Rock mass grade intelligent classification method and system based on while-drilling multi-physics field information
By acquiring and processing multi-physics field information during drilling, and using a neural network model for intelligent rock mass classification, the problem of low classification accuracy in deep rock engineering has been solved, and real-time and accurate rock mass classification has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA HYDROELECTRIC ENGINEERING CONSULTING GROUP CHENGDU RESEARCH HYDROELECTRIC INVESTIGATION DESIGN AND INSTITUTE
- Filing Date
- 2026-01-27
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies have low accuracy in rock mass classification in deep rock engineering, and cannot meet the requirements for real-time and accurate classification. Traditional methods also suffer from problems such as long cycle time, high cost, and strong subjectivity.
By acquiring multi-physics field information during drilling, including mechanical field, vibration field, seepage field and temperature field information, preprocessing and feature extraction are performed, and a neural network model with one-dimensional convolutional neural network, long short-term memory network and attention mechanism is used for intelligent classification to achieve real-time identification of rock mass grade.
It improves the accuracy and real-time performance of deep rock mass classification, enabling accurate identification of rock mass grades during drilling, providing a reliable basis for engineering design, and reducing human intervention and subjective errors.
Smart Images
Figure CN122045927A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rock mass classification technology, and in particular to a method and system for intelligent classification of rock mass grades based on multi-physics field information during drilling. Background Technology
[0002] With the increasing number of deep rock mass engineering projects, the maximum burial depth has exceeded 2,000 meters. Deep rock masses are in a special environment of "three highs and one disturbance" (high ground stress, high ground temperature, high water pressure and excavation disturbance). Their rock mass mechanical properties (such as compressive strength and integrity) are significantly different from those of shallow rock masses, and they are prone to disasters such as rock bursts, water inrushes, and large deformations. Therefore, accurate and real-time classification of deep rock mass engineering levels while drilling has become a core prerequisite for ensuring the rationality of engineering design, construction safety and operational stability.
[0003] The core objective of rock mass engineering classification is to quantify key rock mass parameters (such as strength, integrity, and stability) to classify rock mass categories with different engineering adaptability (such as categories I-V in the BQ classification and excellent-poor in the RMR classification), providing a basis for excavation scheme optimization, support structure selection, and disaster risk assessment.
[0004] The BQ grading method (GB / T 50218-2014, "Engineering Rock Mass Grading Standard") relies on laboratory tests to obtain the saturated uniaxial compressive strength (Rc) and rock mass integrity coefficient (Kv) of the rock mass. The grading index is calculated using the formula BQ=100+3Rc+250Kv. This method requires a large number of core samples (the core sampling rate at depth is often less than 50%), and the laboratory test cycle is as long as 3-7 days, resulting in serious lag and failing to meet the needs of real-time decision-making while drilling.
[0005] RMR grading method (Bieniawski, 1973): It scores based on six parameters including core quality index (RQD), joint spacing, and joint condition. It relies on engineers' on-site observation and subjective judgment, which has large human error and is difficult to accurately score under deep and complex geological conditions (such as hidden joints and broken rock mass).
[0006] Traditional classification methods mainly rely on manual geological logging, borehole coring, and laboratory testing, which have drawbacks such as long cycles, high costs, and strong subjectivity, making them difficult to meet the rapid exploration needs of modern deep rock engineering. To overcome the limitations of traditional methods, Measurement While Drilling (MWD) technology has become a hot topic in the industry. MWD technology uses sensors deployed on the drilling rig to collect physical parameters (such as oil pressure, drill pressure, torque, and rotational speed) in real time during the drilling process, obtaining information about the rock mass inside the borehole without stopping drilling for sampling, thus achieving "drilling and judging simultaneously."
[0007] For example, the invention patent application CN116879527A, entitled "Method and System for Real-Time Determination of Basic Rock Mass Quality Classification Based on Drilling Parameters," focuses on the mechanical field. Based on rock fracture mechanics and the law of conservation of energy, it derives "key rock-breaking indicators" (impact energy, rotational energy, and drilling rate consumed per unit volume of rock fracture). Rock mass classification is achieved by establishing a predictive model (such as a BP neural network) between these key indicators and basic rock mass quality indicators (saturated uniaxial compressive strength Rc, integrity coefficient Kv). This patent is the first to correlate mechanical drilling parameters with basic rock mass quality indicators (BQ), explicitly considering the identification of rock masses with different degrees of fracture (from intact to extremely fractured). The physical meaning is clear, and the model has a certain degree of interpretability.
[0008] This scheme primarily relies on mechanical field parameters (drilling pressure, torque, and rotational speed), essentially making it a "strong mechanical model." It severely lacks direct and effective means of sensing rock mass structure (such as fractures) and permeability. Key physical field information such as vibration, seepage, and temperature is not incorporated into the system. The BP neural network used is a relatively traditional shallow model with weak feature extraction and sequence modeling capabilities, making it difficult to capture the complex spatiotemporal patterns and multi-field coupling effects during drilling.
[0009] In summary, existing methods mostly rely on empirical formulas or simple statistical models, which are insufficient in describing the nonlinear characteristics of complex deep rock masses, and their universality and robustness are difficult to guarantee. They lack effective automatic feature extraction capabilities, cannot achieve end-to-end intelligent identification, involve many manual interventions, and are highly subjective. At the same time, data processing and interpretation are lagging behind, which cannot meet the real-time decision-making needs of the construction process, and the early warning capabilities are limited, making it difficult to meet the "real-time, accurate, and engineering-oriented" classification requirements of deep rock mass engineering. Summary of the Invention
[0010] The technical problem solved by this invention is to provide a method and system for intelligent classification of rock mass grades based on multi-physics field information during drilling, thereby solving the problem of low accuracy in existing deep rock mass classification.
[0011] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a rock mass grade intelligent classification method based on drilling multi-physics field information, comprising the following steps: S1. Obtain drilling multiphysics information, which includes mechanical field information, vibration field information, seepage field information and temperature field information; S2. Preprocess the acquired multiphysics information obtained during drilling, including data cleaning, noise reduction and depth alignment. S3. Extract the features corresponding to the preprocessed physical field information. The features corresponding to the physical field information include mechanical field features, vibration field features, seepage field features, and temperature field features. The mechanical field features include mechanical field time-domain features and mechanical field frequency-domain features. The vibration field features include vibration field time-domain features, vibration field frequency-domain features, and vibration field time-frequency-domain features. S4. Using the features corresponding to the physical field information as input data and the rock mass grade corresponding to the depth as output, train the neural network model to obtain an intelligent classification model. S5. Collect multi-physics field information while drilling, preprocess it, extract the features corresponding to the preprocessed physical field information, classify it using an intelligent classification model, and obtain the rock mass grade that varies with depth.
[0012] Furthermore, the mechanical field information includes the counter-torque generated when the drill bit breaks the rock, the axial pressure applied to the drill bit, and the rotational speed of the drill pipe; the vibration field information includes axial, radial, and tangential vibration accelerations during drilling; the seepage field information includes mud pressure or gas flow rate; and the temperature field information includes drill bit temperature.
[0013] Furthermore, the counter-torque generated when the drill bit breaks rock is obtained by a torque sensor installed on the drill pipe or power head; the axial pressure applied to the drill bit is obtained by a feed pressure sensor or feed pressure sensor installed on the feed cylinder; the rotational speed of the drill pipe is obtained by a speed sensor; the axial, radial, and tangential vibration accelerations are obtained by a triaxial accelerometer; the mud pressure is obtained by a mud pressure sensor installed at the orifice or bottom of the mud circulation system; the gas flow rate is obtained by a gas flow sensor installed in the air circulation system; and the drill bit temperature is obtained by a high-temperature resistant temperature sensor installed inside the drill bit or on the drill pipe.
[0014] Furthermore, the time-domain characteristics of the mechanical field include mean, variance, root mean square, peak value, kurtosis, and waveform factor; the frequency-domain characteristics of the mechanical field include dominant frequency, frequency centroid, and frequency variance; the time-domain characteristics of the vibration field include peak value, root mean square, impulse factor, and margin factor; the frequency-domain characteristics of the vibration field include one-third octave band spectrum energy and spectral peak value; the time-frequency domain characteristics of the vibration field include time and frequency information; the seepage field characteristics include the mean, variance, and fluctuation amplitude of mud pressure or gas flow rate; and the temperature field characteristics include the mean and rate of change of temperature.
[0015] Furthermore, the neural network model includes a one-dimensional convolutional neural network, a long short-term memory network, and an attention mechanism. The one-dimensional convolutional neural network is used to extract features from the input data, the long short-term memory network is used to find sequence dependencies between rock masses at different depths, and the attention mechanism is used to adaptively focus on depth points for the classification task.
[0016] This invention also provides an intelligent rock mass classification system based on multi-physics field information during drilling, realizing the intelligent rock mass classification method based on multi-physics field information during drilling as described above. The system includes a data acquisition module, a data preprocessing module, a feature extraction module, and an intelligent classification module. The data acquisition module is used to acquire multi-physics field information during drilling, which includes mechanical field information, vibration field information, seepage field information, and temperature field information. The data preprocessing module is used to preprocess the acquired multi-physics field information during drilling, including data cleaning, noise reduction, and depth alignment. The feature extraction module is used to extract features corresponding to the preprocessed physical field information, which include mechanical field features, vibration field features, seepage field features, and temperature field features. The mechanical field features include mechanical field time-domain features and mechanical field frequency-domain features. The vibration field features include vibration field time-domain features, vibration field frequency-domain features, and vibration field time-frequency-domain features. The intelligent classification module stores an intelligent classification model, which takes the features corresponding to the physical field information as input data and outputs the rock mass grade that varies with depth.
[0017] Furthermore, the mechanical field information includes the counter-torque generated when the drill bit breaks the rock, the axial pressure applied to the drill bit, and the rotational speed of the drill pipe; the vibration field information includes axial, radial, and tangential vibration accelerations during drilling; the seepage field information includes mud pressure or gas flow rate; and the temperature field information includes drill bit temperature.
[0018] Furthermore, the counter-torque generated when the drill bit breaks rock is obtained by a torque sensor installed on the drill pipe or power head; the axial pressure applied to the drill bit is obtained by a feed pressure sensor or feed pressure sensor installed on the feed cylinder; the rotational speed of the drill pipe is obtained by a speed sensor; the axial, radial, and tangential vibration accelerations are obtained by a triaxial accelerometer; the mud pressure is obtained by a mud pressure sensor installed at the orifice or bottom of the mud circulation system; the gas flow rate is obtained by a gas flow sensor installed in the air circulation system; and the drill bit temperature is obtained by a high-temperature resistant temperature sensor installed inside the drill bit or on the drill pipe.
[0019] Furthermore, the time-domain characteristics of the mechanical field include mean, variance, root mean square, peak value, kurtosis, and waveform factor; the frequency-domain characteristics of the mechanical field include dominant frequency, frequency centroid, and frequency variance; the time-domain characteristics of the vibration field include peak value, root mean square, impulse factor, and margin factor; the frequency-domain characteristics of the vibration field include one-third octave band spectrum energy and spectral peak value; the time-frequency domain characteristics of the vibration field include time and frequency information; the seepage field characteristics include the mean, variance, and fluctuation amplitude of mud pressure or gas flow rate; and the temperature field characteristics include the mean and rate of change of temperature.
[0020] Furthermore, the neural network model includes a one-dimensional convolutional neural network, a long short-term memory network, and an attention mechanism. The one-dimensional convolutional neural network is used to extract features from the input data, the long short-term memory network is used to find sequence dependencies between rock masses at different depths, and the attention mechanism is used to adaptively focus on depth points for the classification task.
[0021] The beneficial effects of this invention are as follows: This invention provides an intelligent classification method and system for rock mass grades based on drilling multi-physics field information. By acquiring drilling multi-physics field information, including mechanical field information, vibration field information, seepage field information, and temperature field information, the acquired drilling multi-physics field information is preprocessed, including data cleaning, noise reduction, and depth alignment. Features corresponding to the preprocessed physical field information are extracted. The intelligent classification model uses the features corresponding to the physical field information as input data and outputs the rock mass grade that varies with depth, thus solving the problem of low accuracy in existing deep rock mass classification. Attached Figure Description
[0022] Figure 1 This is a flowchart illustrating an intelligent rock mass classification method based on drilling multi-physics information provided by the present invention. Figure 2 This is a schematic diagram of the structure of an intelligent rock mass classification system based on drilling multi-physics field information provided by the present invention. Detailed Implementation
[0023] This invention addresses the problem of low accuracy in existing deep rock mass classification methods by providing an intelligent rock mass classification method based on multi-physics field information during drilling, such as... Figure 1 As shown, it includes the following steps: S1. Obtain drilling multiphysics information, which includes mechanical field information, vibration field information, seepage field information and temperature field information.
[0024] Specifically, the mechanical field information includes the counter-torque generated when the drill bit breaks the rock, the axial pressure applied to the drill bit, and the rotational speed of the drill pipe; the vibration field information includes axial, radial, and tangential vibration accelerations during drilling, which is key to reflecting the integrity and heterogeneity of the rock mass; the seepage field information includes mud pressure or gas flow rate, which reflects the permeability and water-bearing fractures of the rock mass; and the temperature field information includes the drill bit temperature, which reflects the degree of frictional heat generation and the thermal characteristics of the geological body.
[0025] The counter-torque generated when the drill bit breaks rock is obtained by a torque sensor installed on the drill pipe or power head; the axial pressure applied to the drill bit is obtained by a feed pressure sensor or feed pressure sensor installed on the feed cylinder; the rotational speed of the drill pipe is obtained by a speed sensor; the axial, radial, and tangential vibration accelerations are obtained by a triaxial accelerometer; the mud pressure is obtained by a mud pressure sensor installed at the orifice or bottom of the mud circulation system; the gas flow rate is obtained by a gas flow sensor installed in the air circulation system; and the drill bit temperature is obtained by a high-temperature sensor installed inside the drill bit or on the drill pipe.
[0026] S2. Preprocess the acquired multiphysics information obtained during drilling. The preprocessing includes data cleaning, noise reduction, and depth alignment.
[0027] Specifically, data cleaning includes handling issues such as missing data values and anomalous jumps, such as momentary sensor malfunctions.
[0028] Noise reduction includes using wavelet thresholding or empirical mode decomposition (EMD) on vibration field information to effectively separate useful vibration components related to rock mass fracture and suppress mechanical vibration noise; for mechanical field information and seepage field information, low-pass digital filters (such as Butterworth filters) are used to smooth high-frequency electrical noise.
[0029] Depth alignment: The multiphysics information while drilling is precisely aligned with the drilling depth (obtained by drill pipe counting or encoder) to ensure that various types of data at the same depth point are synchronized in time. Then, the continuous multiphysics information data stream while drilling is sliced according to a fixed depth window to obtain different data windows. Each data window is used as an independent analysis sample. The depth window can be 10 cm or 20 cm, etc.
[0030] S3. Extract the features corresponding to the preprocessed physical field information. The features corresponding to the physical field information include mechanical field features, vibration field features, seepage field features, and temperature field features. The mechanical field features include mechanical field time-domain features and mechanical field frequency-domain features. The vibration field features include vibration field time-domain features, vibration field frequency-domain features, and vibration field time-frequency-domain features.
[0031] Specifically, for the same data window, corresponding features are extracted to obtain the features corresponding to the physical field information at different depths. The time-domain features of the mechanical field include mean, variance, root mean square, peak value, kurtosis, and waveform factor. For example, high kurtosis indicates a large number of impact pulses in the torque, possibly corresponding to hard, fractured rock masses. The frequency-domain features of the mechanical field include dominant frequency, frequency centroid, and frequency variance, reflecting the periodic fluctuation characteristics of drilling. The time-domain features of the vibration field include peak value, root mean square, impulse factor, and margin factor, reflecting the overall energy and impact intensity of the vibration. The frequency-domain features of the vibration field include one-third octave band spectral energy and spectral peak value, reflecting different rock mass structures. The time-frequency features of the vibration field include time... The frequency information is extremely effective in capturing transient events such as crossing fracture surfaces and weak interlayers; the seepage field characteristics include the mean, variance, and fluctuation amplitude of mud pressure or gas flow rate, which can identify sudden pressure drops (which may indicate fracture leakage) or sudden pressure increases (which may indicate high-pressure aquifers); the temperature field characteristics include the mean and rate of change of temperature, which can identify abnormal temperature rises caused by frictional heating or temperature changes when crossing rock layers with different thermal conductivity. In this way, the properties of rock masses can be accurately reflected by mechanical field characteristics, vibration field characteristics, seepage field characteristics, and temperature field characteristics, thereby improving the accuracy of rock mass classification.
[0032] S4. Using the features corresponding to the physical field information as input data and the rock mass grade at the corresponding depth as output, train the neural network model to obtain an intelligent classification model.
[0033] Specifically, the neural network model includes a one-dimensional convolutional neural network (1D-CNN), a long short-term memory network (LSTM), and an attention mechanism. The 1D-CNN is used to extract features from the input data, specifically, it is responsible for automatically extracting local sensitive features from the physical field information (especially vibration waveforms, torque sequences, etc.) of each depth window. The convolutional kernel can slide across the physical field information of the depth window of the sequence, effectively capturing features such as impact pulses and periodic fluctuations. The LSTM network is used to find sequence dependencies between rock masses at different depths, that is, there is a geological correlation between the rock mass grade at the current depth and the sequence of rock masses drilled above it. The attention mechanism is used to adaptively focus on the depth points of the classification task, thereby improving the model's performance and interpretability.
[0034] During training, the rock mass grades corresponding to the depth are obtained through existing reliable methods, such as the existing BQ grading or RMR grading, to form a correspondence between the features corresponding to the physical field information and the rock mass grades that are accurately matched according to the depth. These are used as paired samples to form a training set, which is used to train the neural network model. This results in an intelligent classification model that takes the features corresponding to the physical field information as input data and outputs the rock mass grades corresponding to different depths of the physical field information.
[0035] S5. Collect multi-physics field information while drilling, preprocess it, extract the features corresponding to the preprocessed physical field information, classify it using an intelligent classification model, and obtain the rock mass grade that varies with depth.
[0036] Specifically, the rock mass grade that varies with depth can be visualized by plotting a rock mass grade variation curve with depth. At the same time, based on the classification results, a color-coded comprehensive geological profile model can be automatically generated, clearly identifying the rock mass grade at different depths with different colors, providing engineers with the most direct decision support.
[0037] This invention also provides an intelligent rock mass classification system based on multi-physics information during drilling, realizing the intelligent rock mass classification method based on multi-physics information during drilling as described above. The system, as... Figure 2 As shown, the system includes a data acquisition module, a data preprocessing module, a feature extraction module, and an intelligent classification module. The data acquisition module acquires multi-physics information while drilling, including mechanical field information, vibration field information, seepage field information, and temperature field information. The data preprocessing module preprocesses the acquired multi-physics information while drilling, including data cleaning, noise reduction, and depth alignment. The feature extraction module extracts features corresponding to the preprocessed physical field information, including mechanical field features, vibration field features, seepage field features, and temperature field features. The mechanical field features include time-domain and frequency-domain features; the vibration field features include time-domain, frequency-domain, and time-frequency-domain features. The intelligent classification module stores an intelligent classification model, which takes the features corresponding to the physical field information as input data and outputs the rock mass grade that varies with depth.
[0038] Furthermore, the mechanical field information includes the counter-torque generated when the drill bit breaks the rock, the axial pressure applied to the drill bit, and the rotational speed of the drill pipe; the vibration field information includes axial, radial, and tangential vibration accelerations during drilling; the seepage field information includes mud pressure or gas flow rate; and the temperature field information includes drill bit temperature.
[0039] Furthermore, the counter-torque generated when the drill bit breaks rock is obtained by a torque sensor installed on the drill pipe or power head; the axial pressure applied to the drill bit is obtained by a feed pressure sensor or feed pressure sensor installed on the feed cylinder; the rotational speed of the drill pipe is obtained by a speed sensor; the axial, radial, and tangential vibration accelerations are obtained by a triaxial accelerometer; the mud pressure is obtained by a mud pressure sensor installed at the orifice or bottom of the mud circulation system; the gas flow rate is obtained by a gas flow sensor installed in the air circulation system; and the drill bit temperature is obtained by a high-temperature resistant temperature sensor installed inside the drill bit or on the drill pipe.
[0040] Furthermore, the time-domain characteristics of the mechanical field include mean, variance, root mean square, peak value, kurtosis, and waveform factor; the frequency-domain characteristics of the mechanical field include dominant frequency, frequency centroid, and frequency variance; the time-domain characteristics of the vibration field include peak value, root mean square, impulse factor, and margin factor; the frequency-domain characteristics of the vibration field include one-third octave band spectrum energy and spectral peak value; the time-frequency domain characteristics of the vibration field include time and frequency information; the seepage field characteristics include the mean, variance, and fluctuation amplitude of mud pressure or gas flow rate; and the temperature field characteristics include the mean and rate of change of temperature.
[0041] Furthermore, the neural network model includes a one-dimensional convolutional neural network, a long short-term memory network, and an attention mechanism. The one-dimensional convolutional neural network is used to extract features from the input data, the long short-term memory network is used to find sequence dependencies between rock masses at different depths, and the attention mechanism is used to adaptively focus on depth points for the classification task.
Claims
1. A rock mass grade intelligent classification method based on drilling multi-physical field information, characterized in that, The method includes the following steps: S1. Obtain drilling multiphysics information, which includes mechanical field information, vibration field information, seepage field information and temperature field information; S2. Preprocess the acquired multiphysics information obtained during drilling, including data cleaning, noise reduction and depth alignment. S3. Extract the features corresponding to the preprocessed physical field information. The features corresponding to the physical field information include mechanical field features, vibration field features, seepage field features, and temperature field features. The mechanical field features include mechanical field time-domain features and mechanical field frequency-domain features. The vibration field features include vibration field time-domain features, vibration field frequency-domain features, and vibration field time-frequency-domain features. S4. Using the features corresponding to the physical field information as input data and the rock mass grade corresponding to the depth as output, train the neural network model to obtain an intelligent classification model. S5. Collect multi-physics field information while drilling, preprocess it, extract the features corresponding to the preprocessed physical field information, classify it using an intelligent classification model, and obtain the rock mass grade that varies with depth.
2. The rock mass grade intelligent classification method based on drilling multi-physical field information according to claim 1, characterized in that, The mechanical field information includes the counter-torque generated when the drill bit breaks the rock, the axial pressure applied to the drill bit, and the rotational speed of the drill pipe; the vibration field information includes axial, radial, and tangential vibration accelerations during drilling; the seepage field information includes mud pressure or gas flow rate; and the temperature field information includes drill bit temperature.
3. The intelligent rock mass classification method based on drilling multi-physics information according to claim 2, characterized in that, The counter-torque generated when the drill bit breaks rock is obtained by a torque sensor installed on the drill pipe or power head; the axial pressure applied to the drill bit is obtained by a feed pressure sensor or feed pressure sensor installed on the feed cylinder; the rotational speed of the drill pipe is obtained by a speed sensor; the axial, radial, and tangential vibration accelerations are obtained by a triaxial accelerometer; the mud pressure is obtained by a mud pressure sensor installed at the orifice or bottom of the mud circulation system; the gas flow rate is obtained by a gas flow sensor installed in the air circulation system; and the drill bit temperature is obtained by a high-temperature sensor installed inside the drill bit or on the drill pipe.
4. The intelligent rock mass classification method based on multi-physics information during drilling as described in claim 1, characterized in that, The time-domain characteristics of the mechanical field include mean, variance, root mean square, peak value, kurtosis, and waveform factor; the frequency-domain characteristics of the mechanical field include dominant frequency, frequency centroid, and frequency variance; the time-domain characteristics of the vibration field include peak value, root mean square, impulse factor, and margin factor; the frequency-domain characteristics of the vibration field include one-third octave band spectral energy and spectral peak value; the time-frequency domain characteristics of the vibration field include time and frequency information; the seepage field characteristics include mean, variance, and fluctuation amplitude of mud pressure or gas flow rate; and the temperature field characteristics include mean and rate of change of temperature.
5. The intelligent rock mass classification method based on multi-physics information during drilling as described in claim 1, characterized in that, The neural network model includes a one-dimensional convolutional neural network, a long short-term memory network, and an attention mechanism. The one-dimensional convolutional neural network is used to extract features from the input data, the long short-term memory network is used to find sequence dependencies between rock masses at different depths, and the attention mechanism is used to adaptively focus on depth points for the classification task.
6. A rock mass classification system based on multi-physics field information during drilling, characterized in that, To implement the method as described in claim 1, the system includes a data acquisition module, a data preprocessing module, a feature extraction module, and an intelligent classification module; the data acquisition module is used to acquire multi-physics information while drilling, which includes mechanical field information, vibration field information, seepage field information, and temperature field information; the data preprocessing module is used to preprocess the acquired multi-physics information while drilling, which includes data cleaning, noise reduction, and depth alignment. The feature extraction module is used to extract features corresponding to the preprocessed physical field information. The features corresponding to the physical field information include mechanical field features, vibration field features, seepage field features, and temperature field features. The mechanical field features include mechanical field time-domain features and mechanical field frequency-domain features. The vibration field features include vibration field time-domain features, vibration field frequency-domain features, and vibration field time-frequency-domain features. The intelligent classification module stores an intelligent classification model. The intelligent classification model takes the features corresponding to the physical field information as input data and outputs the rock mass grade that varies with depth.
7. The intelligent rock mass classification system based on multi-physics information during drilling as described in claim 6, characterized in that, The mechanical field information includes the counter-torque generated when the drill bit breaks the rock, the axial pressure applied to the drill bit, and the rotational speed of the drill pipe; the vibration field information includes axial, radial, and tangential vibration accelerations during drilling; the seepage field information includes mud pressure or gas flow rate; and the temperature field information includes drill bit temperature.
8. The intelligent rock mass classification system based on multi-physics information during drilling according to claim 7, characterized in that, The counter-torque generated when the drill bit breaks rock is obtained by a torque sensor installed on the drill pipe or power head; the axial pressure applied to the drill bit is obtained by a feed pressure sensor or feed pressure sensor installed on the feed cylinder; the rotational speed of the drill pipe is obtained by a speed sensor; the axial, radial, and tangential vibration accelerations are obtained by a triaxial accelerometer; the mud pressure is obtained by a mud pressure sensor installed at the orifice or bottom of the mud circulation system; the gas flow rate is obtained by a gas flow sensor installed in the air circulation system; and the drill bit temperature is obtained by a high-temperature sensor installed inside the drill bit or on the drill pipe.
9. The intelligent rock mass classification system based on multi-physics information during drilling according to claim 6, characterized in that, The time-domain characteristics of the mechanical field include mean, variance, root mean square, peak value, kurtosis, and waveform factor; the frequency-domain characteristics of the mechanical field include dominant frequency, frequency centroid, and frequency variance; the time-domain characteristics of the vibration field include peak value, root mean square, impulse factor, and margin factor; the frequency-domain characteristics of the vibration field include one-third octave band spectral energy and spectral peak value; the time-frequency domain characteristics of the vibration field include time and frequency information; the seepage field characteristics include mean, variance, and fluctuation amplitude of mud pressure or gas flow rate; and the temperature field characteristics include mean and rate of change of temperature.
10. The intelligent rock mass classification system based on multi-physics information during drilling according to claim 6, characterized in that, The neural network model includes a one-dimensional convolutional neural network, a long short-term memory network, and an attention mechanism. The one-dimensional convolutional neural network is used to extract features from the input data, the long short-term memory network is used to find sequence dependencies between rock masses at different depths, and the attention mechanism is used to adaptively focus on depth points for the classification task.