Rock mechanical parameter acquisition method based on sound wave detection and artificial intelligence

Abstract

The invention discloses a rock mechanical parameter acquisition method based on acoustic detection and artificial intelligence. The method comprises the following steps: step 1, constructing a BP neural network; step 2, constructing a rock mass mechanical parameter database, and inputting mechanical parameters in the database into the BP neural network training in the step 1 to obtain a trained BP neural network training; step 3, obtaining the rock of the rock mass to be detected, measuring the uniaxial compressive strength of the rock mass to be detected, the longitudinal wave velocity of the rock mass to be detected and the wave velocity of the rock mass to be detected with initial damage, and inputting the uniaxial compressive strength and the longitudinal wave velocity of the rock mass to be detected and the wave velocity of the rock mass to be detected with initial damage into the BP neural network trained in the step 2, and calculating the initial damage of the rock mass to be detected based on the trained BP neural network. And outputting the compressive strength sigma c of the rock mass and the elastic modulus E of the rock mass, wherein the compressive strength sigma c of the rock mass and the elastic modulus E of the rock mass are used as reference parameters of a tunnel excavation mode. By the adoption of the method, input parameters are easy to obtain on site, the calculation amount is small, and the method is not affected by the type and the moisture content of the rock.

Description

Technical Field

[0001] The invention belongs to the technical field of geotechnical engineering, and specifically relates to a method for acquiring rock mechanics parameters based on acoustic wave detection and artificial intelligence. Background Art

[0002] Rock wave velocity has the characteristics of convenient, non-destructive and fast testing, and it also has a good correlation with some mechanical parameters of rock mass. At present, acoustic wave testing technology has been widely used in the field of rock mechanics. By conducting indoor mechanical tests on rocks and testing the wave velocity of rocks at the same time, a quantitative relationship between wave velocity and rock mechanical parameters such as compressive strength σc and elastic modulus E can be established. However, the current research on the relationship between rock acoustic wave velocity and mechanical parameters is mainly based on empirical formulas and regression fitting analysis. The empirical formula established in this way is only applicable to a single condition, that is, a single rock type and a single water content state. As long as the rock type and water content state change, the empirical formula established before is no longer applicable. For geotechnical engineering in a complex geological environment, the strata in the work area are usually mixed with multiple rocks and the water content conditions are variable. In this case, a large number of empirical formulas need to be established, the calculation process is complicated, and in the actual calculation, they need to be corrected according to the type and water content of the rock. In addition, the test operation of water content is relatively complicated and can usually only be carried out in the laboratory, making it difficult to directly obtain this parameter at the construction site. Summary of the invention

[0003] The purpose of the present invention is to provide a rock mechanics parameter acquisition method based on acoustic wave detection and artificial intelligence, in which the input parameters are easy to obtain on site, the amount of calculation is small, and the method is not affected by the type and water content of the rock.

[0004] The present invention adopts the following technical solution: a method for obtaining rock mechanical parameters based on acoustic wave detection and artificial intelligence, comprising the following steps:

[0005] Step 1: Construct BP neural network;

[0006] Step 2: construct a rock mass mechanical parameter database, input the mechanical parameters in the database into the BP neural network training in step 1, and obtain a trained BP neural network training;

[0007] Step 3: Obtain rock samples of the rock mass to be tested, measure the uniaxial compressive strength of the rock samples and the longitudinal wave velocity of the rock to be tested, as well as the wave velocity of the rock mass to be tested with initial damage, and use them as inputs into the BP neural network trained in step 2. Based on the trained BP neural network, output the compressive strength σ of the rock mass. c and the elastic modulus E of the rock mass, the compressive strength σ of the rock mass c and the elastic modulus E of the rock mass are used as reference parameters for the tunnel excavation method.

[0008] Furthermore, the uniaxial compressive strength of the rock sample of the rock mass to be tested is measured by a uniaxial compression test.

[0009] Furthermore, the construction process of the rock mass mechanics parameter database in step 2 is as follows:

[0010] Collect rock samples of all types of rock masses. For any type of rock sample, at least three water-containing states and ten damage states are tested to obtain the mechanical parameters of the rock samples under a total of thirty different working conditions. The mechanical parameters are used as inputs of the neural network.

[0011] The three moisture states are represented as: S-1, S-2 and S-3, where S-1 represents the dry state, that is, the moisture content is 0%; S-2 represents the semi-saturated state, that is, the moisture content is 50%; S-3 represents the fully saturated state, that is, the moisture content is 100%;

[0012] The ten damage states are represented as D-1 to D-10;

[0013] The definition of damage state is as follows: a rock specimen is loaded with a continuously increasing load until it cracks, and the load intensity that the rock specimen bears when it cracks is its ultimate strength; ten different load values ​​are selected from small to large between the initial load and the ultimate strength, and the magnitude of the load is based on the ultimate strength of the rock specimen, which is a few percentages of the ultimate strength; the ten different loads correspond to ten damage states of the rock specimen, D-1 to D-10, among which D-1 means no initial damage, which is the damage state under the initial load value.

[0014] Furthermore, the mechanical parameters of the rock sample are: the uniaxial compressive strength of the rock sample, the longitudinal wave velocity of the rock sample and the wave velocity of the rock sample with initial damage.

[0015] Furthermore, ten different load values ​​are selected as follows: the critical point of the plastic stage is used as the initial load value, 95% of the ultimate strength is used as the final load value, and the other load values ​​are selected in sequence within the range of 70% to 95% of the ultimate strength.

[0016] Furthermore, the mechanical parameters of the rock samples are obtained as follows:

[0017] The uniaxial compressive strength of the rock sample is obtained in the following manner: the compressive strength of the rock sample is measured by a point load test, and the compressive strength of the rock sample is used as the uniaxial compressive strength of the rock sample used as input;

[0018] The wave velocity of the rock sample with initial damage is obtained by pre-compressing and cracking the rock sample to obtain the rock sample with initial damage, and testing the wave velocity by the seismic wave method to obtain the result.

[0019] The present invention also discloses a method for establishing a rock mass mechanics parameter database in the above-mentioned rock mass mechanics parameter acquisition method based on acoustic wave detection and artificial intelligence, comprising the following steps:

[0020] Collect rock samples of all types of rock masses. For any type of rock sample, at least three water-containing states and ten damage states are tested to obtain a total of thirty mechanical parameters under different working conditions. The mechanical parameters are used as inputs to the neural network.

[0021] The three moisture states are represented as: S-1, S-2 and S-3, where S-1 represents the dry state, that is, the moisture content is 0%; S-2 represents the semi-saturated state, that is, the moisture content is 50%; S-3 represents the fully saturated state, that is, the moisture content is 100%;

[0022] The ten damage states are represented as D-1 to D-10;

[0023] The definition of damage state is as follows: a rock specimen is loaded with a continuously increasing load until it cracks, and the load intensity that the rock specimen bears when it cracks is its ultimate strength; ten different load values ​​are selected from small to large between the initial load and the ultimate strength, and the magnitude of the load is based on the ultimate strength of the rock specimen, which is a few percentages of the ultimate strength; the ten different loads correspond to ten damage states of the rock specimen, D-1 to D-10, among which D-1 means no initial damage, which is the damage state under the initial load value.

[0024] The beneficial effect of the present invention is that it is a feasible method for calculating rock mechanical parameters of various lithologies and water content conditions without having to establish multiple empirical formulas according to the type and water content of the rock, and the initial strength of the rock σ c0 , initial longitudinal wave velocity v of rock p0 , and the current rock longitudinal wave velocity v p As input parameter, we get the rock strength σ c The amount of calculation is reduced, and the way to obtain the input parameters is relatively simple. It can be obtained through triaxial compression test and wave velocity, and it is easy to carry out on-site for a long time. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is the relationship between rock compressive strength and longitudinal wave velocity;

[0026] Figure 2 is the relationship between rock elastic modulus and P-wave velocity;

[0027] Figure 3 The relationship between rock compressive strength and longitudinal wave velocity for different rocks and different water content states;

[0028] Figure 4 Data presentation for calculating rock strength through neural network;

[0029] Figure 5 Data display of calculating rock elastic modulus through neural network. DETAILED DESCRIPTION

[0030] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] There is an obvious correlation between the wave velocity of rock and its mechanical parameters. When the lithology and water content remain unchanged, the compressive strength of rock and wave velocity have the following relationship: Figure 1 At the same time, there is a similar relationship between the elastic modulus of rock and wave velocity, such as Figure 2 Therefore, a fitting formula can be established through data statistics, which is also a commonly used method at present.

[0032] However, once the lithology or water content changes, the relationship between wave velocity and intensity will change, such as Figure 3 As shown, the fitting formula is no longer applicable. When the water content of the same rock changes, on the one hand, because the propagation speed of sound waves in water (1500m / s) is much greater than the propagation speed in air (340m / s), and on the other hand, because the rock softens when it encounters water, there will be a change in wave velocity with higher speed and lower intensity. For another type of rock, due to different rock components and porosity, the relationship between wave velocity and intensity will also change significantly. Similarly, the relationship between wave velocity and elastic modulus also has such a problem.

[0033] The present invention provides a rock mechanics parameter acquisition method based on acoustic wave detection and artificial intelligence. It uses artificial intelligence algorithms to aggregate test data to establish a database, and uses these data to train neural networks. After the neural network training is completed, the corresponding rock strength and elastic modulus can be calculated after inputting the wave velocity and auxiliary control parameters. The method in the present invention is applicable to rocks of various lithologies and water-bearing conditions. The auxiliary control parameter selected is the uniaxial compressive strength σ of the intact rock, that is, the undamaged rock core. c0 and the P-wave velocity v of intact rock p0 , obtained in the laboratory through uniaxial compression test and wave velocity detection. It is necessary to pre-compress and crack the rock specimen to obtain the rock with initial damage and detect its wave velocity v p .

[0034] The present invention also discloses a method for obtaining rock mass mechanical parameters based on acoustic wave detection and artificial intelligence, comprising the following steps:

[0035] Step 1: Construct BP neural network;

[0036] Step 2: construct a rock mass mechanical parameter database, input the mechanical parameters in the database into the BP neural network training in step 1, and obtain a trained BP neural network training; a CNN neural network may also be used;

[0037] Step 3: Obtain rock samples of the rock mass to be tested, measure the uniaxial compressive strength of the rock samples of the rock mass to be tested, the longitudinal wave velocity of the rock to be tested, and the wave velocity of the rock mass to be tested with initial damage, and input them into the BP neural network trained in step 2. Based on the trained BP neural network, output the compressive strength σ of the rock mass. c and the elastic modulus E of the rock mass, the compressive strength σ of the rock mass c The elastic modulus E of the rock mass is used as a reference parameter for the tunnel excavation method. The uniaxial compressive strength of the rock sample to be tested is measured by a uniaxial compression test. The wave velocity of the rock mass to be tested with initial damage is measured by the seismic wave method.

[0038] The construction process of the rock mass mechanics parameter database in step 2 is as follows:

[0039] Collect rock samples of all types of rock masses, and test at least three water-containing states and ten damage states for any type of rock sample, so as to obtain mechanical parameters of the rock samples under a total of thirty different working conditions; the mechanical parameters are used as inputs of the neural network;

[0040] The three water content states are represented in sequence as follows: S-1, S-2 and S-3, wherein S-1 represents a dry state, i.e., a water content of 0%; S-2 represents a semi-saturated state, i.e., a water content of 50%; and S-3 represents a fully saturated state, i.e., a water content of 100%.

[0041] The ten damage states are represented as D-1 to D-10 in sequence;

[0042] The damage state is defined as follows: a rock sample is loaded with a continuously increasing load until it cracks, and the load intensity borne by the rock sample when it cracks is its ultimate strength; ten different load values ​​are selected from small to large between the initial load and the ultimate strength, and the magnitude of the load is based on the ultimate strength of the rock sample, and is a percentage of the ultimate strength; the ten different loads correspond to ten damage states D-1 to D-10 of the rock sample in turn, among which D-1 means no initial damage, which is the damage state under the initial load value.

[0043] The above ten different load values ​​are selected as follows: the critical point of the plastic stage is taken as the initial load value, 95% of the ultimate strength is taken as the final load value, and the other load values ​​are selected in sequence within the range of 70% to 95% of the ultimate strength.

[0044] Specifically, the value of the critical point in the plastic stage is generally about 70% of the ultimate strength. The value of the critical point in the plastic stage is taken as the initial load value. The corresponding state under the initial load value is D-1, that is, the state with no initial damage; the next load value adjacent to the initial load value, its increment can be set to 2.5%, or 5%, or other values. The increment is not fixed. The load value after the increment is within the specified range, that is, it meets the requirements, until 95% of the ultimate strength is taken as the final load value.

[0045] The mechanical parameters of the above rock samples are: uniaxial compressive strength of the rock samples, longitudinal wave velocity of the rock samples and wave velocity of the rock samples with initial damage.

[0046] The uniaxial compressive strength of the rock sample is obtained by measuring the compressive strength of the rock sample through a point load test, and using the compressive strength of the rock sample as the uniaxial compressive strength for input. The wave velocity of the rock sample with initial damage is obtained by pre-compression cracking the rock sample to obtain a rock sample with initial damage, and testing the wave velocity by the seismic wave method.

[0047] The BP neural network training input layer has three input variables, namely the mechanical parameters mentioned above. The number of nodes in the hidden layer of the BP neural network is 10. Increasing the number of nodes has no significant effect on the accuracy of the predicted data. The output layer has two output variables, the compressive strength of the rock mass σ c and the elastic modulus E of the rock mass.

[0048] The present invention discloses a method for establishing a rock mass mechanics parameter database in the above-mentioned rock mass mechanics parameter acquisition method based on acoustic wave detection and artificial intelligence, comprising the following steps:

[0049] Collect rock samples of all types of rock masses. For any type of rock sample, at least three water-containing states and ten damage states are tested to obtain a total of thirty mechanical parameters under different working conditions. The mechanical parameters are used as inputs to the neural network.

[0050] The three moisture states are represented as: S-1, S-2 and S-3, where S-1 represents the dry state, that is, the moisture content is 0%; S-2 represents the semi-saturated state, that is, the moisture content is 50%; S-3 represents the fully saturated state, that is, the moisture content is 100%;

[0051] The ten damage states are represented as D-1 to D-10;

[0052] The definition of damage state is as follows: a rock specimen is loaded with a continuously increasing load until it cracks, and the load intensity that the rock specimen bears when it cracks is its ultimate strength; ten different load values ​​are selected from small to large between the initial load and the ultimate strength, and the magnitude of the load is based on the ultimate strength of the rock specimen, which is a few percentages of the ultimate strength; the ten different loads correspond to ten damage states of the rock specimen, D-1 to D-10, among which D-1 means no initial damage, which is the damage state under the initial load value.

[0053] The pre-compression cracked rock specimens are collectively referred to as rocks with initial damage, and the pre-compression cracked rocks with initial damage are used to simulate the rock mass.

[0054] A database was established by using pre-compression cracked rock simulation with initial damage to predict the mechanical parameters of rock mass at the construction site. The mechanical parameters under the above thirty different working conditions are as follows: D-1~D-10 under S-1 state, ten corresponding working conditions are obtained; D-1~D-10 under S-2 state, ten corresponding working conditions are obtained; D-1~D-10 under S-3 state, ten corresponding working conditions are obtained; each working condition corresponds to a set of mechanical parameters, and each set of mechanical parameters includes: the uniaxial compressive strength σ of intact rock c0 , the longitudinal wave velocity v of intact rock p0 and the wave velocity v of the rock mass with initial damage p The D-1 group of rocks has no initial damage, so its longitudinal wave velocity and compressive strength are consistent with the initial state, that is, v p =v p0 , σ c =σ c0 .

[0055] As a specific example, samples of all rocks included in the work area were collected, and data as shown in Table 1 were obtained through indoor uniaxial compression tests and longitudinal wave velocity detection, with 30 sets of data for each type of rock. All data were used as training sets to train the neural network. The mechanical parameters are as follows:

[0056] Table 1 Examples of rock groups and related parameters

[0057]

[0058]

[0059]

[0060] By using the method of the present invention, since no drilling and coring is performed, the water content of the rock is the natural water content. In this way, the uniaxial compressive strength σ of the intact rock under the current rock mass water content condition can be obtained. c0 , the longitudinal wave velocity v of intact rockp0 In the subsequent prediction of rock mechanical parameters, the influence of water content does not need to be considered when collecting rock data on site.

[0061] When it is necessary to predict the rock mechanical parameters on site, it is necessary to first identify the type of rock, and then input the uniaxial compressive strength of this type of rock and the longitudinal wave velocity of this type of rock based on the existing test data, and then perform wave velocity detection on the rock on site to obtain the wave velocity of the rock mass with initial damage. For the rock on site, the wave velocity of the rock mass with initial damage is obtained in the following way: test the wave velocity by the seismic wave method. When all three data are obtained, the data of the input layer can be imported based on the trained neural network, that is, the uniaxial compressive strength σ of this type of rock c0 , the longitudinal wave velocity v of this type of rock p0 , wave velocity v of rock mass with initial damage p , the corresponding stress value is calculated by the trained neural network without manual calculation. The longitudinal wave velocity v of this type of rock p0 is the initial longitudinal wave velocity.

[0062] like Figure 4 As shown, the black dots in the figure are training data. This embodiment tests three different rocks in total. According to the principle of data selection and grouping, 30 sets of data for each rock, a total of 90 sets of data. Contains wave velocity, strength and elastic modulus under different water-containing conditions of the three rocks. After the neural network training is completed, 9 rock data are selected to test the accuracy of the method in the present invention. The red dots in the figure represent the predicted data, and the blue dots represent the real data obtained through indoor uniaxial compression tests. The data of the two are relatively close, indicating that the method in the present invention has good calculation accuracy.

[0063] The elastic modulus is calculated in the same way, such as Figure 5 As shown, the black points are training data, the red points represent predicted data, and the blue points represent real data obtained through indoor uniaxial compression tests. The data of the two are relatively close, indicating that the method of the present invention has good calculation accuracy.

Claims

1. A method for obtaining rock mechanical parameters based on acoustic wave detection and artificial intelligence, characterized in that: The steps include: Step 1: Construct BP neural network; Step 2: construct a rock mass mechanical parameter database, input the mechanical parameters in the database into the BP neural network training in step 1, and obtain a trained BP neural network training; Step 3: Obtain rock samples of the rock mass to be tested, measure the uniaxial compressive strength of the rock samples of the rock mass to be tested, the longitudinal wave velocity of the rock to be tested, and the wave velocity of the rock mass to be tested with initial damage, and input them into the BP neural network trained in step 2. Based on the trained BP neural network, output the compressive strength σ of the rock mass. c and the elastic modulus E of the rock mass, the compressive strength σ of the rock mass c and the elastic modulus E of the rock mass are used as reference parameters for the tunnel excavation method.

2. A method for obtaining rock mass mechanical parameters based on acoustic wave detection and artificial intelligence as claimed in claim 1, characterized in that: The uniaxial compressive strength of the rock sample to be tested is measured by a uniaxial compressive test.

3. A method for obtaining rock mass mechanical parameters based on acoustic wave detection and artificial intelligence as claimed in claim 2, characterized in that: The construction process of the rock mass mechanics parameter database in step 2 is as follows: Collect rock samples of all types of rock masses, and test at least three water-containing states and ten damage states for any type of rock sample, so as to obtain mechanical parameters of the rock samples under a total of thirty different working conditions; the mechanical parameters are used as inputs of the neural network; The three water content states are represented in sequence as follows: S-1, S-2 and S-3, wherein S-1 represents a dry state, i.e., a water content of 0%; S-2 represents a semi-saturated state, i.e., a water content of 50%; and S-3 represents a fully saturated state, i.e., a water content of 100%. The ten damage states are represented as D-1 to D-10 in sequence; The damage state is defined as follows: a rock sample is loaded with a continuously increasing load until it cracks, and the load intensity borne by the rock sample when it cracks is its ultimate strength; ten different load values ​​are selected from small to large between the initial load and the ultimate strength, and the magnitude of the load is based on the ultimate strength of the rock sample, and is a percentage of the ultimate strength; the ten different loads correspond to ten damage states D-1 to D-10 of the rock sample in turn, among which D-1 means no initial damage, which is the damage state under the initial load value.

4. A method for obtaining rock mass mechanical parameters based on acoustic wave detection and artificial intelligence as claimed in claim 3, characterized in that: The mechanical parameters of the rock sample are: the uniaxial compressive strength of the rock sample, the longitudinal wave velocity of the rock sample and the wave velocity of the rock sample with initial damage.

5. A method for obtaining rock mass mechanical parameters based on acoustic wave detection and artificial intelligence as claimed in claim 4, characterized in that: The ten different load values ​​are selected as follows: the critical point of the plastic stage is used as the initial load value, 95% of the ultimate strength is used as the final load value, and the other load values ​​are selected in sequence within the range of 70% to 95% of the ultimate strength.

6. A method for obtaining rock mass mechanical parameters based on acoustic wave detection and artificial intelligence as claimed in claim 5, characterized in that: The mechanical parameters of the rock sample are obtained in the following way: The uniaxial compressive strength of the rock sample is obtained in the following manner: the compressive strength of the rock sample is measured by a point load test, and the compressive strength of the rock sample is used as the uniaxial compressive strength of the rock sample for input; The wave velocity of the rock sample with initial damage is obtained by pre-compressing and cracking the rock sample to obtain the rock sample with initial damage, and testing the wave velocity by seismic wave method.

7. A method for establishing a rock mass mechanics parameter database in a rock mass mechanics parameter acquisition method based on acoustic wave detection and artificial intelligence as claimed in any one of claims 1 to 6, characterized in that: These include: Collect rock samples of all types of rock masses, and test at least three water-containing states and ten damage states for any type of rock sample, to obtain mechanical parameters under a total of thirty different working conditions; the mechanical parameters are used as inputs of the neural network; The three water content states are represented in sequence as follows: S-1, S-2 and S-3, wherein S-1 represents a dry state, i.e., a water content of 0%; S-2 represents a semi-saturated state, i.e., a water content of 50%; and S-3 represents a fully saturated state, i.e., a water content of 100%. The ten damage states are represented as D-1 to D-10 in sequence; The damage state is defined as follows: a rock sample is loaded with a continuously increasing load until it cracks, and the load intensity borne by the rock sample when it cracks is its ultimate strength; ten different load values ​​are selected from small to large between the initial load and the ultimate strength, and the magnitude of the load is based on the ultimate strength of the rock sample, and is a percentage of the ultimate strength; the ten different loads correspond to ten damage states D-1 to D-10 of the rock sample in turn, among which D-1 means no initial damage, which is the damage state under the initial load value.

Images