Thermal shock cooling rock rotary cutting damage prediction method

Abstract

The application discloses a hot shock cooling rock rotary cutting damage prediction method, solves the technical problem that the prior art cannot realize high-precision and high-generalization prediction of hot shock cooling rock rotary cutting damage parameters while ensuring physical mechanism consistency. The method first acquires hot shock cooling rock rotary cutting sample data under different initial temperature conditions, constructs a hot shock cooling rock rotary cutting failure criterion RCFC considering thermal damage evolution and temperature-dependent fracture mechanics characteristics, and then embeds the RCFC as a hard physical constraint into a physical information neural network model, and outputs the prediction results of key damage parameters such as rotary cutting energy and rotary speed after training. The application realizes the deep fusion of data driving and rock cutting damage physical mechanism, significantly improves the prediction accuracy and physical consistency, and still has good robustness under the condition of sparse data, and can provide technical support for drilling parameter design and construction safety control of deep high-temperature rock mass engineering.

Description

Technical Field

[0001] This invention belongs to the field of rock mechanics and underground engineering technology, specifically relating to a method for predicting the failure of rotating cutting rocks due to thermal shock cooling. Background Technology

[0002] With the rapid development of deep underground engineering (depth > 1000m), geothermal resource development of hot dry rock, construction of deep underground oil and gas storage facilities, and geological disposal of high-level radioactive waste, deep rock masses are generally located in high-temperature environments ranging from 25℃ to 600℃ or even higher. This significantly increases the complexity and technical difficulty of drilling and rock rotary cutting and crushing operations. Under high-temperature conditions, the mismatch of thermal expansion coefficients of different mineral components within the rock generates non-uniform thermal stress, inducing microstructural deterioration and the initiation, propagation, and penetration of primary / secondary microcracks. This leads to non-linear degradation of key mechanical parameters such as uniaxial compressive strength, elastic modulus, and fracture toughness, resulting in increased drilling specific energy, accelerated drill bit wear, and a significant reduction in rock breaking efficiency. This has become a core bottleneck restricting the efficient construction of deep high-temperature rock mass engineering projects.

[0003] Thermal shock cooling-assisted rock breaking technology involves the instantaneous injection of a low-temperature cooling medium into a high-temperature rock mass. This induces strong thermo-mechanical coupling damage through a dramatic temperature gradient, accelerating the propagation and penetration of internal microcracks. This effectively improves rock drillability and reduces rock breaking energy consumption, making it one of the mainstream technologies for efficient high-temperature rock mass fracturing. However, thermal shock cooling significantly alters the rock's failure mode and cutting response, posing a significant challenge to the accurate prediction of rotary cutting parameters.

[0004] Currently, predictions and analyses of rock rotational cutting processes mostly employ empirical statistical models, linear cutting response models, or simplified mechanical models based on the ideal elastoplastic assumption. These models are largely based on cutting test data of rock masses at room temperature and do not fully consider the multi-physics effects during thermal shock cooling, such as the evolution of thermal damage, temperature-related fracture toughness degradation, and crack-stress field coupling. They cannot accurately characterize the evolution of the rock rotational cutting response from near-linear to strongly nonlinear under thermal shock cooling conditions, nor can they quantitatively represent the failure boundaries and damage thresholds of rock rotational cutting under different degrees of thermal damage. This results in a significant decrease in prediction accuracy and engineering applicability under high-temperature thermal shock cooling conditions.

[0005] In recent years, pure data-driven machine learning methods, represented by neural networks, have been gradually applied to the fields of rock mechanics parameter inversion and drilling response prediction. Their powerful nonlinear fitting capabilities can improve prediction results under complex working conditions to a certain extent. However, these methods rely entirely on the statistical regularities of experimental data and lack hard constraints on the physical mechanisms of rock cutting failure. In engineering scenarios where experimental data on thermal shock cooling rock cutting is scarce and working conditions are not fully covered, they are prone to producing prediction results that violate the basic laws of thermodynamics and fracture mechanics. Furthermore, the models have inherent defects such as poor interpretability, insufficient generalization ability, and low robustness, and cannot meet the needs of accurate prediction and safe control of rotary cutting parameters in high-temperature rock engineering.

[0006] In summary, existing technologies cannot simultaneously meet the multiple requirements of consistency of physical mechanisms, prediction accuracy, and generalization ability in predicting the failure of rotating rock under thermal shock cooling. There is an urgent need to develop a prediction method that integrates the failure mechanism of fracture mechanics and the advantages of data-driven approaches to achieve accurate and reliable prediction of rock rotating cutting parameters under high-temperature thermal shock cooling conditions, and to provide technical support for efficient drilling and safe construction in deep high-temperature rock engineering. Summary of the Invention

[0007] The purpose of this invention is to provide a method for predicting the failure of rotating rocks subjected to thermal shock, which solves the technical problem that existing technologies cannot achieve high-precision and high-generalization prediction of failure parameters of rotating rocks subjected to thermal shock while ensuring the consistency of physical mechanisms.

[0008] The technical solution adopted in this invention is a method for predicting the failure of rock rotational cutting caused by thermal shock cooling, comprising the following steps: S1: Obtain rotary cutting sample data of thermally shock cooled rocks under different initial temperature conditions, and establish a rotary cutting sample dataset of thermally shock cooled rocks; S2: Based on the fracture mechanics characteristics, thermal damage parameters and temperature-related fracture toughness parameters of thermally shock cooled rocks, construct the failure criteria for rotary cutting of thermally shock cooled rocks; S3: Construct a physical information neural network model, embed the rotary cutting failure criterion as a physical constraint term into the physical information neural network model, and establish a rotary cutting parameter prediction model; S4: The prediction model is trained based on the dataset obtained in S1, and the predicted parameters of the rotating cutting of the rock by thermal shock are predicted using the trained prediction model.

[0009] The invention is further characterized by: First, the sample data obtained from S1 is normalized. Then, a total loss function is constructed, which is a weighted sum of the data loss term and the physical loss term.

[0010] in, These are the weighting coefficients of the loss function; Data loss items From the mean square error term and regularization term constitute:

[0011]

[0012]

[0013] in, For rotary cutting energy prediction data, For the first A set of experimental data on rotary cutting energy. For the first The weights of the trainable parameters. This represents the expectation operator and the physical loss term. L RCFC for:

[0014] in, For rotary cutting energy prediction data, Data for calculating the energy physics model of rotary cutting. This represents the expectation operator.

[0015] The weight coefficient of the loss function is set to 0.46.

[0016] S1 specifically refers to: The sample is heated to a preset initial temperature. After reaching the preset temperature, the sample is subjected to thermal shock cooling using a cooling medium to obtain different initial temperatures. T Rock samples were subjected to thermal shock cooling under certain conditions, followed by a rotary cutting test to collect drill pressure data during the drilling process. F w Cutting force F t and rotational speed ; Cutting force collected from rotary cutting test and drilling pressure Calculate rotary cutting energy and drilling strength :

[0017]

[0018] in, For cutting force, For drilling pressure, The rotating cutting area is the area of ​​rotation. A The expression is as follows:

[0019] in, Where is the drill bit radius. The drill bit tip angle, For the affected length; Based on the different initial temperatures mentioned above and drilling strength obtained under different initial temperature conditions Rotary cutting energy and rotational speed Establish a sample dataset of rock rotation cutting samples cooled by thermal shock.

[0020] The preset initial temperature is 25℃, 200℃, 400℃ or 600℃; the cooling medium is low temperature water; the rotary cutting test is carried out in a digital drilling test system, and confining pressure is applied during the test to simulate the in-situ stress environment.

[0021] The failure criterion for rotary cutting in S2 is:

[0022] in, For rotary cutting energy, For drilling strength, For the breakage angle, The initial damage variable is temperature-related. and b are coefficients related to thermal damage. The damage coefficient is based on the crack. is the coefficient of friction of intact rock.

[0023] S3 specifically refers to: The parameter generation network employs a multilayer perceptron structure, using the initial temperature... T As input, the hidden layer state is obtained through inter-layer recursion:

[0024] in, For activation function, For the first k Layer output vector, and The first k Layer weight matrix and bias vector; The physical information neural network uses the hyperbolic tangent function as the activation function to linearly map the last hidden state to the network output vector. Γ:

[0025] in, and These are the output layer weight matrix and bias vector, respectively. For the first Layer output vector; The output vector is further represented as:

[0026] in, and These are the expansion coefficients of the basis functions. , and These are temperature-related physical parameters. The initial temperature; The prediction expressions for rotary cutting energy and rotational speed are as follows:

[0027]

[0028] in, For the first A B-spline basis function, and The first The expansion coefficients of the basis functions The number of basis functions; Rotary cutting energy With drilling strength The physical relationship can be simplified as follows:

[0029] in, The critical rotary cutting energy. For the first i Drilling intensity; Define the physical residual as:

[0030] in, For rotary cutting energy prediction data, Calculation data for the physical model of energy in rotary cutting; By minimizing the physical residuals, the network output gradually approximates the low-dimensional physical manifold defined by RCFC.

[0031] In S4, the Adam optimization algorithm is used to train the rotary cutting parameter prediction model; the basis function type is B-spline, the number of basis functions is 12, the temperature network has 2 hidden layers with 64 neurons per layer, the activation function is tanh, the learning rate is 0.001, the number of training rounds is 3000, and the physical constraint form is quadratic; the prediction result includes rotary cutting energy. and rotational speed .

[0032] The beneficial effects of this invention are: This invention constructs a thermal shock-cooled rock rotary cutting failure criterion (RCFC) that considers thermal damage evolution and temperature-related fracture mechanics characteristics, and embeds it as a hard physical constraint into a physical information neural network model. This achieves a deep integration of data-driven approaches and the physical mechanism of rock cutting failure, effectively overcoming the core defects of traditional empirical / simplified mechanical models that cannot accurately characterize the thermo-mechanical coupled nonlinear cutting response, and pure data-driven models that lack physical constraints, leading to prediction results that deviate from physical laws and have insufficient generalization ability. It significantly improves the prediction accuracy and physical consistency of key failure parameters such as thermal shock-cooled rock rotary cutting energy and rotation speed, while ensuring the robustness and generalization ability of the model under sparse experimental data conditions. It can provide reliable technical support for the optimization of thermal shock-assisted rock breaking technology, drilling parameter design, and construction safety control in deep high-temperature rock engineering. Attached Figure Description

[0033] Figure 1 The rotary cutting energy under different initial temperature conditions in the method of this invention With drilling strength Schematic diagram of the relationship curve; Figure 2 This is a schematic diagram of the model framework for predicting the failure parameters of thermally cooled rock rotational cutting using failure criteria and physical information neural networks in the method of this invention; Figure 3 This is a schematic diagram illustrating the impact of the weight coefficients of the loss function on the model training process in the method of this invention; Figure 4 This is a schematic diagram illustrating the evolution of the total loss in the method of this invention; Figure 5 The rotary cutting energy in the method of this invention A diagram showing the comparison between predicted and experimental values; Figure 6 The rotational speed in the method of this invention A diagram showing the comparison between predicted and experimental values; Figure 7 The rotary cutting energy in the method of this invention A schematic diagram illustrating the range of predicted value variations; Figure 8The rotational speed in the method of this invention A schematic diagram illustrating the range of predicted value variations; Figure 9 The rotary cutting energy in the method of this invention A diagram showing the comparison of error indicators; Figure 10 The rotational speed in the method of this invention A diagram showing the comparison of error indicators. Detailed Implementation

[0034] Example 1 The method for predicting rock rotational cutting failure under thermal shock disclosed in this invention includes the following steps: S1: Obtain rotary cutting sample data of rocks under thermal shock cooling under different initial temperature conditions (sample data are used to characterize the rotary cutting response characteristics of rocks under thermal shock cooling conditions) and establish a sample dataset of rotary cutting of rocks under thermal shock cooling. S2: Based on the fracture mechanics characteristics, thermal damage parameters, and temperature-related fracture toughness parameters of thermally shock cooled rocks, a failure criterion for rotary cutting of thermally shock cooled rocks is constructed (to characterize the nonlinear relationship between rotary cutting energy and drilling intensity under thermal shock cooling conditions). S3: Construct a physical information neural network model, embed the rotary cutting failure criterion as a physical constraint term into the physical information neural network model, and establish a rotary cutting parameter prediction model; S4: The prediction model is trained based on the dataset obtained in S1, and the predicted parameters of the rotating cutting of the rock by thermal shock are predicted using the trained prediction model.

[0035] Example 2 Based on Example 1, this example provides a detailed explanation of S1 and S2.

[0036] In step S1, the rock sample is processed into a standard cylindrical specimen. The specimen is then heated to a preset initial temperature. After reaching the preset temperature, the specimen is subjected to thermal shock cooling using a cooling medium to obtain thermally shock-cooled rock specimens under different initial temperature conditions. Finally, a rotary cutting test is performed on the thermally shock-cooled rock specimen to collect the drilling pressure during the drilling process. Cutting force and rotational speed ω Parameters were used to establish a sample dataset of thermal shock cooled rock rotary cutting.

[0037] The rock sample is preferably a marble sample, and the sample size is preferably 50 mm in diameter and 100 mm in height; the preset initial temperature is preferably set to 25℃, 200℃, 400℃ and 600℃; the cooling medium is preferably low temperature water; the rotary cutting test is preferably carried out in a digital drilling test system, and confining pressure is applied during the test to simulate the in-situ stress environment.

[0038] Cutting force collected from rotary cutting test Calculating the rotary cutting energy with axial load and drilling strength Their expressions are as follows:

[0039]

[0040] in, For cutting force, This is the axial pressing force. This refers to the rotary cutting area. The expression is as follows:

[0041] in, Where is the drill bit radius. The drill bit tip angle, The affected length.

[0042] This yields the initial temperature required for subsequent model training and validation. T Drilling strength Rotary cutting energy and rotational speed ω Sample data, etc.

[0043] In S2, based on fracture mechanics theory, thermal cracks and primary defects formed inside the rock after thermal shock cooling are considered as initial cracks; under the combined action of drill bit rotation cutting and axial loading, the normal stress on the slip surface of the initial crack is... and shear stress The expression is:

[0044]

[0045] in, For rotary cutting energy, P For drilling strength, Given the initial crack inclination angle, and based on the crack slip condition, the effective shear stress that induces crack propagation. The expression is:

[0046] in For shear stress, The normal stress on the initial crack slip surface, and These are the friction coefficients of the cracked surface and the intact rock, respectively. , For rotary cutting energy, P For drilling strength, This represents the initial crack inclination angle. As the crack tends to propagate at both ends, Type II stress intensity factor at both ends of the induced initial slip crack It can be represented as:

[0047] in, The initial crack length is half. and For fracture toughness parameters, As intrinsic parameters, The effective shear stress is used to induce crack propagation. Damage variables are introduced to characterize the crack density and geometric propagation characteristics induced by thermal shock cooling. Its expression is:

[0048] in, The length of the wing crack. The initial crack length is half. The number of cracks, Crack density, This represents the initial crack inclination angle. The initial damage variables are obtained before the wing crack is activated. :

[0049] in, The initial crack length is half. The number of cracks, Crack density, The initial crack inclination angle is given.

[0050] When taking the effective crack inclination angle When the damage variable is reduced to:

[0051] in, As the initial damage variable, As a damage variable, The initial crack length is half. The number of cracks, Crack density, The length of the wing crack. The initial crack inclination angle is given.

[0052] Critical rotary cutting energy under thermal shock cooling conditions Based on the wing crack model, its expression is:

[0053] in, and b The coefficient related to thermal damage, The initial damage variable is temperature-related. The initial crack length is half. The friction coefficient of intact rock, combined with the Type II stress intensity factor. and critical rotary cutting energy The corrected expression for the stress intensity factor is:

[0054] in, and For fracture toughness parameters, and b The coefficient related to thermal damage, The initial damage variable is temperature-related. The initial crack length is half. is the coefficient of friction of intact rock.

[0055] Define the damage coefficient based on the crack. for:

[0056] in, For fracture toughness parameters, and b The coefficient related to thermal damage, The initial damage variable is temperature-related. The coefficient of friction of intact rock. This is the critical rotary cutting energy.

[0057] To characterize the rotary cutting energy under thermal shock cooling conditions With drilling strength The nonlinear relationship between them is first used to construct an initial rotary cutting failure criterion:

[0058] in, For the breakage angle, For the nonlinear coefficients related to thermal shock cooling damage, This is the critical rotary cutting energy. Based on the critical condition... We can obtain:

[0059] After substituting, we get:

[0060] And the peak rotary cutting energy is obtained:

[0061] in, Peak rotary cutting energy, For the breakage angle, For the nonlinear coefficients related to thermal shock cooling damage, The critical rotary cutting energy. This represents the critical drilling intensity.

[0062] Considering the increase in thermal cracks and the decrease in critical drilling strength caused by thermal shock cooling, a crack damage coefficient is introduced. ,by The rotary cutting failure criterion is modified to obtain the thermal shock cooled rock rotary cutting failure criterion RCFC:

[0063] in, For rotary cutting energy, For drilling strength, For the breakage angle, The initial damage variable is temperature-related. and b The coefficient related to thermal damage, The damage coefficient is based on the crack. The coefficient of friction of intact rock, and the coefficient of the quadratic term. It is given by the following formula:

[0064] Example 3 Based on Example 2, this example provides a detailed explanation of S3 and S4.

[0065] In S3, the parameter generation network uses a multilayer perceptron structure, with the initial temperature as the starting point. T As input, the hidden layer state is obtained through inter-layer recursion, and its expression is:

[0066] in, For activation function, For the first Layer output vector, and The first Layer weight matrix and bias vector.

[0067] The network uses the hyperbolic tangent function as the activation function and linearly maps the last hidden state to the network output vector. :

[0068] in, and These are the output layer weight matrix and bias vector, respectively. For the first Layer output vector.

[0069] The output vector is further represented as:

[0070] in, and These are the expansion coefficients of the basis functions. , and These are temperature-related physical parameters. T This is the initial temperature.

[0071] The basis function calculation layer uses B-spline basis functions to assess drilling intensity. To unfold, rotating cutting energy Predictive expression for rotational speed They are respectively:

[0072]

[0073] in, For the first A B-spline basis function, and The first The expansion coefficients of the basis functions The number of basis functions.

[0074] To embed the RCFC constructed by S2 into the network training process, the rotational cutting energy is... With drilling strength The physical relationship can be simplified as follows:

[0075] in, The critical rotary cutting energy. For the first i Drilling intensity.

[0076] And define the physical residual as:

[0077] in, For rotary cutting energy prediction data, Calculation data for the physical model of rotary cutting energy.

[0078] By minimizing the physical residuals, the network output gradually approximates the low-dimensional physical manifold defined by RCFC.

[0079] Step 4: Based on the rotary cutting sample data obtained in S1, the physical information neural network model is trained to obtain the trained prediction model. The prediction model is then used to output the predicted results of the rotary cutting parameters for thermally cooled rocks. The predicted rotary cutting parameters include the rotary cutting energy. and rotational speed .

[0080] First, the sample data obtained in step 1 is normalized to avoid output variables. and Differences in numerical magnitude affect model training; then, a total loss function is constructed. The weighted sum of the data loss term and the physical loss term is expressed as follows:

[0081] in, These are the weighting coefficients of the loss function, and the data loss term. From the mean square error term and regularization term constitute:

[0082] in,

[0083]

[0084] in: For rotary cutting energy prediction data, For the first A set of experimental data on rotary cutting energy. For the first The weights of the trainable parameters. This represents the expectation operator, which is implemented by approximating the mean of a batch of training samples. The physical loss term... The expression is:

[0085] in, For rotary cutting energy prediction data, Data for calculating the energy physics model of rotary cutting. This represents the expectation operator.

[0086] In S4, the Adam optimization algorithm is used to train the physical information neural network model. Preferably, the basis function type is B-spline, the number of basis functions is 12, the temperature network has 2 hidden layers with 64 neurons per layer, the activation function is tanh, the learning rate is 0.001, the number of training rounds is 3000, and the physical constraint form is quadratic. Furthermore, the optimal value of the loss function weight coefficient is determined to be 0.46 through verification. The trained model outputs the prediction results of the rotational cutting parameters of rock under different thermal shock cooling temperatures, including the rotational cutting energy. and rotational speed By comparing the predicted values ​​with the measured values, it is possible to predict the parameters and characterize the failure evolution of the rotating cutting behavior of rocks cooled by thermal shock.

[0087] Example 4 According to the aforementioned steps of the method of the present invention, the prediction results of the rotary cutting parameters of marble cooled by thermal shock at 200℃ are output using the prediction model. The steps are as follows: S1, conduct rotary cutting experiments to obtain rotary cutting sample data of thermally shock cooled rocks under different initial temperature conditions, including initial temperature. Drilling strength Rotary cutting energy and rotational speed .

[0088] S2. Based on the fracture mechanics characteristics, thermal damage parameters, and temperature-related fracture toughness parameters of thermally shock cooled rocks, a failure criterion for rotary cutting of thermally shock cooled rocks is constructed.

[0089] S3 constructs a physical information neural network model for predicting the parameters of rotating cutting of rocks under thermal shock, and embeds the rotating cutting failure criterion constructed in S2 as a physical constraint term into the physical information neural network model.

[0090] S4, based on the rotary cutting sample data obtained in S1, trains the physical information neural network model to obtain the trained prediction model, and uses the prediction model to output the predicted results of the rotary cutting parameters (rotary cutting energy) of the rock cooled by thermal shock. and rotational speed ).

[0091] The relevant data are shown in Tables 1, 2 and 3 below, and the specific analysis process is described in the experimental verification below.

[0092] Example 5 Referring to Example 1, following the steps outlined above, the prediction model was used to output the predicted results of the rotary cutting parameters for marble cooled by thermal shock at 400°C. The relevant data are shown in Tables 1, 2, and 3 below. The specific analysis process is described in the experimental verification below.

[0093] Example 6 Referring to Example 1, following the steps outlined above, the prediction model was used to output the predicted results of the rotary cutting parameters for marble cooled by thermal shock at 600℃. The relevant data are shown in Tables 1, 2 and 3 below. The specific analysis process is described in the experimental verification below.

[0094] Experimental verification: 1) Thermal shock cooling rotary cutting test.

[0095] To verify the predictive effect of the method of this invention on the rotary cutting parameters of rock subjected to thermal shock cooling, rotary cutting tests were conducted on marble samples under different initial temperature conditions. The samples were machined into standard cylindrical specimens with a diameter of 50 mm and a height of 100 mm. The preset initial temperatures were 25℃, 200℃, 400℃, and 600℃, and the cooling medium was low-temperature water. After thermal shock cooling, rotary cutting tests were conducted in a digital drilling test system, and confining pressure was applied during the tests to simulate the in-situ stress environment. Drilling pressure was collected during the tests. Cutting force Displacement and rotational speed Using parameters such as [parameters], a dataset of samples of thermally shock-cooled rotary cutting rocks was established. The experimental conditions and modeling inputs remained consistent with those described in S1 to S4 above.

[0096] First, rotary cutting tests were conducted on rocks cooled by thermal shock under different initial temperature conditions to calculate the rotary cutting energy. and drilling strength Establish a system including initial temperature Drilling strength Rotary cutting energy and rotational speed The sample database was used. The results show that as the initial temperature increases, thermal shock cooling leads to the continuous development of internal cracks and damage in the rock, and the rotary cutting response gradually exhibits obvious nonlinear characteristics.

[0097] Second, further study the rotary cutting energy under different initial temperature conditions. With drilling strength The relationship is plotted as a curve, such as... Figure 1 As shown. The results indicate that under normal temperature conditions... The curvature of the relationship curve is significantly lower than that after thermal shock cooling, indicating that the thermal damage induced by thermal shock cooling significantly enhances the nonlinear characteristics of the rotary cutting response. To characterize this change, a linear cutting response is introduced as an asymptote for comparative analysis, where the slope of the asymptote is... The slopes at 25℃, 200℃, 400℃, and 600℃ are 0.69, 0.62, 0.56, and 0.48, respectively. The above slope trends are consistent with... The consistent trend of the nonlinear envelope variation indicates that thermal shock cooling alters the rock rotational cutting failure mechanism, and verifies the effectiveness of the thermal shock cooling rock rotational cutting failure criterion RCFC constructed in S2 for rotational cutting energy. With drilling strength The rationality of representing the nonlinear relationship between them.

[0098] Third, the physical and mechanical parameters of the samples after thermal shock cooling were tested to obtain porosity, longitudinal wave velocity, compressive strength, elastic modulus, and fracture toughness parameters under different temperature conditions, as shown in Table 1. Based on these parameters, the temperature-related damage evolution characteristics were determined. These parameters were used to construct the thermal shock cooling rock rotational cutting failure criterion (RCFC) in S2 and as the physical constraint input for the physical information neural network model in S3.

[0099] Fourth, construct the Thermal Shock Cooling Rock Rotation Cutting Failure Criterion (RCFC) according to S2 above, and construct the Physical Information Neural Network Model according to S3 above. The specific framework is shown in [link to framework]. Figure 2 The model uses B-spline basis functions, with 12 basis functions. The temperature network has two hidden layers, each with 64 neurons. The activation function is tanh, the optimizer is Adam, the learning rate is 0.001, the training epochs are 3000, and the physical constraints are quadratic. Model parameters are shown in Table 2.

[0100] Table 1

[0101] Table 2

[0102] 2) Model training and prediction result verification.

[0103] To investigate the impact of physical constraints on the model's prediction performance, the weighting coefficients in the total loss function were analyzed. A comparative analysis was conducted. The total loss function is a weighted average of the data loss term and the physical loss term. During model training, this is adjusted... The values ​​of the weighting coefficients were varied, and the convergence characteristics and prediction results were compared under different conditions. The results show that the weighting coefficients... It has a significant impact on model training stability and prediction accuracy, when When the data is too small, the model is more inclined to fit the data; when the data is too small, the model is more inclined to fit the data. When the value is too large, the model is more constrained by physical limitations, and the local sample fitting ability decreases; through verification, the optimal value of the loss function weight coefficient is determined to be 0.46.

[0104] like Figure 3 , Figure 4 The model was trained under optimal weighting conditions to obtain a prediction model for the parameters of thermally shock-cooled rock rotary cutting. The results show that... When the value is 0.46, the total loss decreases faster and the training process is more stable, indicating that the method of the present invention achieves a good balance between data consistency and physical consistency.

[0105] The trained model was used to study the rotary cutting energy under different initial temperature conditions. and rotational speed Predictions were made, and the predicted results were compared with the experimentally measured results. The results show that the predicted values ​​and measured values ​​generally maintain good consistency, and can accurately characterize the variation law of rotary cutting parameters of rock cooled by thermal shock under different temperature conditions. Figure 5 and Figure 6 It can be seen that the rotary cutting energy and rotational speed The predicted values ​​and the measured values ​​are close in distribution, and no obvious systematic deviation is observed.

[0106] At the same time, by Figure 7 and Figure 8 It can be seen that the rotary cutting energy under different initial temperature conditions and rotational speed The fluctuation range is basically consistent with the trend of the measured results, indicating that the method of the present invention can not only reflect the average variation characteristics of the rotary cutting parameters, but also reflect the discreteness and stability changes of the rotary cutting response under thermal shock cooling conditions.

[0107] 3) Verification of error indicators.

[0108] To further evaluate the prediction accuracy of the method of this invention, error analysis was performed on the prediction results of the training set and validation set. The coefficient of determination was used. The root mean square error (RMSE), mean absolute error (MAE), and coefficient of variation (CV) are indicators of the effectiveness of rotary cutting energy. and rotational speed The prediction results were quantitatively evaluated. See Table 3. Figure 9 and Figure 10 .

[0109] Table 3

[0110] The results show that the method of the present invention reduces the rotary cutting energy. and rotational speed Both tests demonstrate high prediction accuracy, with similar error levels between the training and validation sets, indicating that the model does not exhibit significant overfitting and possesses good generalization ability and stability. Furthermore, embedding the Thermal Shock Cooling Rock Rotation Cutting Failure Criterion (RCFC) into the physical information neural network model effectively constrains the model output, maintaining good prediction reliability even under limited sample conditions.

[0111] This invention studies the rock rotational cutting response under thermal shock cooling conditions through thermal shock cooling rotational cutting experiments, physical and mechanical parameter testing, and physical information neural network modeling. The rotational cutting failure criterion RCFC is embedded into the prediction model, showing that: (1) As the initial temperature rises, the thermal shock cooling effect causes cracks and damage inside the rock to develop continuously, and the rotational cutting energy and rotational speed The gradual emergence of obvious nonlinear variation characteristics indicates that thermal damage evolution has a significant impact on the rock rotation cutting failure process; (2) After combining the thermal shock cooling rock rotary cutting failure criterion RCFC with the physical information neural network, the model prediction values ​​and experimental measured values ​​generally maintain good consistency, and can accurately characterize the rotary cutting energy under different temperature conditions. and rotational speed The changing pattern; (3) Error analysis results show that the method of the present invention has good control over rotary cutting energy. and rotational speed Both models exhibit high prediction accuracy, and the error levels of the training and validation sets are similar, indicating that the model does not exhibit significant overfitting and has good generalization ability and stability. (4) The method of the present invention has high prediction accuracy, good physical consistency and strong engineering applicability, and can provide an effective method for the analysis and performance evaluation of rotary cutting parameters in high temperature rock drilling, thermal shock cooling rock breaking and deep underground engineering.

Claims

1. A method for predicting failure of rock undergoing rotary cutting due to thermal shock cooling, characterized in that, Includes the following steps: S1: Obtain rotary cutting sample data of thermally shock cooled rocks under different initial temperature conditions, and establish a rotary cutting sample dataset of thermally shock cooled rocks; S2: Based on the fracture mechanics characteristics, thermal damage parameters and temperature-related fracture toughness parameters of thermally shock cooled rocks, construct the failure criteria for rotary cutting of thermally shock cooled rocks; S3: Construct a physical information neural network model, embed the rotary cutting failure criterion as a physical constraint term into the physical information neural network model, and establish a rotary cutting parameter prediction model; S4: The prediction model is trained based on the dataset obtained in S1, and the predicted parameters of the rotating cutting of the rock by thermal shock are predicted using the trained prediction model.

2. The method for predicting rock rotational cutting failure by thermal shock cooling according to claim 1, characterized in that, First, the sample data obtained from S1 is normalized. Then, a total loss function is constructed, which is a weighted sum of the data loss term and the physical loss term. in, These are the weighting coefficients of the loss function; Data loss items From the mean square error term and regularization term constitute: in, For rotary cutting energy prediction data, For the first A set of experimental data on rotary cutting energy. For the first The weights of the trainable parameters. This represents the expectation operator and the physical loss term. L RCFC for: in, For rotary cutting energy prediction data, Data for calculating the energy physics model of rotary cutting. This represents the expectation operator.

3. The method for predicting rock rotational cutting failure by thermal shock cooling according to claim 2, characterized in that, The weight coefficient of the loss function is set to 0.

46.

4. The method for predicting rock rotational cutting failure by thermal shock cooling according to claim 3, characterized in that, S1 specifically refers to: The sample is heated to a preset initial temperature. After reaching the preset temperature, the sample is subjected to thermal shock cooling using a cooling medium to obtain different initial temperatures. T Rock samples were subjected to thermal shock cooling under certain conditions, followed by a rotary cutting test to collect drill pressure data during the drilling process. F w Cutting force F t and rotational speed ; Cutting force collected from rotary cutting test and drilling pressure Calculate rotary cutting energy and drilling strength : in, For cutting force, For drilling pressure, The rotating cutting area is the area of ​​rotation. A The expression is as follows: in, Where is the drill bit radius. The drill bit tip angle, For the affected length; Based on the different initial temperatures mentioned above and drilling strength obtained under different initial temperature conditions Rotary cutting energy and rotational speed Establish a sample dataset of rock rotation cutting samples cooled by thermal shock.

5. The method for predicting rock rotational cutting failure by thermal shock cooling according to claim 4, characterized in that, The preset initial temperature is 25℃, 200℃, 400℃ or 600℃; the cooling medium is low temperature water; the rotary cutting test is carried out in a digital drilling test system, and confining pressure is applied during the test to simulate the in-situ stress environment.

6. The method for predicting rock rotational cutting failure by thermal shock cooling according to claim 5, characterized in that, The failure criterion for rotary cutting in S2 is: in, For rotary cutting energy, For drilling strength, For the breakage angle, The initial damage variable is temperature-related. and b are coefficients related to thermal damage. The damage coefficient is based on the crack. is the coefficient of friction of intact rock.

7. The method for predicting rock rotational cutting failure by thermal shock cooling according to claim 6, characterized in that, S3 specifically refers to: The parameter generation network employs a multilayer perceptron structure, using the initial temperature... T As input, the hidden layer state is obtained through inter-layer recursion: in, For activation function, For the first k Layer output vector, and The first k Layer weight matrix and bias vector; The physical information neural network uses the hyperbolic tangent function as the activation function to linearly map the last hidden state to the network output vector. Γ : in, and These are the output layer weight matrix and bias vector, respectively. For the first Layer output vector; The output vector is further represented as: in, and These are the expansion coefficients of the basis functions. , and These are temperature-related physical parameters. The initial temperature; The prediction expressions for rotary cutting energy and rotational speed are as follows: in, For the first A B-spline basis function, and The first The expansion coefficients of the basis functions The number of basis functions; Rotary cutting energy With drilling strength The physical relationship can be simplified as follows: in, The critical rotary cutting energy. For the first i Drilling intensity; Define the physical residual as: in, For rotary cutting energy prediction data, Calculation data for the physical model of energy in rotary cutting; By minimizing the physical residuals, the network output gradually approximates the low-dimensional physical manifold defined by RCFC.

8. The method for predicting rock rotational cutting failure by thermal shock cooling according to claim 7, characterized in that, In step S4, the Adam optimization algorithm is used to train the rotary cutting parameter prediction model; the basis function type is B-spline, the number of basis functions is 12, the temperature network has 2 hidden layers with 64 neurons per layer, the activation function is tanh, the learning rate is 0.001, the number of training rounds is 3000, and the physical constraint form is quadratic; the prediction result includes rotary cutting energy. and rotational speed .

Images