Evaluation method and system for composite impact multi-tooth rock breaking test
By conducting composite impact multi-tooth rock breaking tests, the main controlling factors were identified and a neural network model was established. The nonlinear relationship was processed through integrated learning, which solved the problems of high time consumption and low accuracy in traditional rock breaking tests. This enabled efficient and accurate rock breaking prediction, providing optimized parameters for drilling in deep hard rock formations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2024-11-06
- Publication Date
- 2026-05-08
AI Technical Summary
Traditional rock fracturing tests are time-consuming and costly, and are difficult to apply directly to engineering sites, resulting in low accuracy in rock fracturing prediction, which affects engineering design and risk control.
By conducting composite impact multi-tooth rock breaking tests, the main controlling factors were identified as variable parameters, a neural network model was established, and BP neural network, RF model and XGBoost model were integrated for learning to predict rock damage morphology.
It enables efficient and accurate prediction of rock-breaking effects in complex geological environments, provides optimized parameter references for composite drilling in deep hard rock formations, and improves the operability and prediction accuracy of engineering applications.
Smart Images
Figure CN121997690A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of rock crushing technology, and specifically relates to an evaluation method and system for composite impact multi-tooth rock crushing test. Background Technology
[0002] With the continuous development of oil and gas exploration and mining technologies, how to effectively improve drilling and mining efficiency and reduce rock damage is a key issue in current engineering technology. Correspondingly, rock fracturing test, as an important link in oil and gas exploration and development, provides necessary rock mechanics parameters for key engineering links such as drilling, fracturing, and well completion, thereby helping to optimize the design of production enhancement measures and reduce wellbore instability and engineering risks.
[0003] However, traditional rock fracturing tests rely on a large amount of experimental data and empirical formulas, which are both time-consuming and costly. Furthermore, the evaluation of the rock fracturing effect is often based on indoor tests or numerical simulation methods. Since there are many factors that cause rock fracturing damage and they are not linearly related, this method is difficult to apply directly in engineering fields. It also affects the accuracy of rock fracturing test predictions and the evaluation conclusions of the rock.
[0004] Therefore, overcoming the shortcomings of the existing technology is an urgent problem to be solved in this technical field. Summary of the Invention
[0005] To address the above problems, this invention proposes an evaluation method for composite impact multi-tooth rock breaking tests, comprising the following steps:
[0006] Through composite impact tests, the main controlling factors that cause differences in rock damage state were selected and determined as variable parameters;
[0007] Based on the composite impact test, the test results of rock damage and the corresponding variable parameters are obtained as sample data. The variable parameters are changed to carry out the next set of composite impact tests until all sample data under different parameters are obtained, and a dataset is established.
[0008] Based on the dataset, establish the neural network structure and activation function, and train the neural network to form a neural network model;
[0009] Input specific variable parameters to be tested into the neural network model, output the corresponding test prediction results, and complete the test evaluation.
[0010] Furthermore, the variable parameters include: sleeve impact velocity, sleeve rotation speed, drill bit shape, drill bit inclination angle, drill bit diameter, rock type, formation Poisson's ratio, elastic modulus, and confining pressure.
[0011] Furthermore, the test results included: penetration depth, damage area, and fragmentation volume.
[0012] Furthermore, in the process of establishing the neural network structure based on the dataset, the dataset is divided into a training set, a validation set, and a test set, with the ratio of training set:test set:validation set = 70-80%:10-15%:10-15%;
[0013] The training set is used to train the neural network model;
[0014] The validation set is used to tune and optimize model parameters;
[0015] The test set is used to test the model's generalization ability.
[0016] Furthermore, in the process of establishing the neural network structure based on the dataset;
[0017] The neural network structure includes an input layer, an output layer, and hidden layers. The number of neurons in the input and output layers is consistent with the variable parameters and the number of experimental results. The number of hidden layers is at least one.
[0018] Furthermore, the activation function is a non-saturating activation function.
[0019] Furthermore, the neural network model includes a first-layer ensemble learning framework and a second-layer ensemble learning framework.
[0020] The first-layer ensemble learning framework includes a BP neural network, a RF model, and an XGBoost model. The input parameters of the first-layer ensemble learning framework are variable parameters, and the output parameters are the calculation results of the BP neural network, the calculation results of the RF model, and the calculation results of the XGBoost model.
[0021] The second-layer ensemble learning framework is used to fuse the computation results of the BP neural network, the RF model, and the XGBoost model to output the experimental prediction results.
[0022] This invention also proposes an evaluation system for composite impact multi-tooth rock breaking tests, comprising:
[0023] The parameter selection module is used to select and determine the main controlling factors that cause differences in rock damage state as variable parameters through composite impact tests.
[0024] The dataset creation module, based on composite impact tests, obtains the test results of rock damage and the corresponding variable parameters as sample data. It changes the variable parameters to conduct the next set of composite impact tests until all sample data under different parameters are obtained, and then creates the dataset.
[0025] The model building module establishes the neural network structure and activation function based on the dataset and trains the neural network. The neural network is used to learn from the database and predict the damage morphology of rocks after impact tests.
[0026] The model computation module takes specific variable parameters to be tested as input to the neural network model, outputs the corresponding experimental prediction results, and completes the experimental evaluation.
[0027] Furthermore, the variable parameters include: sleeve impact speed, sleeve rotation speed, drill bit shape, drill bit inclination angle, drill bit diameter, rock type, formation Poisson's ratio, elastic modulus, and confining pressure.
[0028] The test results include: penetration depth, damage area, and fragmentation volume.
[0029] Furthermore, the neural network model includes a first-layer ensemble learning framework and a second-layer ensemble learning framework.
[0030] The first-layer ensemble learning framework includes a BP neural network, a RF model, and an XGBoost model. The input parameters of the first-layer ensemble learning framework are variable parameters, and the output parameters are the calculation results of the BP neural network, the calculation results of the RF model, and the calculation results of the XGBoost model.
[0031] The second-layer ensemble learning framework is used to fuse the computation results of the BP neural network, the RF model, and the XGBoost model to output the experimental prediction results.
[0032] Compared with the prior art, the embodiments of the present invention have at least the following advantages:
[0033] The evaluation method and system for composite impact multi-tooth rock breaking tests of this invention first determines the main controlling factors affecting the test before constructing a neural network model to ensure the comprehensiveness of the input parameters. Test results are obtained based on continuous adjustment of the main controlling factors, and a dataset of rock breaking effects is established through rock breaking tests. Then, the neural network model automatically learns features from multiple datasets. Based on understanding the main controlling factors, it achieves the processing of complex nonlinear relationship data, enabling efficient and accurate prediction even in complex geological environments. This provides an effective reference for optimizing engineering and tool parameters during composite drilling in deep hard rock formations and is operable in practical engineering applications.
[0034] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description and the drawings. Attached Figure Description
[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0036] Figure 1 A flowchart illustrating the evaluation method of the composite impact multi-tooth rock breaking test in an embodiment of the present invention is shown.
[0037] Figure 2 A block diagram of the evaluation system for the composite impact multi-tooth rock breaking test in an embodiment of the present invention is shown;
[0038] Figure 3 A schematic diagram of the test scan for the evaluation method of the composite impact multi-tooth rock breaking test in an embodiment of the present invention is shown;
[0039] Figure 4 A schematic diagram of the neural network structure of the evaluation method for the composite impact multi-tooth rock breaking test in an embodiment of the present invention is shown. Detailed Implementation
[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0041] This invention provides an evaluation method and system for composite impact multi-tooth rock breaking tests. Figure 1 A flowchart illustrating the evaluation method for the composite impact multi-tooth rock-breaking test in an embodiment of the present invention is shown. Figure 1 The evaluation method for composite impact multi-tooth rock breaking test includes the following steps:
[0042] Through composite impact tests, the main controlling factors that cause differences in rock damage state were selected and determined as variable parameters;
[0043] Based on the composite impact test, the test results of rock damage and the corresponding variable parameters are obtained as sample data. The variable parameters are changed to conduct the next set of composite impact tests until all sample data under different parameters are obtained, and a dataset is established.
[0044] The neural network structure and activation function are established based on the dataset, and the neural network is trained to form a neural network model. The neural network model is used to learn from the database and predict the test results of rock damage after impact test.
[0045] Input specific variable parameters to be tested into the neural network model, output the corresponding test prediction results, complete the test evaluation, and provide a reference for composite drilling in deep hard rock formations through the evaluation data.
[0046] Furthermore, in practice, the accuracy of the neural network model can be tested and optimized by comparing the predicted results with the actual experimental results.
[0047] The evaluation method for composite impact multi-tooth rock breaking test of the present invention first determines the main controlling factors affecting the test before constructing the neural network model to ensure the comprehensiveness of the input parameters. The test results are obtained based on the continuous adjustment of the main controlling factors. A dataset of rock breaking effect is established through rock breaking test. Then, the neural network model automatically learns features from multiple data. Based on the grasp of the main controlling factors, it realizes the processing of complex nonlinear relationship data, so as to maintain efficient and accurate prediction even when facing complex geological environments. It provides an effective reference for the optimization of engineering and tool parameters in the process of composite drilling in deep hard rock formations and has operability in practical engineering applications.
[0048] Specifically, the variable parameters can be divided into engineering parameters, impact tool parameters, and rock parameters;
[0049] The engineering parameters used in the experiment included: sleeve impact speed and sleeve torsional speed;
[0050] Tool parameters include: drill tooth type, drill tooth inclination angle, and drill tooth diameter;
[0051] Rock parameters include: rock type, Poisson's ratio of the formation, elastic modulus, and confining pressure.
[0052] The specific parameter selections are shown in Table 1.
[0053] Table 1. Parameter values affecting the depth and volume of impact craters.
[0054] Parameter name Parameter variation range Change step size Impact velocity (m / s) 5-9 1 Torsional speed (rpm) 40-60 10 Drill tooth type 3 (Planar teeth / Triangular teeth / Axe-shaped teeth) 1 Drill tooth inclination angle 10°-30° 5° Drill tooth diameter (mm) 13 / 16 / 19 3 Rock types 3 (Granite / Sandstone / Limestone) 1 Poisson's ratio of formations 0.1-0.3 0.05 Elastic modulus (10 GPa) 1.5-8.5 1 Confining pressure (MPa) 20-60 20
[0055] The settings and value ranges of the nine main control factors are all derived from actual well logging data, which has comprehensiveness and wide applicability. Among them, the change step size refers to the amount of change of the variable in each iteration or calculation.
[0056] Correspondingly, refer to Figure 3The test results selected the penetration depth, damage area, and fragmentation volume. All of the above data can be directly collected using a three-dimensional scanning device during the test.
[0057] Among them, by comparing and analyzing the rock-breaking effects of different drill bits and drill bit combinations based on the characteristics of drill bit geometry, inclination angle, and diameter, engineers can be provided with a basis for optimizing drill bit design, drill bit shape, and parameter configuration, ensuring the best drilling effect in different rock formations.
[0058] To address the diversity of rock types, this technology incorporates rock characteristic parameters (rock type, elastic modulus, Poisson's ratio, confining pressure) to conduct a detailed evaluation of the compatibility between percussion drilling tools, drill bits, and formations. This technology can adjust the parameters of composite percussion drilling tools and optimize drill bit types for different geological characteristics, ensuring optimal compatibility between the drilling tools and the formation, thereby improving the reliability and economy of drilling operations.
[0059] For details, please refer to Figure 4 In the process of establishing the neural network structure based on the dataset, the dataset is divided into:
[0060] The training set is used to train the neural network model;
[0061] The validation set is used to tune and optimize model parameters;
[0062] The test set is used to test the model's generalization ability;
[0063] The training set:test set:validation set ratio is 70-80%:10-15%:10-15%. Based on the conventional ratio of 60:20:20, increasing the proportion of the training set reduces the possibility of information leakage and improves the efficiency of the response model.
[0064] Meanwhile, in the process of establishing the neural network structure based on the dataset,
[0065] The neural network structure is determined by using variable parameters as input parameters and experimental results as output parameters. The neural network structure includes an input layer, an output layer, and hidden layers.
[0066] The number of neurons in the input layer is determined by the dimension of the input parameters. Each neuron corresponds to a feature of the input data, that is, each neuron corresponds to a variable parameter. Referring to Table 1, based on the selection of nine variable parameters, the number of neurons in the input layer in this embodiment is nine.
[0067] Correspondingly, the number of neurons in the output layer is determined by the nature of the task. In this application, the number of neurons in the output layer is 3, which corresponds to the penetration depth, damage area, and fragmentation volume.
[0068] The number of neurons in the hidden layer can be adjusted experimentally, but the number of hidden layers should be at least one, preferably one or two.
[0069] By carefully selecting the number of neurons in the input, output, and hidden layers, the network structure can be matched with the problem requirements, thereby improving the model's performance. Furthermore, by reasonably setting the number of hidden layers, a balance can be achieved between learning ability and generalization ability, avoiding overfitting the training data or underfitting and failing to learn complex patterns. An appropriate number of neurons helps reduce the consumption of computational resources, especially when the network is small, which can reduce the time cost of training and inference.
[0070] Specifically, the activation functions are non-saturating activation functions, including ReLU and SELU, which effectively accelerate the convergence speed of the model. Among them, the ReLU (Rectified Linear Unit) activation function can effectively solve the gradient vanishing problem in the neural network training process, thereby improving the model's ability to learn complex data.
[0071] Specifically, in this application, the neural network model includes a first-layer ensemble learning framework and a second-layer ensemble learning framework;
[0072] The first-layer ensemble learning framework includes a BP neural network, a RF model, and an XGBoost model. The BP neural network, RF model, and XGBoost model form the first-layer ensemble learning framework through ensemble learning. The input parameters of the first-layer ensemble learning framework are variable parameters, and the output parameters are the calculation results of the BP neural network, the calculation results of the RF model, and the calculation results of the XGBoost model.
[0073] The second-layer ensemble learning framework takes the computational results from the first layer as input. That is, the second-layer ensemble learning framework is used to fuse the computational results of the BP neural network, the RF model, and the XGBoost model to output the experimental prediction results.
[0074] Among them, during the training of the BP neural network model;
[0075] It's important to note that the BP neural network uses the steepest gradient descent method to backpropagate error information. Its learning process includes two stages: forward propagation of input information and backward adjustment of error information. In the forward propagation stage, input information is passed from the input layer through the hidden layers to the output layer. If the desired output is not obtained at the output layer, the output error will propagate backward layer by layer until it reaches the input layer. Along the way, the connection weights and thresholds between neurons in each layer are modified to gradually minimize the error. The forward propagation and backward adjustment processes alternate until the network output error is reduced to a preset error range or a preset number of learning iterations is reached.
[0076] For example, a three-layer neural network structure consists of engineering parameters (impact velocity, torsional velocity), drill tooth parameters (drill tooth type, inclination angle, diameter), and rock parameters (rock type, Poisson's ratio, elastic modulus, confining pressure).
[0077] During the forward propagation phase, the input information enters the network from the input layer, passes through the calculations of each layer in turn, and is transmitted to the output layer, ultimately obtaining the penetration depth, damage area, and rock-breaking volume of the output layer.
[0078] If the expected output value is not obtained at the output layer, the sum of squared errors between the output and the expected value is taken as the objective function. Backpropagation is then initiated, and the partial derivatives of the objective function with respect to the weights of each neuron are calculated layer by layer. This forms the gradient of the objective function with respect to the weight vector, which serves as the basis for modifying the weights. The network's learning is completed during this weight modification process. When the error reaches the expected value, the neural network's learning process ends.
[0079] For example, during the training of an RF model;
[0080] First, variable parameters are used as model inputs, including engineering parameters, drill bit parameters, and rock parameters, as follows: engineering parameters (impact velocity, torsional velocity), drill bit parameters (drill bit type, inclination angle, diameter), and rock parameters (rock type, Poisson's ratio, elastic modulus, confining pressure); the output targets of the RF model are penetration depth, damage area, and rock breaking volume.
[0081] The RF model captures the nonlinear relationship between variable parameters and experimental results by integrating multiple decision trees;
[0082] This allows the RF model to gradually optimize its prediction accuracy during training by automatically selecting features and splitting nodes.
[0083] Furthermore, during the training phase of the RF model, the accuracy of the RF model can be improved by adjusting its hyperparameters, including the number of decision trees and the maximum depth. Cross-validation methods (such as K-fold cross-validation) can also be used to validate the RF model, ensuring that it has good generalization ability on different datasets.
[0084] Trained and validated RF models can be used to predict actual rock-breaking processes. When new engineering parameters, drill bit parameters, and rock parameters are input, the model outputs predicted penetration depth, damage area, and rock-breaking volume, helping engineers evaluate rock-breaking performance under different conditions. Furthermore, by analyzing the importance of features in the model, it is possible to identify which key controlling factors have the greatest impact on experimental results, thereby guiding the optimization and adjustment of engineering parameters. This improves efficiency and reduces experimental costs.
[0085] For example, when training using an XGBoost model,
[0086] The tree model of XGBoost can be represented as:
[0087]
[0088] Where F represents the regression tree space, f k Let f represent the k-th subtree. k (x i ) represents the output of the corresponding subtree k, x i Corresponding to the input sample, This is the output of the algorithm, where K is the number of iterations.
[0089] Its objective function can be expressed as:
[0090] O(θ)=L(θ)+Ω(θ) (2)
[0091] Where O(θ) is the objective function, L(θ) is the training error term, and Ω(θ) is the constraint regularization term, which is used to constrain the complexity of the algorithm and improve its generalization ability.
[0092] The XGBoost model is trained using an additive approach, allowing... Let represent the output of the i-th tree in the t-th iteration, then:
[0093]
[0094] in, f is the output of the i-th tree in the (t-1)-th iteration. t (x i Let be the predicted value to be added. Further, the objective function of the algorithm is expressed as:
[0095]
[0096] Among them, y i The true value is given by m, where m is the sample size, and O is the distance between the two samples. t Let be the objective function of the algorithm. To facilitate rapid optimization, the algorithm approximates the objective function using a second-order Taylor expansion:
[0097]
[0098] Where g i and h i Let represent the first and second derivatives of the error function, respectively. Substituting the optimal value of the leaf node, the number of leaf nodes in the t-th iteration is denoted as T. t And the evaluation value is labeled as w j,t Therefore, the objective function is:
[0099]
[0100] Where r and λ are regularization coefficients.
[0101] The following is a detailed description combining variable parameters and experimental results:
[0102] First, variable parameters need to be prepared as model inputs, including engineering parameters, drill bit parameters, and rock parameters, specifically: engineering parameters (impact velocity, torsional velocity), drill bit parameters (drill bit type, inclination angle, diameter), and rock parameters (rock type, Poisson's ratio, elastic modulus, confining pressure). In the data preparation stage, the raw data is first processed as follows:
[0103] Numerical variables, such as impact velocity, elastic modulus, and Poisson's ratio, are directly used as model inputs.
[0104] Categorical variable processing: such as drill tooth type and rock type, use one-hot encoding or label encoding to convert categorical data into numerical format;
[0105] Based on the above processing, data standardization is performed, and numerical features are standardized or normalized to improve model convergence speed and performance.
[0106] In this process, variable parameters are used as input features, and the output targets of the model are penetration depth, damage area, and rock-breaking volume.
[0107] The specific steps are as follows:
[0108] Input data: Input the main control factors such as impact velocity, torsional velocity, drill tooth type, inclination angle, diameter, rock type, Poisson's ratio, elastic modulus, and confining pressure into the model.
[0109] Output data: The model's prediction targets include penetration depth, damage area, and rock break volume.
[0110] Objective function: Based on the actual application requirements, select an appropriate loss function for optimization, such as mean squared error (MSE) to minimize the difference between the predicted value and the true value.
[0111] When new engineering parameters, drill bit parameters, and rock parameters are used as input, the trained and validated XGBoost model can output corresponding predictions of penetration depth, damage area, and rock-breaking volume. This helps engineers evaluate the rock-breaking effect, provides important guidance for engineering optimization, and helps adjust drill bit parameters to improve rock-breaking efficiency.
[0112] After the three models are trained, the calculation results of the three models are weighted and averaged (the weights are provided by the validation set) to output the experimental prediction results, thus achieving the effect of integrating all models.
[0113] This application constructs a neural network model by integrating BP neural networks, RF models, and XGBoost models through ensemble learning. It leverages the strengths of each model to improve prediction accuracy and generalization ability. Furthermore, the reasonable selection of nine key control factors enhances the comprehensiveness of the experimental prediction results. This effectively reduces errors in damage morphology prediction, making the prediction results more reliable and accurate, covering a wider range of working conditions, and improving the universality of the method. Secondly, the introduction of the ensemble neural network provides more possibilities for the model to handle complex nonlinear relationships, enabling it to maintain efficient and accurate predictions even in complex geological environments.
[0114] To better illustrate the practical application of the present invention, the dataset will be further described.
[0115] In actual geological exploration and drilling work, different working environments and rock strata structures make the rock breaking process extremely complex. Referring to Table 1 above, this embodiment will explain the process from the perspective of nine variable parameters: sleeve impact velocity, sleeve rotation speed, drill bit shape, drill bit inclination angle, drill bit diameter, rock type, formation Poisson's ratio, and confining pressure.
[0116] In the rock-breaking process, the performance and parameters of the drilling tools play a crucial role in the rock-breaking effect. Therefore, this embodiment selected the following sets of tool parameters for experimentation.
[0117] Sleeve impact speed: In this embodiment, three different impact speeds are set, namely 5m / s, 7m / s and 9m / s. The sleeve impact speed directly affects the kinetic energy of the drill bit when it hits the rock, thus determining the force exerted by the drill bit on the rock surface. By setting these three speeds, we can study how the damage morphology of the rock varies with the change of kinetic energy at different speeds.
[0118] Sleeve rotation speed: In this embodiment, two rotation speeds, 40 rpm and 60 rpm, were selected to observe the effect of increasing the rotation speed on rock damage.
[0119] Drill tooth shape: Flat drill teeth and triangular drill teeth are selected. Triangular drill teeth can generate a higher stress concentration effect and are suitable for harder rocks, while flat drill teeth are more suitable for softer rock layers.
[0120] Drill tooth inclination angle: Set the drill tooth inclination angle to 10 degrees and 30 degrees to determine the angle of contact between the drill bit and the rock and the direction of the force exerted by the drill bit.
[0121] Drill tooth diameter: Two common drill tooth diameters, 13 mm and 16 mm, were selected for the experiment to analyze the influence of drill tooth diameter on rock damage morphology.
[0122] Besides tool parameters, rock parameters also have a significant impact on the rock fracturing process. Especially under actual geological conditions, different rock types, formation physical properties, and external stress states can all significantly affect the rock fracturing process. This embodiment selected the following environmental parameters:
[0123] Rock types: Three different rock samples were selected in this embodiment: granite, limestone, and sandstone. Granite is a relatively hard rock, limestone is a medium-hard rock, and sandstone is a relatively soft rock. The mechanical properties of different rocks have a significant impact on the damage morphology during the rock fracturing process.
[0124] Formation Poisson's ratio: Poisson's ratio is one of the important mechanical parameters of rock materials, which determines the ratio between the lateral strain and the longitudinal strain of the rock during the stress process. In this embodiment, different Poisson's ratio ranges are set for different types of rocks, from 0.1 to 0.3, to simulate the deformation characteristics of rocks under different stress states.
[0125] Elastic modulus: This is a physical quantity that represents the stiffness of a material; it measures the material's resistance to deformation under stress. The larger the elastic modulus, the smaller the deformation of the material under the same stress, and vice versa. For rocks, the elastic modulus reflects the rock's rigidity and compressive strength; the elastic modulus of rocks ranges from 1.5 × 10⁻⁶ GPa to 8.5 × 10⁻⁶ GPa.
[0126] Confining pressure magnitude: Confining pressure refers to the external constraint pressure exerted on rock during the fracturing process. In actual engineering, the confining pressure varies depending on the depth of the rock. This embodiment sets three different confining pressure values: 5 MPa, 10 MPa, and 15 MPa.
[0127] After determining the experimental parameters, a composite impact rock-breaking experiment was conducted. In each experiment, one or more sets of parameters were changed, and the corresponding rock damage morphology was recorded. The specific rock damage morphology was acquired using a three-dimensional scanning device, including: penetration depth, damage area, and fracture volume.
[0128] The results of each experiment, including all variable parameters and their corresponding experimental results, are saved as a complete sample dataset. Through multiple experiments, a large-scale dataset covering multiple continuously variable parameter combinations is established.
[0129] Based on the same concept, referencing Figure 2 This application also proposes an evaluation system for a composite impact multi-tooth rock-breaking test device, comprising:
[0130] The parameter selection module is used to select and determine the main controlling factors that cause differences in rock damage state as variable parameters through composite impact tests.
[0131] The dataset creation module, based on the composite impact test, obtains the test results of rock damage and the corresponding variable parameters as sample data, changes the variable parameters to conduct the next set of composite impact tests, until all sample data under different parameters are obtained, and then establishes the dataset.
[0132] The model building module establishes the neural network structure and activation function based on the dataset and trains the neural network. The neural network is used to learn from the database and predict the damage morphology of rocks after impact tests.
[0133] The model computation module takes specific variable parameters to be tested as input to the neural network model, outputs the corresponding experimental prediction results, and completes the experimental evaluation.
[0134] Regarding the system in the above embodiments, the specific methods and technical effects of each part performing operations have been described in detail in the embodiments related to the method, and will not be elaborated here.
[0135] In summary, the evaluation method for composite impact multi-tooth rock breaking tests of this invention first identifies nine key factors affecting the test before constructing the neural network model to ensure the comprehensiveness of the input parameters. A dataset of rock breaking effects is then established through rock breaking tests. Next, an integrated learning approach is used to fuse BP neural networks, RF models, and XGBoost models to construct a neural network model. This leverages the advantages of each model to improve prediction accuracy. Based on understanding the key factors, it enables the processing of complex nonlinear relational data, maintaining efficient and accurate predictions even in complex geological environments. This provides a valuable reference for optimizing engineering and tool parameters during composite drilling in deep hard rock formations, expanding the application scope of the method and system of this application.
[0136] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An evaluation method for composite impact multi-tooth rock-breaking tests, characterized in that, Includes the following steps: Through composite impact tests, the main controlling factors that cause differences in rock damage state were selected and determined as variable parameters; Based on the composite impact test, the test results of rock damage and the corresponding variable parameters are obtained as sample data. The variable parameters are changed to conduct the next set of composite impact tests until all sample data under different parameters are obtained, and a dataset is established. Based on the dataset, establish the neural network structure and activation function, and train the neural network to form a neural network model; Input specific variable parameters to be tested into the neural network model, output the corresponding test prediction results, and complete the test evaluation.
2. The evaluation method for the composite impact multi-tooth rock-breaking test according to claim 1, characterized in that, The variable parameters include: sleeve impact speed, sleeve rotation speed, drill bit shape, drill bit inclination angle, drill bit diameter, rock type, formation Poisson's ratio, elastic modulus, and confining pressure.
3. The evaluation method for the composite impact multi-tooth rock-breaking test according to claim 2, characterized in that, The test results include: penetration depth, damage area, and fragmentation volume.
4. The evaluation method for the composite impact multi-tooth rock-breaking test according to claim 1, characterized in that, In the process of establishing the neural network structure based on the dataset, the dataset is divided into training set, validation set, and test set, and the ratio of training set:test set:validation set is 70-80%:10-15%:10-15%. The training set is used to train the neural network model; The validation set is used to adjust and optimize model parameters; The test set is used to test the model's generalization ability.
5. The evaluation method for the composite impact multi-tooth rock-breaking test according to claim 4, characterized in that, In the process of establishing the neural network structure based on the dataset; The neural network structure includes an input layer, an output layer, and hidden layers. The number of neurons in the input and output layers is consistent with the variable parameters and the number of experimental results. The number of hidden layers is at least one.
6. The evaluation method for the composite impact multi-tooth rock-breaking test according to claim 5, characterized in that, The activation function is a non-saturating activation function.
7. The evaluation method for the composite impact multi-tooth rock-breaking test according to claim 6, characterized in that, The neural network model includes a first-layer ensemble learning framework and a second-layer ensemble learning framework. The first-layer ensemble learning framework includes a BP neural network, an RF model, and an XGBoost model. The input parameters of the first-layer ensemble learning framework are variable parameters, and the output parameters are the calculation results of the BP neural network, the calculation results of the RF model, and the calculation results of the XGBoost model. The second-layer ensemble learning framework is used to fuse the computation results of the BP neural network, the RF model, and the XGBoost model to output the experimental prediction results.
8. An evaluation system for a composite impact multi-tooth rock-breaking test device, characterized in that, include: The parameter selection module is used to select and determine the main controlling factors that cause differences in rock damage state as variable parameters through composite impact tests. The dataset creation module, based on the composite impact test, obtains the test results of rock damage and the corresponding variable parameters as sample data, changes the variable parameters to conduct the next set of composite impact tests, until all sample data under different parameters are obtained, and then establishes the dataset. The model building module establishes the neural network structure and activation function based on the dataset and trains the neural network. The neural network is used to learn from the database and predict the damage morphology of rocks after impact tests. The model computation module takes specific variable parameters to be tested as input to the neural network model, outputs the corresponding experimental prediction results, and completes the experimental evaluation.
9. The evaluation system for the composite impact multi-tooth rock-breaking test device according to claim 8, characterized in that, The variable parameters include: sleeve impact speed, sleeve rotation speed, drill tooth shape, drill tooth inclination angle, drill tooth diameter, rock type, formation Poisson's ratio, elastic modulus, and confining pressure. The test results include: penetration depth, damage area, and fragmentation volume.
10. The evaluation system for the composite impact multi-tooth rock-breaking test device according to claim 9, characterized in that, The neural network model includes a first-layer ensemble learning framework and a second-layer ensemble learning framework. The first-layer ensemble learning framework includes a BP neural network, an RF model, and an XGBoost model. The input parameters of the first-layer ensemble learning framework are variable parameters, and the output parameters are the calculation results of the BP neural network, the calculation results of the RF model, and the calculation results of the XGBoost model. The second-layer ensemble learning framework is used to fuse the computation results of the BP neural network, the RF model, and the XGBoost model to output the experimental prediction results.