weak, loose and plastic rock masses, and it is only allowed to be used within the elastic range of the rock (within 60% of the rock failure strength). The focal mechanism analysis method (Focal Mechanism of Earthquake) is based on the linear elastic theory, and takes the discrete focal mechanism of each microseismic event and the dislocation direction of the rupture surface as input parameters. The ground stress inversion based on the real section slip vector can obtain the ground stress in a statistical sense. This method can only determine the stress changes caused by earthquakes in the source area, the spatial tectonic stress direction of a large area, and the relative size of the three principal stresses, but cannot obtain absolute values.Since each geostress testing method has its limitations, in recent years, multiple methods are often combined to measure the geostress of a region, and geostress testing methods combining field measurements, numerical simulations, and mechanical modeling have been developed to adapt to various complex geological conditions and improve data accuracy.

[0005] Deep rocks and shallow soft rocks in the earth often show obvious nonlinear characteristics. However, the current geostress testing technology based on elastic theory is difficult to accurately measure the geostress in deep soft rock areas. It is urgent to study the nonlinear stress-strain relationship of rocks under high temperature and high confining pressure and establish a geostress measurement method suitable for deep rocks and shallow soft rocks based on this, so as to provide scientific and technological support for deep earth scientific exploration and underground engineering construction. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a method for measuring ground stress based on indentation technology and machine learning to solve the problems in the prior art. The technical solution adopted by the present invention is:

[0007] The ground stress measurement method based on indentation technology and machine learning includes the following steps:

[0008] Step 1: Drill the core and record the original core temperature, and store the core in a moisturizing manner;

[0009] Step 2: Processing the core into samples;

[0010] Step 3: Calculate the pressure of the overlying rock layer above the core;

[0011] Step 4: Heat the core sample to the original core temperature, perform a shallow indentation test on the sample with a conical indenter, obtain the indentation load-displacement curve, and obtain the core equivalent elastic model;

[0012] Step 5: Heat the sample to the original core temperature, set the minimum horizontal stress and the maximum horizontal principal stress, perform a deep indentation test on the sample with a conical indenter, and use the test data as the actual training sample for machine learning;

[0013] Step 6: Expand the actual training samples into training sets; input the training sets into the neural network for network training to obtain the inverse problem model of geostress;

[0014] Step 7: Conduct an in-situ indentation test to obtain an indentation load-displacement curve, calculate the curvature of the loading curve of the in-situ indentation load-displacement curve, the slope of the maximum indentation depth of the unloading curve, and the ratio of residual work to total work; input the trained neural network to obtain the output value of the maximum horizontal principal stress dimensionless value and the minimum horizontal principal stress dimensionless value, and calculate the minimum horizontal principal stress and the maximum horizontal principal stress according to the core equivalent elastic model obtained in the fourth step;

[0015] Step 8: Use imaging logging instruments to measure the main crack direction of the bottom indentation of the borehole, which is the direction of the maximum horizontal principal stress.

[0016] The present invention has the following beneficial effects:

[0017] The present invention constructs a method for measuring the geostress of deep earth rocks based on indentation technology and machine learning; compared with the current conventional rock mechanical property measurement technology, the indentation technology is easy to operate and can perform in-situ testing on the mechanical behavior of deep earth rocks; deep earth rocks have ductile deformation characteristics, and the adopted Bayesian neural network can construct a nonlinear mechanical model, which can not only characterize the elastic deformation behavior of deep earth rocks, but also characterize their plastic deformation, thus overcoming the shortcomings of the current geostress measurement method based on elastic theory and further expanding the measurement range of deep earth geostress. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of the indentation load-displacement curve of the present invention;

[0019] Figure 2 It is a schematic diagram of the inverse problem model of geostress based on Bayesian neural network of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the embodiments of the present invention Figure 1-Figure 2 , the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0021] The ground stress measurement method based on indentation technology and machine learning includes the following steps:

[0022] Step 1: Drill the core and record the original core temperature, and store the core to keep it moist.

[0023] Step 2: Process the core into samples (5cm×5cm×10cm).

[0024] Step 3: Calculate the pressure of the overlying rock layer above the core according to formula (1).

[0025] p=ρgh r (1)

[0026] Where ρ is the rock density, g is the gravitational acceleration, and h is r is the thickness of the overlying rock.

[0027] Step 4: Heat the core sample to the original core temperature, and use a conical indenter with an angle of 60 degrees to perform a shallow indentation test on the sample (the ratio of the indentation depth h to the average scale L of the microstructure of the core sample is greater than 5, and the ratio to the side length b of the cross section of the sample is less than 0.2), obtain the indentation load-displacement curve, and obtain the equivalent elastic model of the core according to Formula 2-Formula 6.

[0028]

[0029] Where: E,υ are Young's modulus and Poisson's ratio of soft rock respectively;

[0030] E i ,v i are Young’s modulus and Poisson’s ratio of the indenter, respectively;

[0031] E * Equivalent Young's modulus;

[0032] The slope of the S-unloading curve;

[0033] β is a constant related to the geometry of the indenter, and the value for a conical indenter is 1.05;

[0034] P u Head unloading pressure;

[0035] h indentation depth;

[0036] h max Maximum indentation depth;

[0037] A The contact area between the indenter and the core sample;

[0038] h c The contact depth between the indenter and the rock sample.

[0039] Step 5: Heat the sample to the original core temperature and set the minimum horizontal stress to be ρgh r ,1.2ρgh r ,1.6ρgh r ,2ρgh r ,2.4ρgh r , the maximum horizontal principal stress is 0.8ρgh r ,1.44ρgh r ,2.56ρgh r ,4ρgh r ,5.76ρgh r A conical indenter with a vertex angle of 60 degrees was used to perform a deep indentation test on the sample. The ratio of the indentation depth h to the side length b of the sample cross section was greater than 0.2 and less than 0.5, and a series of indentation load-displacement curves were obtained. The curvature C of the loading curve and the slope of the maximum indentation depth of the unloading curve were calculated for each load-displacement curve. and the ratio of residual work to total work Equivalent eigenvalues ​​can be found in Figure 1 , as actual training samples for machine learning.

[0040] Specifically, the curvature C of the loading curve of the indentation load-displacement curve of the core sample and the slope of the maximum indentation depth of the unloading curve are and the ratio of residual work to total work The relationship between the eigenvalues ​​and core mechanical parameters is as follows:

[0041]

[0042] Step 6: If Figure 2 , an inverse model of geostress problem is constructed based on Bayesian neural network. The above actual data training set is expanded by generative adversarial network. The training set is input into the neural network for network training, and the network weights of the Bayesian neural network are learned by minimizing the KL divergence of the variational distribution of the Bayesian neural network and the KL divergence of the posterior distribution of the Bayesian neural network, and the inverse model of geostress problem is obtained.

[0043] Step 7: Perform an in-situ indentation test to obtain the indentation load-displacement curve, and calculate the curvature C0 of the loading curve and the slope of the maximum indentation depth of the unloading curve. and the ratio of residual work to total work Equivalent eigenvalues ​​are input into the trained neural network to obtain the dimensionless value of the maximum horizontal principal stress output value and the dimensionless value of the minimum horizontal principal stress According to the equivalent elastic model E obtained in the fourth step * , calculate the minimum horizontal principal stress σ h and the maximum horizontal principal stress σ H .

[0044] Step 8: Use imaging logging instruments to measure the direction of the main crack at the bottom of the borehole, which is the direction of the maximum horizontal principal stress.

[0045] Borehole ultrasonic imaging has the characteristics of high resolution, high precision, and can obtain clear images even in turbid well fluids. It can provide a large amount of useful information required for basic engineering analysis, and therefore has been widely used. For example, the HIRAT (High Resolution Acoustic Televiewer) of the borehole imaging system of Robertson Geologging in the UK. High-resolution acoustic television is a device specially designed to provide qualitative images of the borehole wall. Because it uses ultrasound instead of visible light, it can work in turbid well water, expanding its scope of application. At the same time, due to its high resolution and accurate azimuth determination, it is particularly suitable for determining the details of the inclination, dip and fragmentation development of rock fractures, analyzing geological structures, and providing a reliable basis for geological, hydrogeological research or major engineering evaluation. The main parts of the logging system are as follows: Figure 1 The Micrologger2 is the most powerful portable logging system on the market today, equipped with a USB port for laptop connection, dual DSP (digital signal processing) processors and built-in drilling video support module. It is lightweight, smaller than an average laptop, and powerful, supporting all RG probes and cameras, including the latest acoustic and optical imaging systems. Micrologger2 only needs a PC, a probe and a winch from RG or a third party to provide high-quality logging and drilling video data in any situation.

[0046] The embodiments described above are only descriptions of the preferred modes of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations, modifications, and substitutions made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for measuring ground stress based on indentation technology and machine learning, characterized in that: The following steps are involved: Step 1: Drill the core and record the original core temperature, and store the core in a moisturizing manner; Step 2: Processing the core into samples; Step 3: Calculate the pressure of the overlying rock layer above the core; Step 4: Heat the core sample to the original core temperature, perform a shallow indentation test on the sample with a conical indenter, obtain the indentation load-displacement curve, and obtain the core equivalent elastic model; Step 5: Heat the sample to the original core temperature, set the minimum horizontal stress and the maximum horizontal principal stress, perform a deep indentation test on the sample with a conical indenter, and use the test data as the actual training sample for machine learning; Step 6: Construct a neural network training set based on actual test data; input the training set into the neural network for network training to obtain an inverse problem model of geostress; Step 7: Conduct an in-situ indentation test to obtain an indentation load-displacement curve, calculate the curvature of the loading curve of the in-situ indentation load-displacement curve, the slope of the maximum indentation depth of the unloading curve, and the ratio of residual work to total work; input the trained neural network to obtain the output value of the maximum horizontal principal stress dimensionless value and the minimum horizontal principal stress dimensionless value, and calculate the minimum horizontal principal stress and the maximum horizontal principal stress according to the core equivalent elastic model obtained in the fourth step; Step 8: Use imaging logging instruments to measure the main crack direction of the borehole bottom indentation, which is the direction of the maximum horizontal principal stress.

2. The method for measuring ground stress based on indentation technology and machine learning according to claim 1, characterized in that: In step 3, the pressure of the overlying rock layer above the core is calculated according to formula (1); p=ρgh r (1) Where ρ is the rock density, g is the gravitational acceleration, and h is r is the thickness of the overlying rock.

3. The method for measuring ground stress based on indentation technology and machine learning according to claim 1, characterized in that: In step 4, the equivalent elastic model of the core is obtained by the following formula; Where: E and v are Young's modulus and Poisson's ratio of soft rock respectively; E i ,v i are Young’s modulus and Poisson’s ratio of the indenter, respectively; E * Equivalent Young's modulus; The slope of the S-unloading curve; β is a constant related to the geometry of the indenter; P u Head unloading pressure; h indentation depth; h max Maximum indentation depth; A The contact area between the indenter and the core sample; h c The contact depth between the indenter and the rock sample.

4. The method for measuring ground stress based on indentation technology and machine learning according to claim 3 is characterized in that: In step five, a series of indentation load-displacement curves are obtained through deep indentation tests; the curvature of the displacement curve, the maximum indentation depth slope of the unloading curve, and the ratio of residual work to total work of each load are calculated and used as actual training samples for machine learning.

5. The method for measuring ground stress based on indentation technology and machine learning according to claim 4 is characterized in that: Core sample indentation load-displacement curve loading curve curvature C, unloading curve maximum indentation depth slope and the ratio of residual work to total work The relationship with the core mechanical parameters is as follows:

6. The method for measuring ground stress based on indentation technology and machine learning according to claim 1, characterized in that: Step six includes: constructing an inverse geostress problem model based on a Bayesian neural network; using a generative adversarial network to expand actual training samples into a training set; inputting the training set into the neural network for network training, and learning the network weights of the Bayesian neural network by minimizing the KL divergence of the variational distribution of the Bayesian neural network and the KL divergence of the posterior distribution of the Bayesian neural network to obtain an inverse geostress problem model.

7. The method for measuring ground stress based on indentation technology and machine learning according to claim 1, characterized in that: When the shallow indentation test is performed in step 4, the ratio of the indentation depth h to the average microstructure scale L of the core sample is greater than 5, and the ratio to the side length b of the sample cross section is less than 0.

2.

8. The method for measuring ground stress based on indentation technology and machine learning according to claim 1, characterized in that: In step 5, the minimum horizontal stresses are set as: ρgh r ,1.2ρgh r ,1.6ρgh r ,2ρgh r ,2.4ρgh r The maximum horizontal principal stresses are set as follows: 0.8ρgh r ,1.44ρgh r ,2.56ρgh r ,4ρgh r ,5.76ρgh r Where ρ is the rock density, g is the gravitational acceleration, h r is the thickness of the overlying rock.

Images