Rock stratum drilling in-situ identification method based on physical information machine learning and application thereof

Abstract

The invention discloses a rock stratum drilling in-situ identification method based on physical information machine learning and an application thereof, which are used for solving the technical problem of in-situ sensing of unknown rock mass properties in a drilling process. Multi-source signal parameters including drilling pressure, drilling rod torque, drilling rod rotating speed, drilling speed and drilling vibration and acoustic emission signals are collected in the drilling process, CEEMDAN-VMD combined denoising and feature extraction are carried out, and fused feature vectors are input into a physical information machine learning model; and outputting the strength prediction value of the drilled rock stratum and the prediction value of the rock mass category on line in real time. The method is applied to innovative combination of freeze-thaw environment simulation, rock drilling process test and lithology intelligent identification, an advanced means is provided for drillability evaluation of freeze-thaw rock mass, data support can be provided for blasting parameter intelligent design and slope stability early warning, and the method has good engineering applicability and popularization value.

Description

TECHNICAL FIELD

[0001] The present application relates to a rock stratum drilling in-situ identification method based on physical information machine learning and its application, belonging to the cross technical field of rock mechanics, geological engineering and intelligent mining. BACKGROUND

[0002] In the process of drilling and blasting excavation in open-pit mines in cold regions, the traditional rock mass property evaluation method relies on post-drilling coring and laboratory testing, which is cumbersome and time-consuming, and cannot realize in-situ identification of rock properties during drilling, resulting in lagging design of blasting parameters and insufficient basis for slope stability judgment, which seriously affects mining efficiency and operation safety.

[0003] The use of sensor technology can achieve in-situ testing of rock mass strength during drilling. For example, Chinese patent application No. CN119321965A discloses a borehole rock mass in-situ testing device and its testing method. It tests the rock mass at multiple points around the borehole at a certain position in the borehole, evaluates the differences in stress state of the rock mass around the borehole through the differences in double-probe penetration characteristics at multiple angles, and then identifies the mechanical properties of the rock mass under different stress states. The acquisition of mechanical properties relies on traditional mathematical calculations, which often produce large errors due to poor measurement accuracy or sudden noise changes. Moreover, it is difficult to obtain factual image data during drilling due to the environmental characteristics of the borehole during drilling, so it is also difficult to identify different lithology formations in the drilling direction.

[0004] In particular, in the drilling of frozen strata in cold regions, the lithology of the strata structure in cold regions is often different from that of conventional strata due to repeated freezing and thawing. Therefore, the identification and drilling of frozen strata structure in cold regions are key and difficult technologies in geological mining. There are some low-temperature drilling devices or rock mass testing systems in the prior art, but they lack environmental simulation capability and the testing process is separated from drilling, making it difficult to meet the needs of precise and rapid sensing and self-adaptive control of deteriorated rock mass in freezing and thawing.

[0005] Therefore, there is an urgent need for a comprehensive and precise testing technology that can integrate cold region stratum freezing and thawing environment simulation, multi-parameter acquisition during drilling, and lithology intelligent identification function to provide technical support for safe and efficient mining in cold region mining. SUMMARY

[0006] The technical problem solved by the present application is to provide a rock stratum drilling in-situ identification method based on physical information machine learning and its application, which can realize in-situ stratum identification during drilling in geological mining.

[0007] The present application adopts the following technical solutions:

[0008] This invention first discloses an in-situ rock strata identification method based on physical information machine learning, comprising the following steps:

[0009] Step S1: During the drilling process, collect multi-source signal parameters including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals;

[0010] Step S2: Perform two mode decompositions on the drilling vibration and acoustic emission signals using CEEMDAN and VMD to obtain dynamic signal characteristics related to rock mass fracture.

[0011] Step S3: After standardizing the remaining multi-source signal parameters, combine them with the dynamic signal features in S2 to form a fused feature vector;

[0012] Step S4: Input the fused feature vector from step S3 into the pre-trained physical information machine learning model. The physical information machine learning model uses an LSTM network as its backbone and constructs a physical inversion equation between drilling pressure, drill rod torque, drill rod rotation speed, drilling speed and rock strength based on the Businesk elastic half-space theory to perform physical constraints. It outputs the predicted strength value of the drilled rock layer and the rock mass category.

[0013] In the in-situ rock strata identification method based on physical information machine learning of the present invention, specifically, step S2 includes the following sub-steps:

[0014] Sub-step S21: Perform preliminary decomposition of drilling vibration and acoustic emission signals using CEEMDAN to obtain a series of intrinsic mode functions:

[0015] ;

[0016] in, The original drilling vibration and acoustic emission signals, For intrinsic mode functions, K1 is the residual, and K1 is the total number of intrinsic mode functions obtained from the CEEMDAN decomposition.

[0017] Sub-step S22: Perform secondary decomposition of the key IMF components with high energy in the drilling vibration and acoustic emission signals using variational mode decomposition (VMD) to accurately separate the characteristic frequency bands related to the fracturing of the freeze-thaw rock mass.

[0018] ;

[0019] in, The key IMF variables selected from the intrinsic mode functions in step S21 are those with a large proportion of IMF component energy. K represents the modes after VMD decomposition, and K2 represents the number of modes in VMD decomposition.

[0020] Sub-step S23: Extract dynamic signal features, including time-domain features, frequency-domain features, and energy entropy, from the modes after decomposition of drilling vibration and acoustic emission signals. The time-domain features are as follows:

[0021] , ;

[0022] The frequency domain characteristics are as follows:

[0023] , ;

[0024] The energy entropy is as follows:

[0025] ;

[0026] Wherein, RMS represents the mode after VMD decomposition. The root mean square value is a statistical representation of the signal's time-domain energy. Peak represents the mode after VMD decomposition. The peak value is given by N, where N is the number of signal sampling points for the k-th mode after VMD decomposition, FC is the centroid frequency, and MSF is the mean square frequency. The k-th mode after VMD decomposition The power spectral density, The instantaneous frequency value in the frequency domain. For the k-th VMD mode energy, Let H be the energy percentage of the k-th VMD mode, and H be the energy entropy.

[0027] In the in-situ rock strata identification method based on physical information machine learning of the present invention, specifically, the fused feature vector in step S3 is obtained through the following sub-steps:

[0028] Sub-step S31: Feature dimension organization. The standardized multi-source signal parameter features are organized into a one-dimensional vector with four dimensions: drilling pressure, drill rod torque, drill rod rotation speed, and drilling speed. The dynamic signal features are organized into a K2×5-dimensional vector, where K2 is the VMD decomposition mode number, which includes the five features after mode decomposition.

[0029] Sub-step S32: Dimensional alignment, dividing the samples for feature fusion according to the time window of synchronous acquisition of multi-source signal parameters;

[0030] Sub-step S33: Redundancy removal. Pearson correlation analysis is used. If the correlation coefficient between dynamic signal features exceeds the set value, dynamic signal features with greater mutual information to the rock layer strength label are retained.

[0031] Sub-step S34: Vector concatenation. The standardized multi-source parameter features and the de-redundant dynamic signal features are concatenated in sequence to form a unified fused feature vector.

[0032] In the in-situ rock strata identification method based on physical information machine learning of the present invention, the training process of the physical information machine learning model is as follows:

[0033] Step K1: Obtain multiple sets of historical drilling parameters, including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals. Label each set of historical drilling parameters with corresponding laboratory standard rock mass parameter labels, including the strength prediction value of the drilled rock strata and the rock mass category. Divide the multiple sets of historical drilling parameters into training set, validation set, and test set.

[0034] Step K2: Perform two mode decompositions on the drilling vibration and acoustic emission signals of each set of historical drilling parameters using CEEMDAN and VMD to obtain dynamic signal characteristics related to rock mass fracture.

[0035] Step K3: After standardizing and feature processing the remaining historical drilling parameters, combine them with the dynamic signal features of the drilling vibration and acoustic emission signals in the same group to form a fused feature vector;

[0036] Step K4: Input the fused feature vector of historical drilling parameters from the training set into the physical information machine learning model for training. The physical information machine learning model uses an LSTM network as its backbone and constructs a physical inversion equation between drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and rock strength based on the Buschnesk elastic half-space theory.

[0037] ;

[0038] Wherein, UCS is the calculated rock strength value obtained from the inversion, WOB is the drilling pressure, RPM is the drill pipe rotation speed, ROP is the drilling speed, and Torque is the drill pipe torque. , These are physical proportionality coefficients, representing the contribution weights of the product of drilling pressure and drill pipe rotation speed, and drill pipe torque to the rock strength inversion, respectively. This is the intercept term, used to correct systematic deviations between the model and actual operating conditions;

[0039] The loss function for embedding the physical inversion equation into the physical information machine learning model is as follows:

[0040] ;

[0041] in, It is data and The driving loss, It is the actual value of rock layer strength in the laboratory standard rock mass parameter label. It is the rock strength prediction value from the physical information machine learning model. It is the physical constraint loss, that is, the calculated rock strength value UCS obtained through the physical inversion equation. It is a trade-off parameter;

[0042] The physical information machine learning model outputs the predicted strength value of the drilled rock strata and the rock mass type.

[0043] Step K5: Using the Adam optimizer, input the validation set and test set into the physical information machine learning model for validation and testing, and obtain the trained physical information machine learning model.

[0044] Furthermore, the present invention also discloses an in-situ identification method for drilling in frozen strata in cold regions, which uses the above-mentioned in-situ identification method for rock strata based on physical information machine learning to identify the rock strata strength value and rock mass type during the drilling process in frozen strata in cold regions.

[0045] Furthermore, the present invention also discloses an experimental apparatus for realizing the above-mentioned in-situ identification method for drilling in frozen strata in cold regions, including a drilling system, a cold environment simulation box, and a multi-source drilling information acquisition system;

[0046] The drilling system uses a hydraulic rock drilling rig;

[0047] The cold environment simulation box is fixed relative to the drilling system. The sample assembly after freeze-thaw is fixed inside. The hydraulic rock drilling machine of the drilling system is connected to the drill rod and passes through one side of the cold environment simulation box to drill the sample assembly. A pressure system is set outside the cold environment simulation box to clamp and fix the sample assembly. The pressure system has a pressure head that applies pressure to the sample assembly in the normal and tangential directions from the drilling direction.

[0048] The multi-source drilling information acquisition system includes:

[0049] A pressure sensor is installed on the impact mechanism of the hydraulic rock drilling rig to collect the drilling pressure during the drilling process.

[0050] Torque and speed sensors are installed at the connection between the rotary mechanism and the drill rod of the hydraulic rock drilling rig to collect the drill rod torque and drill rod speed during the drilling process of the hydraulic rock drilling rig.

[0051] A displacement sensor is installed on the hydraulic rock drilling rig to collect the moving displacement of the hydraulic rock drilling rig during the drilling process and obtain the drilling speed.

[0052] Vibration and acoustic emission sensors are installed on the pressure head of the pressurization system to collect drilling vibration and acoustic emission signals generated by the hydraulic rock drilling rig during the rock breaking process of the sample assembly.

[0053] The multi-source drilling information acquisition system synchronously acquires multi-source signal parameters such as drilling pressure, drill rod torque, drill rod speed, drilling speed, and drilling vibration and acoustic emission signals during the drilling process of the hydraulic rock drilling rig. The in-situ identification method for drilling frozen strata in cold regions as described in claim 5 is used to obtain the rock strength value and rock mass type of the sample assembly.

[0054] In the above-described experimental apparatus of the present invention, the sample assembly further includes a rock mass sample and clamping plates. The two clamping plates clamp the rock mass sample in the drilling direction by means of a tie rod, and drilling holes for the drill rod to pass through are provided on the clamping plates. After the sample assembly is placed in the cold environment simulation chamber, the pressure head of the pressurization system symmetrically clamps the side of the rock mass sample from the normal and tangential directions of the drilling direction through the pads. The vibration and acoustic emission sensors are arranged on the pads.

[0055] In the above-mentioned experimental apparatus of the present invention, the inner wall of the cold environment simulation chamber is provided with heat insulation material, the side wall is provided with a cold injection hole for injecting refrigerant into the cold environment simulation chamber, and a temperature sensor is provided on the outer wall of the chamber and connected to a temperature display outside the cold environment simulation chamber.

[0056] The experimental apparatus described above in this invention further includes a freeze-thaw cycle chamber for conducting cyclic freeze-thaw simulation experiments on the sample assembly.

[0057] This invention also discloses an intelligent drilling control method. During the drilling process, the drilling system collects multi-source signal parameters in real time, including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals. The real-time rock strata strength value and rock mass category of the drilling strata are obtained by using the in-situ rock strata identification method based on physical information machine learning described above. The drilling system adjusts the matching drilling parameters according to the real-time rock strata strength value and rock mass category to adapt.

[0058] The above-mentioned technical solution of the present invention has the following beneficial effects:

[0059] (1) The in-situ rock strata identification method based on physical information machine learning provided by this invention adopts an improved CEEMD-VMD joint algorithm and a physical information machine learning model, which can identify and output rock strata type and rock mass strength parameters online in real time. This invention uses an improved CEEMD to perform preliminary decomposition on the vibration and acoustic emission signals collected during the rock strata drilling process to obtain a series of intrinsic mode functions (IMFs). Then, variational mode decomposition (VMD) is used to perform secondary decomposition on the key IMF components with high energy to accurately separate the characteristic frequency bands related to rock mass fracture in the vibration and acoustic emission signals during the drilling process, effectively suppressing the interference of environmental noise on the physical information machine learning model. An attention-enhanced LSTM network is used as the backbone of the physical information machine learning model. The input of the model includes drilling parameters and dynamic signal features of drilling vibration and acoustic emission signals. The LSTM layer is used to effectively learn the time series dependence of the drilling process, while the attention mechanism dynamically assigns different weights to features at different times, so that the model focuses on key drilling stages and improves the identification accuracy of rock strata type and rock mass strength parameters.

[0060] (2) The in-situ identification method for drilling in frozen strata in cold regions provided by this invention utilizes the advantages of the aforementioned in-situ identification method for drilling strata. Through the deep fusion of synchronous acquisition of multi-source drilling signals and physical information machine learning algorithms, it achieves online, real-time, and accurate identification of multiple parameters of the rock mass during drilling in frozen strata in cold regions. The experimental device used integrates a three-dimensional deep fusion experimental platform for simulating the freeze-thaw environment of drilling samples, drilling tests of full-size rock strata samples, and intelligent sensing of multi-source drilling information, realizing high-fidelity reproduction of the field working conditions of frozen strata in cold regions in indoor experiments.

[0061] (3) The intelligent drilling control method provided by this invention, which applies the above-mentioned in-situ rock strata identification method based on physical information machine learning, feeds back the rock strata type and rock mass strength parameter results identified in-situ during the drilling process to drilling decisions. When the drilling process identifies that the rock mass strength has decreased beyond the threshold due to freeze-thaw degradation, the drilling parameters are adjusted in real time. The optimization can be used to guide or directly control the drilling rig for adaptive drilling, and the identified rock strata type and rock mass strength results serve as important bases for blasting parameter optimization or slope stability assessment. It realizes an online closed loop from "drilling response" to "rock mass characteristics" and then to "decision control", providing key technical equipment and data support for safe, efficient and intelligent mining in cold regions. It can be extended to multiple fields such as geological exploration and tunnel excavation.

[0062] In summary, this invention synchronously and in real-time collects drilling parameters such as drilling pressure, drill rod torque, drill rod rotation speed, and drilling speed, as well as multi-dimensional vibration and acoustic emission signals during drilling. Based on a physical information machine learning model, it achieves in-situ intelligent identification of rock strata types and rock mass strength during drilling. The identification and sensing accuracy is precise and efficient. Applied to adaptive feedback of drilling strata and drilling parameter recommendation capabilities, it enables intelligent drilling control, achieving intelligent and efficient drilling construction. Combined with the complex geological conditions of frozen strata in cold regions, this invention provides advanced indoor testing and decision support equipment for the precise and efficient development of mineral resources in cold regions.

[0063] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0064] Figure 1 This is a schematic diagram illustrating the steps of the in-situ rock strata identification method based on physical information machine learning in this invention.

[0065] Figure 2 This is a schematic diagram of the overall experimental setup for implementing an in-situ identification method for drilling through frozen strata in cold regions, as an example.

[0066] Figure 3 This is a three-dimensional schematic diagram of the drilling system in Example 3 drilling the specimen assembly inside the cold environment simulation chamber.

[0067] Figure 4 This is a front view of the drilling system in Example 3 drilling the specimen assembly inside the cold environment simulation chamber.

[0068] Figure 5 This is a top view of the drilling system in Example 3 drilling the specimen assembly inside the cold environment simulation chamber.

[0069] Figure 6 This is a schematic diagram of the specimen assembly being clamped and fixed by the pressurization system inside the cold environment simulation chamber in Example 3.

[0070] Figure 7 This is a schematic diagram of the specimen assembly structure in Example 3.

[0071] Figure 8 This is a schematic diagram of the in-situ rock strata identification method in this embodiment and its overall process in engineering applications.

[0072] The diagram is labeled as follows: 100-Drilling system, 101-Drill rod, 11-Hydraulic rock drilling rig, 12-Propulsion mechanism, 13-Rock drill travel rail;

[0073] 200-Cold environment simulation chamber, 21-Pressure system, 211-Pressure support, 212-Pressure cylinder, 213-Normal indenter, 214-Tangential indenter, 22-Sample placement platform, 23-Base, 231-Base water tank, 232-Base drain pipe;

[0074] 300-Sample assembly, 301-Rock mass sample, 302-Clamping plate, 303-Tie rod, 304-Pad plate;

[0075] 400-Control system, 401-Control console, 41-Multi-source drilling information acquisition system, 411-Displacement sensor, 412-Vibration and acoustic emission sensor, 413-Torque sensor, 414-Speed ​​sensor, 415-Pressure sensor;

[0076] 500-Freeze-thaw cycle chamber. Detailed Implementation

[0077] Example 1

[0078] See Figure 1 and Figure 2 The in-situ rock strata identification method based on physical information machine learning of the present invention specifically includes the following steps:

[0079] Step S1: During the drilling process, collect multi-source signal parameters including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals.

[0080] Step S2: Perform two mode decompositions on the drilling vibration and acoustic emission signals using CEEMDAN and VMD to obtain dynamic signal characteristics related to rock mass fracture.

[0081] This step specifically includes the following sub-steps:

[0082] Sub-step S21: Perform preliminary decomposition of drilling vibration and acoustic emission signals using CEEMDAN to obtain a series of intrinsic mode functions: .

[0083] in, The original drilling vibration and acoustic emission signals, For intrinsic mode functions, K1 represents the residual, and K1 represents the total number of intrinsic mode functions obtained from the CEEMDAN decomposition.

[0084] Sub-step S22: The key IMF components with high energy in drilling vibration and acoustic emission signals are decomposed a second time using variational mode decomposition (VMD) to accurately separate the characteristic frequency bands related to the fracturing of freeze-thawed rock masses, effectively suppressing environmental noise interference. .

[0085] in, The key IMF variables with a large proportion of IMF component energy are selected from the intrinsic mode functions in sub-step S21. K represents the mode after VMD decomposition, and K2 represents the number of modes in VMD decomposition.

[0086] Key IMF variables selected from intrinsic mode functions that have a large proportion of IMF component energy. The screening criteria are based on the energy proportion of the IMF components, and the specific steps are as follows:

[0087] a. Calculate each intrinsic mode function (IMF) in sub-step S21. k The energy E of (t) k The energy calculation formula is: Where N is the number of sampling points for the original drilling vibration and acoustic emission signals.

[0088] b. All IMFs k (t) according to energy E k Sort by size from largest to smallest.

[0089] c. Calculate the intrinsic mode function (IMF) for each intrinsic mode function. k The cumulative energy percentage of (t) , (m≤K1), where K1 is the total number of IMFs in the CEEMDAN decomposition in sub-step S21. The cumulative energy percentage is selected. The first m IMF components, ≥80%, are merged into the key IMF variable s(t) to ensure that the main effective signal components related to the fracturing of the freeze-thaw rock mass are retained and low-energy noise components are removed.

[0090] Sub-step S23: Extract dynamic signal features, including time-domain features, frequency-domain features, and energy entropy, from the modes after decomposition of drilling vibration and acoustic emission signals. The time-domain features are as follows: , .

[0091] The frequency domain characteristics are as follows: , .

[0092] The energy entropy is as follows: .

[0093] Wherein, RMS represents the mode after VMD decomposition. The root mean square value is a statistical representation of the signal's time-domain energy. Peak represents the mode after VMD decomposition. The peak value is given by N, where N is the number of signal sampling points for the k-th mode after VMD decomposition, FC is the centroid frequency (dimensionless), representing the core frequency location where signal energy is concentrated, and MSF is the mean square frequency (Hz²), reflecting the dispersion of the signal frequency distribution. The k-th mode after VMD decomposition The power spectral density, This represents the instantaneous frequency value in the frequency domain, measured in Hertz (Hz). For the k-th VMD mode Energy, measured in volts squared (V²). Let H be the energy percentage of the k-th VMD mode, and H be the energy entropy, which is dimensionless and characterizes the uniformity of energy distribution among modes after VMD decomposition. The smaller the entropy value, the more concentrated the energy; the larger the entropy value, the more dispersed the energy, and the stronger the corresponding noise interference.

[0094] Step S3: Standardize the remaining multi-source signal parameters, such as Z-score standardization. Combine the standardized drilling pressure, drill rod torque, drill rod rotation speed, and drilling speed with the dynamic signal features in S2 to form a fused feature vector, which serves as the unified input to the physical information machine learning model.

[0095] Sub-step S31: Feature dimension organization. The standardized multi-source signal parameter features are organized into a one-dimensional vector with four dimensions: drilling pressure, drill rod torque, drill rod rotation speed, and drilling speed. [WOB] std Torque std RPM std ROP std ], WOB std Standardized drilling pressure, RPM std For standardized drill pipe rotation speed and ROP std For standardized drilling speed, Torque std To obtain the standardized drill pipe torque, the dynamic signal features are organized into a K2×5 dimensional vector, where K2 is the number of VMD decomposition modes, including five features for each mode: time domain features RMS, Peak, frequency domain features FC, MSF, and energy entropy.

[0096] Sub-step S32: Dimensional alignment. Divide the samples for feature fusion according to the time window of synchronous acquisition of multi-source signal parameters to ensure that the number of samples of the two types of features is consistent.

[0097] Sub-step S33: Redundancy removal. Pearson correlation analysis is used. If the correlation coefficient between dynamic signal features exceeds the set value, features with greater mutual information with the rock layer strength label are retained.

[0098] Sub-step S34: Vector concatenation. The standardized multi-source parameter features and the de-redundant dynamic signal features are concatenated sequentially to form a unified fused feature vector F=[WOB]. std Torque std RPM std ROP stdf1, f2, ..., f M ], where M is the dimension of the dynamic signal features after redundancy removal.

[0099] Sub-step S4: Input the fused feature vector from step S3 into the pre-trained physical information machine learning model. The physical information machine learning model uses an LSTM network as its backbone and constructs a physical inversion equation between drilling pressure, drill rod torque, drill rod rotation speed, drilling speed and rock strength based on the Buschnesk elastic half-space theory to perform physical constraints. It outputs the predicted strength value of the drilled rock layer and the rock mass category.

[0100] The training process of the physical information machine learning model is as follows:

[0101] Step K1: Obtain multiple sets of historical drilling parameters, including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals. Label each set of historical drilling parameters with corresponding laboratory standard rock mass parameter labels, including the strength prediction value of the drilled rock strata and the rock mass category. Divide the multiple sets of historical drilling parameters into training set, validation set, and test set.

[0102] Step K2: Perform two mode decompositions on the drilling vibration and acoustic emission signals of each set of historical drilling parameters using CEEMDAN and VMD to obtain dynamic signal characteristics related to rock mass fracture. The process of the two mode decompositions is shown in step S2.

[0103] Step K3: After standardizing the remaining historical drilling parameters of each group, combine them with the dynamic signal features of the drilling vibration and acoustic emission signals of the same group to form a fused feature vector.

[0104] For the drilling pressure, drill pipe torque, drill pipe rotation speed, and drilling speed collected in step K1, Z-score normalization (zero-mean normalization) is used for feature processing. The core purpose is to eliminate the dimensional differences between different parameters, ensuring that all features are on the same order of magnitude, preventing the machine learning model from being dominated by dimensions, and improving the model's convergence speed and generalization ability. Based on the statistical features of the training set (to avoid data leakage from the test set), the sample values ​​of each multi-source signal parameter in the historical drilling parameters are standardized using the following formula: .

[0105] in, The original multi-source signal parameter values ​​for the i-th historical drilling parameter sample, such as the drilling pressure at a certain moment; For the standardized parameter value of the i-th sample, The mean of this parameter in the training set. This represents the standard deviation of the parameter in the training set. If the parameter value of a sample exceeds [μ−3], then... μ+3 The range is considered an outlier, and the μ value of this parameter in the training set is used to replace it. In engineering scenarios, this is usually caused by sensor failure or drilling impact, and interference with the model should be avoided.

[0106] Step K4: Input the fused feature vector of historical drilling parameters from the training set into the physical information machine learning model for training. The physical information machine learning model uses an LSTM network as its backbone and constructs a physical inversion equation between drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and rock strength based on the Buschnesk elastic half-space theory.

[0107] .

[0108] Where UCS is the calculated rock strength value obtained from the inversion, WOB is the drilling pressure, RPM is the drill pipe rotation speed, ROP is the drilling speed, and Torque is the drill pipe torque. , , Let be the coefficients to be determined, where , These are physical proportionality coefficients, representing the contribution weights of the product of drilling pressure and drill pipe rotation speed, and drill pipe torque to the rock strength inversion, respectively. This is the intercept term, used to correct for systematic deviations between the model and actual operating conditions. During model training, the coefficients... , and As trainable parameters, they are optimized synchronously with the LSTM network weights using the same Adam optimizer.

[0109] The loss function for embedding the physical inversion equation into the physical information machine learning model is as follows:

[0110] .

[0111] in, It is data and The driving loss, It is the actual value of rock layer strength in the laboratory standard rock mass parameter label. It is the rock strength prediction value from the physical information machine learning model. It is the physical constraint loss, that is, the calculated rock strength value UCS obtained through the physical inversion equation. It is a tradeoff parameter that controls the importance of physical constraints relative to data fitting. Its value is a positive real number greater than 0, used to adjust the weight of physical constraints in the total loss function. Its optimal value is determined by hyperparameter tuning on the validation set before training.

[0112] The physical information machine learning model outputs predicted strength values ​​and rock mass categories of the drilled rock formations. The model employs an attention-enhanced LSTM network as its backbone. The model input includes a fused feature vector of drilling parameters and dynamic signal features. The LSTM layer effectively learns the time-series dependencies of the drilling process, while the attention mechanism dynamically assigns different weights to features at different times, allowing the model to focus on key drilling stages and improve recognition accuracy.

[0113] Step K5: Using the Adam optimizer, the validation set and test set are input into the physical information machine learning model for validation and verification, respectively, to obtain the trained physical information machine learning model. The initial learning rate of the Adam optimizer is set to 0.001. When the loss on the validation set no longer decreases for 5 consecutive epochs, the learning rate is halved.

[0114] Example 2

[0115] During drilling in frozen strata in cold regions, the strata undergo repeated freezing and thawing cycles, resulting in unpredictable changes in their structural strength compared to ordinary strata. This embodiment uses the in-situ rock strata identification method based on physical information machine learning from Embodiment 1 to identify the rock strata strength values ​​and rock mass types during drilling in frozen strata in cold regions. This allows for the rapid and accurate acquisition of drilling parameters for frozen strata in cold regions, enabling precise drilling in these strata.

[0116] Example 3

[0117] This embodiment discloses an experimental device for an in-situ identification method for drilling in frozen strata in cold regions, so as to realize drilling experiments simulating the geological environment of cold strata.

[0118] like Figures 3-7As shown, the experimental setup in this embodiment includes a drilling system 100, a cold-region environment simulation chamber 200, and a multi-source drilling information acquisition system 41. The drilling system 100 uses a hydraulic rock drilling rig 11; the cold-region environment simulation chamber 200 is fixed relative to the drilling system 100, and a sample assembly 300 after freeze-thaw cycles is fixed inside. The hydraulic rock drilling rig 11 of the drilling system 100 is connected to a drill rod 101, which passes through one side of the cold-region environment simulation chamber to drill into the sample assembly 300. A pressure system 21 is provided outside the cold-region environment simulation chamber 200 to clamp and fix the sample assembly. The pressure system 21 has a pressure head that applies pressure to the sample assembly in the normal and tangential directions from the drilling direction. The multi-source drilling information acquisition system 41 simultaneously acquires multi-source signal parameters during the drilling process of the hydraulic rock drilling rig, including drilling pressure, drill rod torque, drill rod speed, drilling speed, and drilling vibration and acoustic emission signals. These parameters include a displacement sensor 411, a vibration and acoustic emission sensor 412, a torque sensor 413, a speed sensor 414, and a pressure sensor 415. Specifically, the displacement sensor 411 is installed on the hydraulic rock drilling rig to acquire the drilling speed by measuring the displacement of the rig during drilling; the vibration and acoustic emission sensor 412 is installed on the pressure head of the pressurization system to acquire the drilling vibration and acoustic emission signals generated by the hydraulic rock drilling rig during the rock-breaking process of the sample assembly; the torque sensor 413 and the speed sensor 414 are installed at the connection between the rotary mechanism and the drill rod of the hydraulic rock drilling rig to acquire the drill rod torque and drill rod speed during drilling; and the pressure sensor 415 is installed on the impact mechanism of the hydraulic rock drilling rig, mounted at the inlet and outlet of the impact mechanism cylinder, to acquire the drilling pressure during drilling based on the hydraulic difference.

[0119] The drilling system 100 in this embodiment includes a hydraulic rock drill 11 and a propulsion mechanism 12. The hydraulic rock drill 11 is a rock-breaking actuator, mounted on a rock drill travel rail 13 via its base, and can move horizontally along the rail. One end of the drill rod 101 is connected to the shank of the hydraulic rock drill 11, and the other end is equipped with a drill bit. During the drilling experiment, the drill rod 101 and the drill bit pass through the cold environment simulation chamber to perform drilling operations on the sample assembly 300. The hydraulic rock drill 11 is slidably mounted on the frame of the drilling system via the rock drill travel rail 13, which provides travel guidance for the hydraulic rock drill 11 during the drilling process. The propulsion mechanism 12 employs a combination of a propulsion cylinder and a winch wire rope. The propulsion cylinder is connected to the hydraulic rock drill 11, directly pushing the body of the hydraulic rock drill 11 to slide linearly along the rock drill stroke rail 13. The winch wire rope is connected to the other side of the hydraulic rock drill body and is connected to the winch mechanism. When the propulsion cylinder pushes the hydraulic rock drill 11 and the drill rod to drill towards the sample assembly, the winch mechanism drives the winch wire rope to pull the hydraulic rock drill 11 from the other side to assist in drilling. During the process of the propulsion cylinder retracting and pulling the hydraulic rock drill 11 back, the winch mechanism unwinds the winch wire rope to assist the hydraulic rock drill in retracting along the rock drill stroke rail. The displacement sensor 411 is a pull-wire displacement sensor, installed at one end of the rock drill stroke rail 13. The pull rope is connected to the bottom of the hydraulic rock drill. During the drilling process, the hydraulic rock drill moves along the rock drill stroke rail, causing the pull-wire displacement sensor to record the specific drilling stroke displacement, thereby obtaining the drilling speed.

[0120] In this embodiment, the cold environment simulation chamber 200 simulates the cold geological environment during the drilling process of the sample assembly. A low-temperature environment is provided by injecting liquid nitrogen or other refrigerants into the chamber, and a pressurization system 21 is installed to pressurize the sample assembly, simulating the geostress environment of the cold geological formation. Specifically, the cold environment simulation chamber 200 has openable doors on both the side near the drilling direction and the side away from the drilling direction for drilling and for placing and removing the sample assembly 300. A borehole pipe is installed on the door of the cold environment simulation chamber. When the drill rod 101 is drilling, the drill bit passes through the borehole pipe. The drilling surface of the sample assembly 300 comes into contact with the borehole pipe after the sample assembly is placed in the cold environment simulation chamber. The borehole pipe provides sealing, support, and guidance for the drill rod 101 during drilling. An observation window is provided on the door of the cold environment simulation chamber 200 for researchers to observe the drilling process inside the chamber.

[0121] The inner walls of the cold environment simulation chamber 200 are lined with thermal insulation material, and the chamber walls are equipped with cooling injection holes for injecting refrigerant into the chamber. During the drilling experiment, liquid nitrogen or other refrigerants are injected into the chamber through these holes to simulate and maintain the low-temperature environment of the cold-region strata where the sample assembly is located. A display is installed on the outer wall of the chamber door. The display, along with temperature sensors installed inside the chamber, such as a distributed fiber optic temperature measurement network (accuracy ±0.1℃) and a high-precision humidity sensor (RH10-100%), monitors the environmental conditions inside the simulation chamber in real time. The temperature sensors are connected to the display for feedback, showing the real-time internal temperature values ​​and feeding them back to the control system, allowing researchers to precisely control the simulated low-temperature environment.

[0122] The pressurization system 21 installed on the cold environment simulation chamber 200 includes a pressurization bracket 211, a pressurization cylinder 212, a normal pressure head 213, and a tangential pressure head 214. The pressurization bracket 211 adopts a high-rigidity C-shaped frame and is fixedly installed on the outside of the cold environment simulation chamber 200 around the drilling direction of the sample assembly. The pressurization bracket 211 is directly fixedly connected to the frame of the drilling system 100. Taking the horizontal drilling of the sample assembly by the drill rod 101 as an example, with the horizontal drilling direction of the drill rod as the center line, the pressurization bracket 211 surrounds the outside of the cold environment simulation chamber 200. Its multiple pressure heads are set through the outer wall of the cold environment simulation chamber along the normal and tangential directions of the drilling direction. Pressurization holes for the pressure heads to pass through are provided on the outer wall of the cold environment simulation chamber 200. The normal pressure head 213 passes through the pressure hole on the top wall of the cold environment simulation chamber. A vertical pressure cylinder 212 on the pressure support 211, positioned at the top of the chamber, presses and clamps the sample assembly inside the chamber 200 from top to bottom. A tangential pressure head 214 passes through the pressure holes on both sides of the chamber walls. Horizontal pressure cylinders 212 on the pressure support 211, positioned on both sides of the chamber, press and clamp the sample assembly inside the chamber 200 from both sides. Vibration and acoustic emission sensors 412 are integrated on the pressure head that clamps the sample assembly, collecting drilling vibration and acoustic emission signals transmitted by the pressure head during the rock-breaking process. Thus, the pressure system 21 clamps and fixes the sample assembly 300 while simultaneously simulating the geostress environment experienced by the strata where the sample assembly 300 is located. Specifically, as shown... Figure 5 As shown.

[0123] See further Figure 6In this embodiment, the sample assembly 300 includes a rock mass sample 301 and clamping plates 302. The rock mass sample 301 is cut into a square column during sampling. The two clamping plates 302 clamp the two end faces of the rock mass sample 301 in the drilling direction of the rock mass sample through the tie rod 303. Drilling holes for the drill rod to pass through are provided on the clamping plates 301. Since the sample assembly 300 is an isolated rock mass, there is no real compressive stress around it. In this embodiment, after the clamping plates 301 clamp the rock mass sample 301 in the drilling direction, the integrity of the rock mass sample 301 in the drilling direction is enhanced. This can prevent the rock mass sample from breaking off due to the lack of compressive stress behind the rock mass sample after the drill rod enters the interior of the rock mass sample 301. Furthermore, in order to more realistically simulate the geostress environment of the sample assembly 300 during the drilling process, pads 304 are provided on the other four sides of the sample assembly 300. After the sample assembly 300 is placed in the cold environment simulation box 200, the normal pressure head 213 and the tangential pressure head 214 of the pressurization system 21 clamp the sides of the rock mass sample symmetrically from the normal and tangential directions of the drilling direction through the pads 304. The pressure of the pressure head is evenly distributed on the rock mass sample 301 through the pads 304 to simulate the peripheral compression geostress experienced by the sample assembly 300 in the formation.

[0124] The sample assembly 300 is placed on the base 23 at the bottom of the cold environment simulation chamber. The base 23 is directly fixed to the frame of the drilling system 100. Since drilling coolant is injected into the drill rod 101 during the drilling process to cool the drill rod and drill bit, a base water tank 231 for collecting drilling coolant is also provided on the base 23. The base water tank 231 is connected to the base drain pipe 232 and connected to the outside of the cold environment simulation chamber 200 to quickly collect and discharge the coolant during the drilling process from the cold environment simulation chamber 200.

[0125] In the experimental setup of this embodiment, the sample assembly 300 is fixed to the frame of the drilling system via the pressurization system 21. The housing of the cold environment simulation chamber 200 is fixedly mounted on the frame at the location of the sample assembly 300. The low-temperature environment of the frozen strata in the cold region is simulated around the sample assembly 300, and it does not directly bear the drilling force during the drilling process of the drill pipe, thus ensuring the structural stability of the cold environment simulation chamber.

[0126] The sample assembly 300 used in this embodiment can be obtained directly from on-site sampling in cold-region geological construction sites, or it can be obtained in a laboratory by repeatedly freezing and thawing ordinary geological rock samples using a freeze-thaw cycle chamber 500 to simulate the geological structure of different cold-region strata. The former can be used for rapid sampling experiments at on-site geological construction sites in cold regions, while the latter can be used for simulation experiments in the laboratory to simulate the geological structure of different cold-region strata. In the latter experimental setup, the freeze-thaw cycle chamber 500 is placed on one side of the cold-region environment simulation chamber 200, such as... Figure 7As shown, the freeze-thaw cycle chamber 500 is equipped with a temperature control system for repeatedly freezing and thawing the sample assembly, such as using a combination of liquid nitrogen injection and resistance heating for temperature control. The rock mass sample of the sample assembly 300 is placed in the freeze-thaw cycle chamber 500 and subjected to multiple freeze-thaw cycles according to a preset temperature curve and time to simulate the damage and deterioration of rocks in cold-region strata due to seasonal changes, thus preparing experimental rock mass samples that conform to real-world working conditions. In this embodiment, when the freeze-thaw cycle chamber 500 is used to batch produce the sample assemblies 300 for the experimental platform, a sample placement platform 22 is set on the side of the cold-region environment simulation chamber 200 away from the drilling direction. This platform is used to receive and place the sample assemblies after freezing and thawing in the freeze-thaw cycle chamber. The sample assemblies are transferred on the sample placement platform 22 via a sample tray equipped with a sliding rail assembly. The sample assembly 300 is transferred from the freeze-thaw cycle chamber 500 to the sample tray on the test stand 22 by a lifting device. Then, the sample assembly 300 is aligned with the positioning block set on the cold environment simulation chamber 200 by the slide rail assembly. The sample assembly 300 is pushed smoothly and accurately to the designated test position in the cold environment simulation chamber 200 by pushing it.

[0127] In practical applications, a temperature control system can be set up inside the cold environment simulation chamber 200 and the freeze-thaw cycle chamber 500 to repeatedly freeze and thaw the sample assembly. The cold environment simulation chamber 200 can be used to directly simulate the cyclic freeze-thaw process of the sample assembly. However, considering that the freeze-thaw cycle is relatively long, the above-mentioned scheme of setting up the freeze-thaw cycle chamber 500 separately can prepare the frozen and thawed sample assemblies in batches. Only the cold environment simulation chamber 200 is used to simulate the low temperature environment of the cold formation during the drilling process, thereby improving the efficiency of the drilling experiment.

[0128] In this embodiment, a control system 400 is set on the other side of the drilling system. The main body of the control system 400 is a console 401, which integrates a multi-source drilling information acquisition system 41. The multi-source drilling information acquisition system 41 synchronously acquires multi-source signal parameters such as drilling pressure, drill rod torque, drill rod speed, drilling speed, and drilling vibration and acoustic emission signals during the drilling process of the hydraulic rock drilling rig through displacement sensor 411, vibration and acoustic emission sensor 412, torque sensor 413, speed sensor 414, and pressure sensor 415. The sampling frequency is ≥1kHz. The acquired multi-source signal parameters are input to the console 401. The controller in the console 401 analyzes and obtains the rock strength value and rock mass type of the sample assembly through the in-situ identification method for drilling in frozen strata in cold regions in Embodiment 2.

[0129] The experimental method of the above-mentioned experimental apparatus in this embodiment includes the following steps:

[0130] Step 1: Sample Assembly Preparation and Installation. Rock samples are collected in the field in cold regions or prepared using a freeze-thaw cycle chamber. The sample assembly is then placed into the freeze-thaw environment simulation chamber.

[0131] Step 2, Stress Loading. Apply the set geostress pressure to the sample using the pressurizing cylinder and pressure head of the pressurizing system, or apply fissure water pressure by activating the seepage system according to experimental requirements.

[0132] S3. Testing while drilling and synchronous data acquisition. During the temperature stabilization phase of the freeze-thaw environment simulation chamber, the hydraulic rock drilling rig was started to drill the sample assembly, and the multi-source information acquisition system while drilling was triggered simultaneously to collect drilling pressure, drill rod torque, drill rod speed, drilling speed, drilling vibration and acoustic emission signals in real time.

[0133] S4. Data Processing and Real-time Identification. The console performs CEEMD-VMD joint denoising and feature extraction on the acquired raw multi-source signal parameters, inputs the fused feature vector into the physical information machine learning model, and outputs the rock mass category and strength prediction values ​​online in real time.

[0134] Example 4

[0135] This embodiment applies the in-situ rock strata identification method based on physical information machine learning of the present invention to intelligent drilling control during the drilling process. The drilling system collects multi-source signal parameters in real time during drilling, including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals. Using the in-situ rock strata identification method based on physical information machine learning from Embodiment 1, the real-time rock strata strength value and rock mass category are obtained. The drilling system adjusts the matching drilling parameters according to the real-time rock strata strength value and rock mass category to adapt. When the rock mass strength degradation exceeds a threshold, this result can be used as an important basis for subsequent blasting parameter optimization or slope stability assessment.

[0136] like Figure 8 As shown, the physical information machine learning model trained in Example 1 is deployed on the drilling system's industrial control computer. During real-time drilling, the drilling system automatically executes the process of "multi-source signal parameter acquisition → CEEMD-VMD processing → feature extraction → model inference," outputting drilling lithology identification results of rock layer strength values ​​and rock mass types within seconds. It can also generate drilling parameter optimization suggestions (such as "suggest increasing the drill pressure by 2 kN and reducing the rotation speed by 50 rpm"), which are sent to the drilling rig control system via the Modbus TCP protocol to achieve adaptive control. The drilling system's industrial control computer can also be equipped with a digital twin verification interface to support bidirectional verification between experimental data and field data.

[0137] To make the objectives, technical solutions, and advantages 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0138] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0139] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. It should also be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0140] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for in-situ identification of rock strata during drilling based on physical information machine learning, characterized in that... Includes the following steps: Step S1: During the drilling process, collect multi-source signal parameters including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals; Step S2: Perform two mode decompositions on the drilling vibration and acoustic emission signals using CEEMDAN and VMD to obtain dynamic signal characteristics related to rock mass fracture. Step S3: After standardizing the remaining multi-source signal parameters, combine them with the dynamic signal features from step S2 to form a fused feature vector; Step S4: Input the fused feature vector from step S3 into the pre-trained physical information machine learning model. The physical information machine learning model uses an LSTM network as its backbone and constructs a physical inversion equation between drilling pressure, drill rod torque, drill rod rotation speed, drilling speed and rock strength based on the Businesk elastic half-space theory to perform physical constraints. It outputs the predicted strength value of the drilled rock layer and the rock mass category.

2. The in-situ rock strata identification method based on physical information machine learning according to claim 1, characterized in that: Step S2 includes the following sub-steps: Sub-step S21: Perform preliminary decomposition of drilling vibration and acoustic emission signals using CEEMDAN to obtain a series of intrinsic mode functions: ； in, The original drilling vibration and acoustic emission signals, For intrinsic mode functions, K1 is the residual, and K1 is the total number of intrinsic mode functions obtained from the CEEMDAN decomposition. Sub-step S22: Perform secondary decomposition of the key IMF components with high energy in the drilling vibration and acoustic emission signals using variational mode decomposition (VMD) to accurately separate the characteristic frequency bands related to the fracturing of the freeze-thaw rock mass. ； in, The key IMF variables with a large proportion of IMF component energy are selected from the intrinsic mode functions in sub-step S21. K represents the modes after VMD decomposition, and K2 represents the number of modes in VMD decomposition. Sub-step S23: Extract dynamic signal features, including time-domain features, frequency-domain features, and energy entropy, from the modes after decomposition of drilling vibration and acoustic emission signals. The time-domain features are as follows: ， ； The frequency domain characteristics are as follows: ， ； The energy entropy is as follows: ； Wherein, RMS represents the mode after VMD decomposition. The root mean square value, Peak is the mode after VMD decomposition. The peak value is given by N, where N is the number of signal sampling points for the k-th mode after VMD decomposition, FC is the centroid frequency, and MSF is the mean square frequency. Modes after VMD decomposition The power spectral density, The instantaneous frequency value in the frequency domain. For the k-th VMD mode energy, Let H be the energy percentage of the k-th VMD mode, and H be the energy entropy.

3. The in-situ rock strata identification method based on physical information machine learning according to claim 1, characterized in that: The fused feature vector in step S3 is obtained through the following sub-steps: Sub-step S31: Feature dimension organization. The standardized multi-source signal parameter features are organized into a one-dimensional vector with four dimensions: drilling pressure, drill rod torque, drill rod rotation speed, and drilling speed. The dynamic signal features are organized into a K2×5-dimensional vector, where K2 is the VMD decomposition mode number, which includes the five features after mode decomposition. Sub-step S32: Dimensional alignment, dividing the samples for feature fusion according to the time window of synchronous acquisition of multi-source signal parameters; Sub-step S33: Redundancy removal. Pearson correlation analysis is used. If the correlation coefficient between dynamic signal features exceeds the set value, dynamic signal features with greater mutual information to the rock layer strength label are retained. Sub-step S34: Vector concatenation. The standardized multi-source parameter features and the de-redundant dynamic signal features are concatenated in sequence to form a unified fused feature vector.

4. The in-situ rock strata identification method based on physical information machine learning according to claim 1, characterized in that: The training process of the physical information machine learning model is as follows: Step K1: Obtain multiple sets of historical drilling parameters, including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals. Label each set of historical drilling parameters with corresponding laboratory standard rock mass parameter labels, including the strength prediction value of the drilled rock strata and the rock mass category. Divide the multiple sets of historical drilling parameters into training set, validation set, and test set. Step K2: Perform two mode decompositions on the drilling vibration and acoustic emission signals of each set of historical drilling parameters using CEEMDAN and VMD to obtain dynamic signal characteristics related to rock mass fracture. Step K3: After standardizing and feature processing the remaining historical drilling parameters, combine them with the dynamic signal features of the drilling vibration and acoustic emission signals in the same group to form a fused feature vector; Step K4: Input the fused feature vector of historical drilling parameters from the training set into the physical information machine learning model for training. The physical information machine learning model uses an LSTM network as its backbone and constructs a physical inversion equation between drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and rock strength based on the Buschnesk elastic half-space theory. ； Wherein, UCS is the calculated rock strength value obtained from the inversion, WOB is the drilling pressure, RPM is the drill pipe rotation speed, ROP is the drilling speed, and Torque is the drill pipe torque. , These are physical proportionality coefficients, representing the contribution weights of the product of drilling pressure and drill pipe rotation speed, and drill pipe torque to the rock strength inversion, respectively. This is the intercept term, used to correct systematic deviations between the model and actual operating conditions; The loss function for embedding the physical inversion equation into the physical information machine learning model is as follows: ； in, It is data and The driving loss, It is the actual value of rock layer strength in the laboratory standard rock mass parameter label. It is the rock strength prediction value from the physical information machine learning model. It is the physical constraint loss, that is, the calculated rock strength value UCS obtained through the physical inversion equation. It is a trade-off parameter; The physical information machine learning model outputs the predicted strength value of the drilled rock strata and the rock mass type. Step K5: Using the Adam optimizer, input the validation set and test set into the physical information machine learning model for validation and testing, and obtain the trained physical information machine learning model.

5. A method for in-situ identification of frozen strata in cold regions through drilling, characterized by: The in-situ rock strata identification method based on physical information machine learning according to any one of claims 1-4 is used to identify the rock strata strength value and rock mass type during the drilling process in frozen strata in cold regions.

6. An experimental apparatus for implementing the in-situ identification method for drilling through frozen strata in cold regions as described in claim 5, characterized in that... This includes the drilling system, the cold-region environment simulation chamber, and the multi-source drilling information acquisition system; The drilling system uses a hydraulic rock drilling rig; The cold environment simulation box is fixed relative to the drilling system. The sample assembly after freeze-thaw is fixed inside. The hydraulic rock drilling machine of the drilling system is connected to the drill rod and passes through one side of the cold environment simulation box to drill the sample assembly. A pressure system is set outside the cold environment simulation box to clamp and fix the sample assembly. The pressure system has a pressure head that applies pressure to the sample assembly in the normal and tangential directions from the drilling direction. The multi-source drilling information acquisition system includes: A pressure sensor is installed on the impact mechanism of the hydraulic rock drilling rig to collect the drilling pressure during the drilling process. Torque and speed sensors are installed at the connection between the rotary mechanism and the drill rod of the hydraulic rock drilling rig to collect the drill rod torque and drill rod speed during the drilling process of the hydraulic rock drilling rig. A displacement sensor is installed on the hydraulic rock drilling rig to collect the moving displacement of the hydraulic rock drilling rig during the drilling process and obtain the drilling speed. Vibration and acoustic emission sensors are installed on the pressure head of the pressurization system to collect drilling vibration and acoustic emission signals generated by the hydraulic rock drilling rig during the rock breaking process of the sample assembly. The multi-source drilling information acquisition system synchronously acquires multi-source signal parameters such as drilling pressure, drill rod torque, drill rod speed, drilling speed, and drilling vibration and acoustic emission signals during the drilling process of the hydraulic rock drilling rig. The in-situ identification method for drilling frozen strata in cold regions as described in claim 5 is used to obtain the rock strength value and rock mass type of the sample assembly.

7. The experimental apparatus according to claim 6, characterized in that: The sample assembly includes a rock mass sample and clamping plates. The two clamping plates clamp the rock mass sample in the drilling direction by means of a tie rod, and drilling holes for the drill rod to pass through are provided on the clamping plates. After the sample assembly is placed in the cold environment simulation chamber, the pressure head of the pressurization system clamps the side of the rock mass sample symmetrically from the normal and tangential directions of the drilling direction through the pads. The vibration and acoustic emission sensors are arranged on the pads.

8. The experimental apparatus according to claim 6, characterized in that: The inner wall of the cold environment simulation chamber is insulated, and the side wall is provided with a refrigerant injection hole for injecting refrigerant into the chamber. A temperature sensor is installed on the outer wall of the chamber and connected to a temperature display outside the chamber.

9. The experimental apparatus according to claim 6, characterized in that: It also includes a freeze-thaw cycle chamber for conducting cyclic freeze-thaw simulation experiments on the sample assembly.

10. An intelligent drilling control method, characterized in that: During the drilling process, the drilling system collects multi-source signal parameters in real time, including drilling pressure, drill rod torque, drill rod rotation speed, drilling speed, and drilling vibration and acoustic emission signals. It uses the in-situ rock strata identification method based on physical information machine learning as described in any one of claims 1-4 to obtain the real-time rock strata strength value and rock mass category. The drilling system adjusts the matching drilling parameters according to the real-time rock strata strength value and rock mass category to adapt.

Images