Method and system for determining correlation between flatness of end face of rock sample and mechanical index

Abstract

The application provides a method and system for determining the correlation between the end face flatness of a rock sample and mechanical indexes. The method includes: preparing multiple rock samples with different end face flatness, collecting multi-dimensional flatness parameters of the samples, and measuring multiple mechanical indexes, water content, and density of the samples to generate a multi-dimensional data set; constructing a multi-dimensional input feature vector; constructing a correlation model by integrating a random forest model and a BP neural network, training the correlation model with the multi-dimensional data set as training data, and optimizing the model hyperparameters through cross-validation; constructing a differentiated correction model to correct the output of the correlation model for different lithology to obtain a mechanical index prediction curve corresponding to the multi-dimensional input feature vector of the rock sample to be tested. The method establishes a quantitative relationship between the end face flatness of the rock and the mechanical indexes, and can effectively avoid the influence of the end face flatness on the rock mechanical test.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rock mechanics testing, and particularly relates to a method and system for determining the correlation between the end surface flatness of a rock sample and mechanical indexes. BACKGROUND

[0002] At present, in rock mechanics testing, the end surface flatness of a rock sample is a key factor affecting the test results, and it is necessary to determine the quantitative relationship between the end surface flatness and the mechanical indexes of the rock, so as to reduce the errors in the rock mechanics test results.

[0003] However, in the related art, only the influence of joint roughness on rock strength is studied, and the influence mechanism of the end surface flatness on the mechanical indexes of the rock has not been systematically studied, and only the tolerance range of the end surface flatness is specified, resulting in a large discreteness of the rock mechanics test results, and there may be errors in the test results affected by the end surface flatness.

[0004] Therefore, how to establish the correlation between the standard rock end surface flatness and the mechanical indexes has become a problem to be solved at present. SUMMARY

[0005] The present application aims to at least partly solve one of the technical problems in the related art.

[0006] To this end, a first object of the present application is to provide a method for determining the correlation between the end surface flatness of a rock sample and mechanical indexes, which quantifies the end surface flatness parameters of the rock sample, constructs a flatness and mechanical index correlation model and an engineering database, establishes a quantitative relationship between the rock end surface flatness and the mechanical indexes, and can effectively avoid the influence of the end surface flatness on the rock mechanics test, and can be applied to various scenes such as rock mechanics experiments and engineering rock mass stability evaluation.

[0007] A second object of the present application is to provide a system for determining the correlation between the end surface flatness of a rock sample and mechanical indexes.

[0008] A third object of the present application is to provide a non-transitory computer readable storage medium.

[0009] To achieve the above objects, a first aspect of the present application provides a method for determining the correlation between the end surface flatness of a rock sample and mechanical indexes, comprising the following steps:

[0010] A plurality of rock samples with different end surface flatnesses are prepared by using a flatness preparation process, multi-dimensional flatness parameters of the plurality of rock samples are collected, and a plurality of mechanical indexes, water content and density of the plurality of rock samples are determined to generate a multi-dimensional data set;

[0011] construct a multi-dimensional input feature vector based on the flatness parameter, lithology, water content and density;

[0012] An association model is constructed by integrating the random forest model and the BP neural network, the multi-dimensional data set is taken as training data to train the association model, and model hyperparameters are optimized through cross-validation to dynamically adjust feature weights for lithology, wherein the association model takes the multi-dimensional input feature vector as input to predict mechanical indexes corresponding to the input features;

[0013] A differentiated correction model is constructed, the output of the association model is corrected through the differentiated correction model for different lithologies, and based on the differentiated correction model, a mechanical index prediction curve corresponding to the multi-dimensional input feature vector of the rock sample to be tested is obtained.

[0014] Optionally, in an embodiment of the present application, the association model is specifically used to: identify the interaction effect between different parameters in the multi-dimensional input feature vector through a multivariate adaptive regression spline (MARS) algorithm.

[0015] Optionally, in an embodiment of the present application, the dynamic adjustment of the feature weights for lithology includes: setting weight adjustment factors for water content and density for lithology, wherein the weight of water content is increased when the rock sample is soft rock, and the weight of density is increased when the rock sample is hard rock; and setting weight threshold values for water content and density in the random forest model.

[0016] Optionally, in an embodiment of the present application, the correction of the output of the association model for different lithologies through the differentiated correction model includes: data layering of the multi-dimensional input feature vector and standardization processing of the layered data; for soft rock, a first correction formula is constructed with uniaxial compressive strength as the dependent variable and the standardized roughness parameter, water content and density as the independent variables, wherein the first correction formula contains a plurality of lithology-related coefficients to be solved; the first correction formula is subjected to ridge regression fitting to construct a loss function containing an L2 regularization term; the regularization parameter in the loss function is optimized through 5-fold cross-validation to obtain a plurality of fitted lithology-related coefficients.

[0017] Optionally, in an embodiment of the present application, after the layered data is normalized, the method further comprises: for hard rock, screening a plurality of target features sensitive to hard rock from the normalized data, and constructing a second correction formula with uniaxial compressive strength as the dependent variable and the plurality of target features as the independent variable, wherein the second correction formula contains a plurality of to-be-solved lithology-related coefficients; performing least squares fitting on the second correction formula, and constructing a three-dimensional particle flow model PFC, and performing discrete element simulation through the three-dimensional particle flow model PFC to calibrate the lithology-related coefficients corresponding to the density; and expanding the data in the training data set based on the discrete element simulation, and correcting the correlation model through the expanded training data set.

[0018] Optionally, in an embodiment of the present application, the method further comprises: receiving the multi-dimensional input feature vector of the rock sample to be tested through a visual query platform; calling the trained prediction model to output the mechanical index prediction value corresponding to the received multi-dimensional input feature vector, wherein the prediction model comprises the correlation model and the differential correction model; and within a preset flatness range, generating equidistant data points from the mechanical index prediction value at a unit step length, and obtaining a mechanical index prediction curve through difference fitting.

[0019] Optionally, in an embodiment of the present application, after the method of obtaining the mechanical index prediction curve corresponding to the multi-dimensional input feature vector of the rock sample to be tested, the method further comprises: displaying the mechanical index prediction curve through an interactive interface of the visual query platform; and generating a strength correction coefficient table within the preset flatness range.

[0020] Optionally, in an embodiment of the present application, the method of preparing a plurality of rock samples with different end face flatnesses by using a flatness preparation process comprises: adjusting the end face flatness of different rock samples in combination with high-precision grinding and polishing equipment and a pressure sensor.

[0021] To achieve the above purpose, a second aspect of the present application further provides a system for determining the correlation between the end face flatness of a rock sample and a mechanical index, comprising the following modules:

[0022] A collection module is configured to prepare a plurality of rock samples with different end face flatnesses by using a flatness preparation process, collect multi-dimensional flatness parameters of the plurality of rock samples, and measure a plurality of mechanical indexes, water content, and density of the plurality of rock samples to generate a multi-dimensional data set.

[0023] A construction module is configured to construct a multi-dimensional input feature vector based on the flatness parameters, lithology, water content, and density.

[0024] A prediction module is configured to construct a correlation model by integrating a random forest model and a BP neural network, train the correlation model by taking the multidimensional data set as training data, and optimize model hyperparameters by cross-validation to dynamically adjust feature weights for lithology, wherein the correlation model takes the multidimensional input feature vector as input to predict a mechanical index corresponding to the input feature.

[0025] A correction module is configured to construct a differentiated correction model, correct the output of the correlation model by the differentiated correction model for different lithologies, and obtain a mechanical index prediction curve corresponding to the multidimensional input feature vector of the rock sample to be tested based on the differentiated correction model.

[0026] To implement the above-mentioned embodiments, the third aspect of the present application further proposes a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the method for determining the correlation between the flatness of the end surface of the rock sample and the mechanical index in the first aspect.

[0027] The technical scheme provided by the embodiments of the present application at least brings the following beneficial effects: the present application constructs a multidimensional feature vector combining three-dimensional morphology and physical properties in rock mechanics testing, improves the feature system, and combines machine learning and experimental data for correlation prediction, which has higher prediction accuracy than traditional statistical regression models, and realizes the leap from qualitative control of flatness parameters to quantitative prediction of mechanical indexes. Moreover, the present application introduces dynamic weights, cross-domain feature fusion and explainability analysis in the prediction model, improving the adaptability of machine learning models in geotechnical engineering. The feature importance dynamic adjustment mechanism sensitive to lithology is introduced in the random forest algorithm, which is more in line with the objective law of the sensitivity of lithology dominant parameters in rock mechanics than the general machine learning framework. Furthermore, the present application also constructs an engineering database, which can be directly applied to rock mechanics testing laboratories and engineering rock mass stability evaluation scenarios, reducing repeated experiments and test costs caused by flatness errors. Thus, the present application establishes a quantitative relationship between rock end surface flatness and mechanical indexes, which can effectively avoid the influence of end surface flatness on rock mechanics testing, improve the accuracy of quantitative relationship prediction, and is conducive to improving engineering safety and economy.

[0028] Additional aspects and advantages of the application will be set forth in part in the description that follows, and in part will become apparent to those skilled in the art upon examination of the following and / or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0029] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description, taken in conjunction with the accompanying drawings, in which:

[0030] Figure 1A flow chart of a method for determining a correlation between end face flatness and mechanical indexes of a rock sample according to an embodiment of the present application is shown in FIG. 1.

[0031] Figure 2 A flow chart of a differential correction method according to an embodiment of the present application is shown in FIG. 2.

[0032] Figure 3 A flow chart of another differential correction method according to an embodiment of the present application is shown in FIG. 3.

[0033] Figure 4 A flow chart of a method for generating and displaying a mechanical index prediction curve according to an embodiment of the present application is shown in FIG. 4.

[0034] Figure 5 A structural schematic diagram of a system for determining a correlation between end face flatness and mechanical indexes of a rock sample according to an embodiment of the present application is shown in FIG. 5. DETAILED DESCRIPTION

[0035] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which the same or similar elements or elements having the same or similar functions are denoted by the same reference numerals throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0036] A method and system for determining a correlation between end face flatness and mechanical indexes of a rock sample according to an embodiment of the present application are described below with reference to the accompanying drawings.

[0037] Figure 1 A flow chart of a method for determining a correlation between end face flatness and mechanical indexes of a rock sample according to an embodiment of the present application is shown in FIG. 1. Figure 1 The method includes the following steps:

[0038] In step S101, a plurality of rock samples with different end face flatness are prepared by using a flatness preparation process, multi-dimensional flatness parameters of the plurality of rock samples are collected, and a plurality of mechanical indexes, water content and density of the plurality of rock samples are determined to generate a multi-dimensional data set.

[0039] Specifically, the present application first prepares rock samples with different end face flatness by using a flatness preparation process.

[0040] In an embodiment of the present application, a plurality of rock samples with different end face flatness are prepared by using a flatness preparation process, including: adjusting the end face flatness of different rock samples in combination with high-precision grinding and polishing equipment and a pressure sensor.

[0041] Specifically, the end face of the sample is polished by using a high-precision polishing equipment (such as a numerical control grinding machine), and the pressure applied by the polishing equipment is monitored in real time by combining a pressure sensor (monitoring accuracy ±0.1N). In addition, the process parameters such as polishing time (for example, 5-30min) and grinding wheel mesh (for example, 80-1000 mesh) are controlled to realize the preparation of rock samples with different flatness.

[0042] Further, the multi-dimensional flatness parameters of the prepared multiple rock samples are collected by using a multi-dimensional detection method. For example, the end face morphology of the sample can be obtained by using a three-dimensional laser scanner (accuracy ±5μm), and then the roughness parameters (including arithmetic average roughness Ra and maximum height roughness Rz), fractal dimension D and root mean square deviation σ and other flatness parameters can be calculated according to the scanning data. Thus, a basic data set containing more than 5000 samples can be established.

[0043] Further, the multiple mechanical indexes of the multiple rock samples are determined by performing corresponding mechanical experiments on the rock samples with different lithology. For example, for different rock types (such as sandstone, limestone and shale), uniaxial compression experiments (loading rate 0.05-2mm / min), triaxial compression experiments (confining pressure 0-100MPa) and Brazilian splitting experiments can be performed, and the peak strength, elastic modulus and Poisson's ratio and other mechanical indexes of each rock sample are recorded synchronously. Then, the corresponding relationship table of the flatness parameters and the mechanical indexes is generated in combination with the determined flatness parameters of each rock sample. The water content (w) of each rock sample can be determined by using the drying method, and the density (ρ) of each rock sample can be calculated by using the volume measurement method.

[0044] Thus, the multi-dimensional data set of the rock samples is obtained by using multiple methods such as measurement and experiment.

[0045] In step S102, a multi-dimensional input feature vector is constructed based on the flatness parameters, lithology, water content and density.

[0046] Specifically, for the obtained multi-dimensional data set, a multi-dimensional input feature vector containing multiple flatness parameters, lithology, water content and density is constructed by performing feature engineering, so as to determine the input features of the prediction model.

[0047] For example, the multiple flatness parameters (Ra, Rz, D and σ) are combined with the lithology (i.e. rock type, which can include sandstone, limestone and shale), water content (0-20%, which is divided into intervals according to every 5% gradient), and density (2.0-3.0g / cm 3 , accurate to 0.01g / cm 3) and normalization processing are performed to construct a 10-dimensional input feature vector. For example, 3-dimensional lithology one-hot encoding + 4-dimensional flatness + 1-dimensional water content + 1-dimensional density + 1-dimensional other data can be used.

[0048] In the present application, the water content and the density are used as independent variables to participate in the training of the mechanical index prediction model. Unlike the traditional material science model which only uses a single roughness parameter (such as Ra in the field of mechanical processing), the present application constructs a multi-dimensional feature system containing three-dimensional topographic parameters (fractal dimension D and root mean square deviation σ) and physical property parameters (water content and density), breaking through the limitation of single parameter representing surface integrity.

[0049] For example, the gear surface roughness model in mechanical engineering usually only contains Ra and processing parameters, while the present application can also depict the self-similarity of microcracks on the rock surface through the fractal dimension D, and correlate the cementation strength of mineral particles through the density parameter, forming a feature space suitable for the multi-scale failure mechanism of rocks.

[0050] Therefore, the present application constructs a multi-parameter evaluation system to replace the single roughness parameter to accurately represent the three-dimensional topography of the rock sample.

[0051] Step S103, an association model is constructed by integrating the random forest model and the BP neural network, the multi-dimensional data set is used as training data to train the association model, and the model hyperparameters are optimized through cross-validation to dynamically adjust the feature weights for lithology.

[0052] The association model uses the multi-dimensional input feature vector as input to predict the mechanical index corresponding to the input features.

[0053] Specifically, the present application constructs a machine learning model for predicting mechanical indexes, integrates the random forest algorithm (RF) and the BP neural network (BP-NN) for modeling, and the multi-dimensional input feature vector determined in the above step is used as the input feature of the model, that is, the input features of the model include lithology, water content, density and flatness parameters. And the hyperparameters are optimized through cross-validation, for example, the random forest tree depth is set to 20-50, the BP neural network hidden layer node number is set to 10-30, and the feature weights are dynamically adjusted through feature cross-validation.

[0054] In an embodiment of the present application, the constructed association model is specifically used to identify the interaction effect between different parameters in the multi-dimensional input feature vector through the multivariate adaptive regression spline MARS algorithm.

[0055] Specifically, since the application integrates the water content, density and flatness parameters into a unified feature space, the modeling is performed according to the coupling of multiple physical parameters, and the Multivariate Adaptive Regression Splines (MARS) algorithm is used to identify the interaction effects between parameters, such as the synergistic attenuation effect of the product of water content and Ra on strength.

[0056] Therefore, the application can perform multi-factor coupling analysis to improve the prediction accuracy.

[0057] In an embodiment of the application, the feature weight is dynamically adjusted for lithology, including: setting the weight adjustment factor of water content and density for lithology, wherein the weight of water content is increased when the rock sample is soft rock, and the weight of density is increased when the rock sample is hard rock; and setting the weight threshold of water content and density in the random forest model.

[0058] Specifically, when the hyperparameters are optimized by cross-validation in this embodiment, the weight adjustment factor of water content and density is increased. As a possible implementation, a lithology-sensitive feature importance dynamic adjustment mechanism is introduced in the random forest algorithm, for example, the weight of water content in soft rock is increased to 30%, and the weight of density in hard rock is increased to 25%. Moreover, this embodiment sets the feature importance threshold of water content in RF to be greater than or equal to 0.15, and the density to be greater than or equal to 0.1, which specifies the basic range for the dynamic adjustment of the weight.

[0059] Therefore, unlike the traditional model which treats all features with equal weight, this embodiment is more in line with the objective law of lithology dominant parameter sensitivity, and can improve the prediction accuracy by forcing the model to learn the coupling effect of water content and density with flatness.

[0060] Based on the above model construction method, the application generates a training data set required for model training using the multi-dimensional data set obtained in step S101, and trains the above correlation model.

[0061] As an example, a training data set containing 100,000 experimental data is constructed by expanding the data, wherein the stratified sampling is performed according to water content (low, medium, high) and density (light, medium, heavy), the proportion of soft rock (such as shale) samples is 30% (of which different water content samples each account for 10%), the proportion of hard rock (such as granite) is 20% (of which different density samples each account for 5%-8%), and the number of cross-validation is 10. Moreover, the model evaluation indicators are set to include R 2 , root mean square error (RMSE) and mean absolute error (MAE), for example, the model prediction accuracy R 2≥0.92, and the root mean square error RMSE is less than or equal to 3%. The specific model training process can refer to the training manner in the related art, and the present application does not limit this.

[0062] It should be noted that the present application converts the coupling problem of lithology, flatness and environmental parameters in rock mechanics into a nonlinear mapping of a high-dimensional feature space. Compared with the concrete strength prediction model in civil engineering (such as usually using linear regression or a single neural network), the present application realizes, on the one hand, feature cross-validation, and sets the weight adjustment factor of water content and density. On the other hand, by referring to the multi-scale feature extraction idea in the image processing field, cross-domain model fusion is realized, that is, the nonlinear influence of the flatness parameter is captured through the tree structure of the random forest, and the complex coupling relationship between the water content, density and strength is fitted by combining the BP neural network, forming a cross-domain correlation model of "geometric features-physical properties-mechanical responses".

[0063] Step S104, a differential correction model is constructed, the output of the correlation model is corrected through the differential correction model for different lithologies, and based on the differential correction model, a mechanical index prediction curve corresponding to the multi-dimensional input feature vector of the rock sample to be tested is obtained.

[0064] Specifically, since there are various types of rock samples in actual application, and the lithology difference will cause model prediction deviation, therefore, the present application further constructs a differential correction model, which corrects the output of the above correlation model by generating a multi-element correction formula of the mechanical index. Therefore, the prediction model composed of the correlation model and the differential correction model of the present application can more accurately predict the correlation between the end face flatness of the rock sample and the mechanical index.

[0065] Among them, the differential correction model of the present application can correspondingly establish a multi-element correction formula containing water content and density for soft rock (such as shale) and hard rock (such as granite).

[0066] In order to more clearly illustrate the specific implementation manner of the differential correction of the present application for lithology, a correction method proposed in an embodiment of the present application is exemplarily described below. Figure 2 The flow chart of a differential correction method proposed in an embodiment of the present application is shown in Figure 2 The method includes the following steps:

[0067] Step S201, the multi-dimensional input feature vector is data layered, and the layered data is standardized.

[0068] Specifically, the data is layered according to the lithology (soft rock or hard rock), water content (0-20%, 5% as a grade), density (0.1 g / cm 3The data subsets are divided by intervals to ensure that the sample size in each subset is greater than or equal to 500.

[0069] Then, the flatness parameters (Ra and Rz), moisture content (w), and density (p) are subjected to standardization processing to eliminate the dimensional influence.

[0070] In step S202, for soft rock, a first correction formula is constructed with uniaxial compressive strength as the dependent variable and the roughness parameters, moisture content, and density after standardization processing as the independent variables, wherein the first correction formula contains multiple lithology-related coefficients to be solved.

[0071] Specifically, for soft rock, the first correction formula is constructed with the measured uniaxial compressive strength (UCS) as the dependent variable and Ra, moisture content (w), and density (p) as the independent variables, as shown below:

[0072] UCS = a x Ra + b x w + g x p + d

[0073] where a, b, g, and d are lithology-related coefficients.

[0074] In step S203, the first correction formula is subjected to ridge regression fitting to construct a loss function containing an L2 regularization term.

[0075] Specifically, each lithology-related coefficient is solved by ridge regression fitting. Ridge regression fitting is a regularization improvement of the least squares method. When the first correction formula is subjected to ridge regression fitting, a loss function with an L2 regularization term is constructed as shown in the following formula:

[0076]

[0077] where y i is the measured strength of the i-th sample, and l is the regularization parameter.

[0078] In step S204, the regularization parameter in the loss function is optimized by 5-fold cross-validation to obtain the fitted multiple lithology-related coefficients.

[0079] Specifically, 5-fold cross-validation (5-Fold Cross-Validation) is a model evaluation method that can be used to evaluate the performance and generalization ability of machine learning models. In this embodiment, the regularization parameter l in the above loss function is optimized (search range 0.01-10) by 5-fold cross-validation to suppress the possible multicollinearity of moisture content and density (such as the positive correlation between moisture content and density in clay-rich soft rock), so that the validation set R 2 ≥ 0.92.

[0080] Therefore, the embodiment introduces L2 regularization to suppress overfitting by fitting the measured data through ridge regression, and then optimizes the regularization parameter through 5-fold cross-validation, so as to determine the plurality of correlation coefficients in the step S202. The embodiment can also verify the accuracy of the solved correlation coefficients through sensitivity analysis. For example, if the solved β < 0, it is consistent with the rule that the higher the water content, the lower the strength.

[0081] Based on the above embodiment, the application can also perform differential correction on hard rock. Figure 3 The flowchart of another differential correction method proposed by the embodiment of the application is shown in FIG. 6, which comprises the following steps after the step S201 in the above embodiment is executed: Figure 3

[0082] Step S301, for hard rock, screening a plurality of target features sensitive to hard rock from the standardized data, and constructing a second correction formula with uniaxial compressive strength as the dependent variable and the plurality of target features as the independent variables, wherein the second correction formula contains a plurality of to-be-solved lithology correlation coefficients.

[0083] Specifically, for hard rock (such as granite), significant variables (such as Ra and p) are screened through stepwise regression method, and parameters such as water content which are not sensitive to hard rock are removed. Then, the least square method is used to fit the second correction formula shown in the following formula: UCS = a x Ra + g x p + e.

[0084] Step S302, performing least square fitting on the second correction formula, and constructing a three-dimensional particle flow model PFC, and performing discrete element simulation through the three-dimensional particle flow model PFC to calibrate the lithology correlation coefficients corresponding to the density.

[0085] Specifically, the coefficient g corresponding to the density p in the above second correction formula can be calibrated through discrete element simulation, specifically, the particle flow model (Particle Flow Code, PFC) can be used, and the residual standard deviation is required to be less than 3% of the measured value. The particle flow model is a numerical simulation tool based on the discrete element method (DEM), which predicts the macroscopic behavior of materials by simulating the interaction between particles.

[0086] It should be noted that, because the hard rock mineral particles are significantly cemented, the strength is mainly controlled by the contact stiffness and cementing strength between particles, so the discrete element simulation can intuitively depict the micro-crack propagation process through the particle flow model (PFC). The soft rock (such as shale) has complex water-softening characteristics of clay minerals, and the discrete element is difficult to accurately depict the water-rock interaction of nanoscale clay particles, so it needs to rely on measured data.

[0087] ​Thus, the embodiment can determine the plurality of correlation coefficients in step S301 by fitting the measured data and the discrete element virtual data by the least square method, and calibrating the coefficient γ by the discrete element simulation. The embodiment can also verify the accuracy of the solved correlation coefficients by residual analysis, for example, judging whether the residual standard deviation of the simulation result is less than or equal to 3% of the measured value.

[0088] In step S303, the data in the training data set is expanded based on the discrete element simulation, and the correlation model is corrected by the expanded training data set.

[0089] Specifically, when the hard rock is predicted and corrected, the data in the training data set can also be expanded by the discrete element simulation, so that the parameters in the above correlation model are further optimized by the expanded training data set, so as to improve the prediction accuracy for hard rock.

[0090] As a possible implementation, the process of expanding the training data set based on the discrete element simulation includes the following steps:

[0091] First, model construction. Based on the mineral composition of hard rock (such as quartz, feldspar), the particle assembly model is established, the particle size distribution (1-5mm), the contact model (linear parallel cementation model) and the cementation parameters (normal or tangential stiffness and cementation strength) are set.

[0092] Second, parameter calibration. By simulating the uniaxial compression experiment, the cementation parameters are adjusted to make the simulation strength error ≤5% of the measured value, and the quantitative relationship between density (ρ) and cementation strength (such as the cementation strength is increased by 3-6MPa for every 0.1g / cm 3

[0093] Third, data generation. Within the standard flatness range (Ra≤50μm), 20,000 virtual samples (including the corresponding relationship of Ra, ρ and UCS) are generated by changing the particle contact surface roughness to simulate the Ra parameter, and the original training set is mixed in proportion of 1:1, so as to improve the generalization ability of the model to multi-scale damage of hard rock.

[0094] It should be noted that the influence of the lithology, water content and density cross-scale fusion can be corrected by hierarchical sampling and data enhancement to correct the model deviation caused by lithology difference. For soft rock, 20% more experimental samples can be added according to the water content, such as shale samples with water content of 10%, 15% and 20% groups, each adding 500 data. For hard rock, data is supplemented by discrete element simulation.

[0095] ​It should be noted that in the sensitivity analysis in the above embodiment, the feature importance ranking of the random forest (for example, w weight of 30% in soft rock, Ra weight of 25%) can be used to verify the consistency of the coefficient sign and the physical meaning (for example, β < 0 indicates that the higher the water content, the lower the strength). When solving the correlation coefficient, parameter optimization can be performed, for example, by gradient descent method or matrix inversion (ridge trace method) to solve the coefficient that minimizes the loss function. Wherein β < 0 indicates that the strength attenuation is 0.5-1.2 MPa per 1% increase in water content, and γ > 0 indicates that the strength attenuation is 0.5-1.2 MPa per 0.1 g / cm 3 The strength is improved by 2-5 MPa; when Ra > 10 μm and the water content is > 15%, the soft rock strength attenuation rate is improved to 20%-25%; when the density is > 2.6 g / cm 3 The sensitivity of the hard rock to flatness is reduced by 30%.

[0096] Based on the above embodiment, for the prediction model composed of the correlation model and the differential correction model constructed by the present application, model verification can also be performed, for example, 5-fold cross-validation is used to evaluate the fitting accuracy, and the determination coefficient R 2 of the predicted value and the measured value is required to be greater than or equal to 0.92, and the root mean square error (RMSE) is required to be less than or equal to 5%. It can be understood that compared with the ordinary least squares method (OLS) without regularization, the ridge regression in the present application can effectively solve the problem of multicollinearity of rock physical parameters (such as water content and density) that may exist, and improve the model generalization ability. For example, when the water content and the density of the soft rock are positively correlated, the ridge regression can avoid the problem of too large coefficient estimation variance caused by strong correlation between parameters in OLS.

[0097] Therefore, the prediction model established by the present application can accurately predict the mechanical indicators corresponding to different multi-dimensional input feature vectors.

[0098] Further, after the training and adjustment of the prediction model are completed, in actual application, the multi-dimensional input feature vector of the rock sample to be tested can be input into the prediction model to obtain the predicted values of the multiple mechanical indicators output by the prediction model, and then the corresponding mechanical indicator prediction curve can be generated according to the predicted values.

[0099] In order to more clearly illustrate the specific implementation manner of generating and displaying the mechanical indicator prediction curve by the present application, the following will exemplarily illustrate a method proposed in an embodiment of the present application. Figure 4 The flow chart of a method for generating and displaying a mechanical indicator prediction curve proposed in an embodiment of the present application is shown in Figure 4 The method comprises the following steps:

[0100] In step S401, the multi-dimensional input feature vector of the rock sample to be tested is received through the visual query platform.

[0101] Specifically, the embodiment pre-constructs a visual query platform, in which a Web-side interactive interface is developed to support a user to input the flatness parameter, lithology and water content of a rock sample to be tested through the interactive interface. The platform can output the strength range and prediction curve in real time according to the input parameters.

[0102] In the present application, the prediction curve is a quantitative relationship curve, in which the horizontal axis represents the flatness parameter (Ra and Rz, etc.), and the vertical axis represents the mechanical index (such as uniaxial compressive strength UCS, elastic modulus E). Under the conditions of fixed lithology, water content and density, the quantitative influence trend of the flatness parameter on the mechanical index can be reflected.

[0103] Step S402, calling the trained prediction model to output the mechanical index prediction value corresponding to the received multi-dimensional input feature vector.

[0104] The prediction model is the model trained and adjusted as described above, including the correlation model and the differential correction model.

[0105] Specifically, when the visual query platform operates on the input data, data driving is performed, that is, the platform calls the trained prediction model to extract the feature weight corresponding to the input parameter and the regression equation. For example, for a certain sandstone, the following regression equation UCS = -3.2 x Ra - 1.5 x w + 4 x p + 50 is obtained. Further, the mechanical index prediction value corresponding to the sandstone can be output.

[0106] Step S403, within the preset flatness range, using the mechanical index prediction value to generate equidistant data points according to unit step length, and fitting the mechanical index prediction curve through difference.

[0107] Specifically, when generating the curve, the lithology, water content and density parameters are fixed, and within the flatness range allowed by the national standard (Ra≤50μm), equidistant data points are generated with a step length of 5μm, and a continuous curve is obtained through cubic spline interpolation fitting.

[0108] Step S404, displaying the mechanical index prediction curve through the interactive interface of the visual query platform.

[0109] Specifically, after obtaining the mechanical index prediction curve corresponding to the multi-dimensional input feature vector of the rock sample to be tested, front-end display can also be performed. As an example, ECharts or other visualization libraries are used to dynamically draw a "flatness-strength" two-dimensional curve, which is displayed on the interactive interface of the visual query platform. The horizontal axis of the curve is Ra or Rz, and the vertical axis is the mechanical index prediction value, and the measured data scatter points (error ±5%) and the model determination coefficient R 2 .

[0110] Further, after obtaining the mechanical index prediction curve corresponding to the multi-dimensional input feature vector of the rock sample to be tested, the embodiment further includes: generating a strength correction coefficient table within a predetermined flatness range. Specifically, the embodiment can also construct a standard compatibility module according to a large number of prediction results, establish a strength correction formula within the flatness range allowed by the national sample preparation standard (for example, Ra≤50μm), form a standardized prediction system, and automatically generate a strength correction coefficient table that meets the standard, thereby supporting one-key generation of an engineering report.

[0111] Based on the above embodiment, the application can also establish a rock mechanics index engineering database system. The database architecture can use a relational database (MySQL), which can store experimental data (≥100,000), model parameters (≥200 groups), and engineering cases (≥50). The design fields include sample ID, lithology, flatness parameter, mechanical index, and experimental conditions.

[0112] As an example, the database system includes the following modules:

[0113] A data acquisition module for storing experimental data, numerical simulation results, and engineering cases.

[0114] A model prediction module that integrates the prediction model described in the above embodiment, supports inputting lithology, water content, density, and flatness parameters, and supports outputting strength range using a multivariate correction formula.

[0115] A visual interactive module that provides flatness-strength relationship curve display, data export, and API interface functions.

[0116] The model prediction module of the present example has built-in lithology differentiation correction formula, the database covers the flatness range allowed by the national sample preparation standard (Ra≤50μm), and has built-in lithology differentiation correction model. When applied to engineering rock mass stability evaluation, by inputting the end face flatness parameter of the rock sample, the database system can output a strength range that meets the national standard, guiding engineering design.

[0117] In summary, the method for determining the correlation between the flatness of the end face of the rock sample and the mechanical index of the embodiments of the present application constructs a multi-dimensional feature vector combining three-dimensional morphology and physical properties in rock mechanics testing, improves the feature system, and predicts the correlation by combining machine learning with experimental data. Compared with the traditional statistical regression model, the method has higher prediction accuracy, realizes the leap from qualitative control of flatness parameters to quantitative prediction of mechanical indexes, and improves the adaptability of machine learning models in geotechnical engineering by introducing dynamic weights, cross-domain feature fusion, and explainability analysis in the prediction model. In addition, the method introduces a dynamic adjustment mechanism for feature importance sensitivity of lithology in the random forest algorithm, which is more in line with the objective law of the sensitivity of lithology dominant parameters in rock mechanics than the general machine learning framework. Furthermore, the method also constructs an engineering database that can be directly applied to rock mechanics testing laboratories and engineering rock mass stability evaluation scenarios, which can reduce repeated experiments and testing costs caused by flatness errors. Thus, the method establishes a quantitative relationship between the flatness of the end face of the rock and the mechanical index, which can effectively avoid the influence of the end face flatness on rock mechanics testing, improve the accuracy of quantitative relationship prediction, and is conducive to improving engineering safety and economy.

[0118] Based on the above embodiments, in order to more clearly and intuitively describe the method for determining the correlation between the flatness of the end face of the rock sample and the mechanical index of the present application, the specific process in actual application and the achieved effect, the following examples are used to illustrate the method.

[0119] The first embodiment is to predict the uniaxial compressive strength of samples with different lithology. The embodiment includes the following steps:

[0120] First, prepare the sample. Select quartz sandstone (density 2.3 g / cm 3 ), prepare three groups of samples with water content of 5%, 10%, and 15%, and prepare 10 samples with Ra=10 μm, 5 μm, and 2 μm for each group by 80 mesh, 200 mesh, and 500 mesh grinding wheels.

[0121] Second, collect data. The UCS of the sample with water content of 5% and Ra=10 μm is 52 MPa, and the UCS of the sample with water content of 15% and Ra=10 μm is 45 MPa (the strength decreases by about 7-8 MPa for every 5% increase in water content); the UCS of the granite sample (Ra=5 μm) with density of 2.5 g / cm 3 is 125 MPa, and the UCS of the sample with the same lithology and density of 2.7 g / cm 3 is 132 MPa (the strength increases by about 7 MPa for every 0.2 g / cm 3 increase in density).

[0122] Third step, model training. Input data into the random forest model, and train the sandstone correction formula of water content (w) and density (p) as follows: UCS = -3.2 x Ra-1.5 x w+4 x p+50 (R 2 = 0.97), and the granite correction formula as follows: UCS = -2 x Ra+3 x p+55 (R 2 = 0.95).

[0123] Fourth step, engineering application. In a certain slope engineering, the sandstone sample is collected, Ra = 3 pm is measured, the UCS prediction value is 61.6 MPa through database query, and the field test value is 60 MPa, so it can be determined that the error is ≤2.7%.

[0124] Second embodiment, operation of the constructed database platform. This embodiment includes the following steps:

[0125] First step, obtain user input data: select shale, water content 15%, and Ra = 15 pm on the Web interface.

[0126] Second step, system processing. Call the differentiated model, output the uniaxial compressive strength range 25-30 MPa, and generate the strength-flatness relationship curve and experimental data traceability report.

[0127] Third step, verify the results, compare the experimental data of 3 groups of shale samples, and determine that the prediction value and the measured value are consistent ≥92%.

[0128] In order to realize the above embodiment, the application further provides a system for determining the correlation between the end surface flatness of a rock sample and the mechanical index, Figure 5 The structure diagram of the system for determining the correlation between the end surface flatness of a rock sample and the mechanical index according to the embodiment of the application is shown in Figure 5 As shown in the figure, the system includes a collection module 100, a construction module 200, a prediction module 300, and a correction module 400.

[0129] Among them, the collection module 100 is used to prepare multiple rock samples with different end surface flatness by using the flatness preparation process, collect the multi-dimensional flatness parameters of the multiple rock samples, and measure the multiple mechanical indexes, water content and density of the multiple rock samples, to generate a multi-dimensional data set.

[0130] The construction module 200 is used to construct a multi-dimensional input feature vector based on the flatness parameters, lithology, water content and density.

[0131] The prediction module 300 is configured to construct a correlation model by integrating a random forest model and a BP neural network, train the correlation model by taking a multidimensional data set as training data, and optimize model hyperparameters by cross-validation, so as to dynamically adjust feature weights for lithology, wherein the correlation model takes a multidimensional input feature vector as input, and is configured to predict a mechanical index corresponding to the input feature.

[0132] The correction module 400 is configured to construct a differentiated correction model, correct the output of the correlation model by the differentiated correction model for different lithologies, and obtain a mechanical index prediction curve corresponding to the multidimensional input feature vector of the rock sample to be tested based on the differentiated correction model.

[0133] It should be noted that the above description of the embodiment of the method for determining the correlation between the end face flatness of the rock sample and the mechanical index is also applicable to the system of the embodiment, which will not be described here.

[0134] In summary, the system for determining the correlation between the end face flatness of the rock sample and the mechanical index according to the embodiments of the present application establishes a quantitative relationship between the end face flatness of the rock and the mechanical index, can effectively avoid the influence of the end face flatness on the rock mechanical test, and improves the accuracy of the quantitative relationship prediction.

[0135] In order to realize the above-mentioned embodiments, the present application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method for determining the correlation between the end face flatness of the rock sample and the mechanical index according to any one of the above-mentioned embodiments of the first aspect.

[0136] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0137] Moreover, the terms "first", "second", "third", etc. are used herein only to describe different steps or categories of steps in a claim for patent purposes, and are not to be construed as indicating or implying relative importance of one step to another or a quantity of steps. Thus, features defined with "first", "second" or "third" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "plurality" is at least two, for example, two, three, etc., unless otherwise explicitly and specifically limited.

[0138] Any process or method descriptions or blocks in flow charts herein and elsewhere can be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps in the process, and alternate implementations are possible. In some embodiments, the processes or methods described in flow charts herein and elsewhere can be tailored by reordering steps and / or adding or omitting one or more of the described steps, and the order of the steps can or can not be specifically mentioned or critical. One of ordinary skill in the art will recognize that the steps in the processes or methods described herein and elsewhere can be implemented by processor-executable code stored on a computer-readable medium, which can be incorporated in software, applied to the process or method, or otherwise used to implement the process or method.

[0139] Logic and / or steps represented in flow charts herein and elsewhere, for example, can be embodied in computer-readable instructions, statements, or in the form of one or more modules, segments, or portions of code that implement specified logical functions. Such computer-readable instructions can be loaded onto a computer, in one or more ways, and executed thereby. Such computer-readable instructions can include, for example, instructions in a computer-readable form, computer-executable instructions, or instructions that modify the operation of or management of the computer or components thereof, or a combination thereof. For purposes of this application, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. Computer readable medium can typically be a computer- readable storage medium. The computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a propagation medium. The computer-readable storage medium can also include, without limitation, an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CD-ROM). The computer-readable medium can also include, without limitation, a paper or other suitable medium upon which the program is printed, as the program can be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0140] It should be understood that parts of the present application can be realized in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be realized as software or firmware stored in a memory and executed by a suitable instruction execution system. As such, if realized in hardware, and in another embodiment, any one or a combination of the following technologies known in the art can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0141] Those skilled in the art can understand that all or part of the steps carried out by the above-mentioned embodiment methods can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium, and when the programs are executed, one or a combination of the steps of the method embodiments is included.

[0142] In addition, each functional unit in each embodiment of the present application can be integrated in one processing module, or each unit can be physically present separately, or two or more units can be integrated in one module. The above-mentioned integrated module can be realized in the form of hardware or in the form of a software functional module. The integrated module, if realized in the form of a software functional module and sold or used as an independent product, can also be stored in a computer readable storage medium.

[0143] The above-mentioned storage medium can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for determining the correlation between the flatness of the end face of a rock sample and its mechanical properties, characterized in that, Includes the following steps: Multiple rock samples with different end-face flatness were prepared using a flatness preparation process. Multidimensional flatness parameters of the multiple rock samples were collected, and various mechanical properties, water content and density of the multiple rock samples were measured to generate a multidimensional dataset. Based on the smoothness parameters, lithology, water content, and density, a multidimensional input feature vector is constructed; An association model is constructed by integrating a random forest model and a backpropagation neural network. The multidimensional dataset is used as training data to train the association model. The model hyperparameters are optimized by cross-validation to dynamically adjust the feature weights for lithology. The association model uses the multidimensional input feature vector as input to predict the mechanical indices corresponding to the input features. A differential correction model is constructed to correct the output of the correlation model for different lithologies. Based on the differential correction model, the mechanical index prediction curve corresponding to the multidimensional input feature vector of the rock sample to be tested is obtained. The output of the correlation model is modified using the differentiated correction model for different lithologies, including: For soft rock, the multidimensional input feature vector is stratified, and the stratified data is standardized. A first correction formula is constructed with uniaxial compressive strength as the dependent variable and standardized roughness parameter, water content, and density as independent variables. Ridge regression is applied to the first correction formula to construct a loss function containing an L2 regularization term. The regularization parameter in the loss function is optimized through 5-fold cross-validation to obtain multiple lithological correlation coefficients. The first correction formula contains multiple lithological correlation coefficients to be solved. For hard rock, the multidimensional input feature vector is stratified, and the stratified data is standardized. Multiple target features sensitive to hard rock are selected from the standardized data, and a second correction formula is constructed with uniaxial compressive strength as the dependent variable and the multiple target features as independent variables. The second correction formula is fitted using the least squares method, and a three-dimensional particle flow model (PFC) is constructed. Discrete element simulation is performed using the PFC to calibrate the lithological correlation coefficients corresponding to density. The data in the training dataset is expanded based on the discrete element simulation, and the correlation model is corrected using the expanded training dataset. The second correction formula contains multiple lithological correlation coefficients to be solved.

2. The method according to claim 1, characterized in that, The association model is specifically used for: The interaction effects between different parameters in the multidimensional input feature vector are identified using the Multivariate Adaptive Regression Spline (MARS) algorithm.

3. The method according to claim 1, characterized in that, The dynamic adjustment of feature weights for lithology includes: Weighting factors for water content and density were set for lithology. Specifically, the weight of water content was increased when the rock sample was soft rock, and the weight of density was increased when the rock sample was hard rock. In the random forest model, weight thresholds for water content and density are set.

4. The method according to claim 1, characterized in that, The process of obtaining the mechanical index prediction curve corresponding to the multidimensional input feature vector of the rock sample to be tested includes: The multidimensional input feature vector of the rock sample to be tested is received through a visualization query platform; The trained prediction model is invoked to output the predicted mechanical index value corresponding to the currently received multi-dimensional input feature vector, wherein the prediction model includes the correlation model and the differential correction model; Within a preset flatness range, the predicted mechanical index values ​​are used to generate equally spaced data points in unit step sizes, and the predicted mechanical index curve is obtained by difference fitting.

5. The method according to claim 4, characterized in that, After obtaining the mechanical index prediction curve corresponding to the multidimensional input feature vector of the rock sample to be tested, the method further includes: The mechanical index prediction curve is displayed through the interactive interface of the visualization query platform; Within the preset flatness range, a strength correction coefficient table is generated.

6. The method according to claim 1, characterized in that, The preparation of multiple rock samples with different end-face flatness using a flatness preparation process includes: By combining high-precision grinding and polishing equipment with pressure sensors, the flatness of the end face of different rock samples can be adjusted.

7. A system for determining the correlation between the flatness of the end face of a rock sample and its mechanical properties, characterized in that, Includes the following modules: The acquisition module is used to prepare multiple rock samples with different end face flatness using a flatness preparation process, acquire multidimensional flatness parameters of the multiple rock samples, and measure various mechanical properties, water content and density of the multiple rock samples to generate a multidimensional dataset. A construction module is used to construct a multidimensional input feature vector based on the smoothness parameters, lithology, water content, and density; The prediction module is used to construct an association model by integrating a random forest model and a backpropagation neural network, train the association model using the multidimensional dataset as training data, and optimize the model hyperparameters through cross-validation to dynamically adjust the feature weights for lithology. The association model uses the multidimensional input feature vector as input to predict the mechanical indicators corresponding to the input features. The correction module is used to construct a differential correction model, which corrects the output of the correlation model for different lithologies, and obtains the mechanical index prediction curve corresponding to the multidimensional input feature vector of the rock sample to be tested based on the differential correction model. The output of the correlation model is modified using the differentiated correction model for different lithologies, including: For soft rock, the multidimensional input feature vector is stratified, and the stratified data is standardized. A first correction formula is constructed with uniaxial compressive strength as the dependent variable and standardized roughness parameter, water content, and density as independent variables. Ridge regression is applied to the first correction formula to construct a loss function containing an L2 regularization term. The regularization parameter in the loss function is optimized through 5-fold cross-validation to obtain multiple lithological correlation coefficients. The first correction formula contains multiple lithological correlation coefficients to be solved. For hard rock, the multidimensional input feature vector is stratified, and the stratified data is standardized. Multiple target features sensitive to hard rock are selected from the standardized data, and a second correction formula is constructed with uniaxial compressive strength as the dependent variable and the multiple target features as independent variables. The second correction formula is fitted using the least squares method, and a three-dimensional particle flow model (PFC) is constructed. Discrete element simulation is performed using the PFC to calibrate the lithological correlation coefficients corresponding to density. The data in the training dataset is expanded based on the discrete element simulation, and the correlation model is corrected using the expanded training dataset. The second correction formula contains multiple lithological correlation coefficients to be solved.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a method for determining the correlation between the flatness of the rock sample end face and mechanical properties as described in any one of claims 1-6.

Images