A method and system for sensing the material and structure of ore rock while drilling in coordination with touch

Abstract

The application discloses a kind of ore rock material quality and structure and system for drilling collaborative touch sensing method cooperates with.The method of the present application takes multitask time convolution network as core prediction model, and the physical correlation established by rock mechanics theory and drilling mechanism is used as constraint basis, and forms the collaborative sensing framework of "data-driven prediction + physical mechanism constraint + result collaborative optimization".The physical mechanism constraint participates in network parameter learning in the model training stage, limits model output to meet the preset physical mechanics law, the correlation between parameters and the structure surface response law;In the model online inference stage, based on the physical correlation rule, the consistency of the output results is checked and optimized, and the ore rock mechanics parameter prediction result, the structure surface identification result and the positioning result are obtained.The present application realizes in-situ, real-time and continuous sensing during drilling, overcomes the problems of traditional hysteresis and high cost, and provides key technical support for mine production and resource evaluation.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary fields of geotechnical engineering, mining, and artificial intelligence, and specifically relates to a method and system for collaborative penetration testing of mineral and rock materials and structures during drilling. Background Technology

[0002] With the increasing depletion of shallow mineral resources, mineral resource extraction and underground engineering construction are developing towards deeper deposits. The mechanical properties and structural parameters of deep mineral rocks are important indicators for preventing geological disasters in deep engineering projects, optimizing construction techniques, and improving mining efficiency.

[0003] Traditional methods for acquiring ore and rock mechanical parameters rely on laboratory sampling and testing, which suffers from drawbacks such as being time-consuming, destructive, and unable to reflect real-time field conditions. Existing inversion methods often focus on single parameters (such as UCS), failing to achieve simultaneous inversion of multiple parameters and neglecting the mechanical correlation between parameters, leading to one-sided construction guidance. Traditional methods for identifying ore and rock structural surfaces rely on offline means such as geological exploration and borehole photography, resulting in problems such as delayed positioning, high costs, and the inability to provide real-time warnings during drilling. Existing structural surface identification technologies are not integrated with mechanical parameter inversion, making it difficult to establish a synergistic relationship between "structural surfaces and mechanical parameters," thus failing to accurately guide adjustments to construction parameters. Existing multi-parameter inversion methods often employ purely data-driven models (such as LSTM and CNN), lacking constraints from the physical mechanisms of ore and rock drilling, resulting in poor model generalization and weak interpretability. While TCN networks are suitable for time-series data processing, they are not customized for the collaborative needs of multi-parameter ore and rock inversion and structural surface identification, and they do not address the model optimization problem under "multi-task (inversion + identification)" scenarios.

[0004] Therefore, there is an urgent need in this field to develop an online inversion method for mineral and rock multivariate parameters and structural surface identification and localization that integrates physical mechanisms and temporal convolutional networks (TCNs), which can realize in-situ real-time monitoring and inversion of mineral and rock characteristic parameters and accurate localization and perception of structural surfaces. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for collaborative penetration testing of rock material and structure during drilling. By constructing a multi-task TCN network of "shared feature extraction - branch task output", it integrates multi-dimensional physical mechanism parameters and drilling time-series data to achieve online collaborative inversion of multi-dimensional mechanical parameters such as rock strength, elastic modulus, cohesion and internal friction angle, and identification and positioning of structural surfaces, thereby significantly improving the real-time performance and reliability of mine rock mass quality evaluation.

[0006] This invention provides a method for co-sensing the material and structure of mineral rocks during drilling, comprising the following steps:

[0007] S1. Obtain the drilling time series parameters, and perform data cleaning and multi-scale time series window feature extraction on the drilling time series parameters to obtain the input feature dataset;

[0008] S2. An initial mineral and rock material parameter inversion and structure identification and localization model is constructed based on a multi-task temporal convolutional network. The initial mineral and rock material parameter inversion and structure identification and localization model includes a shared temporal feature extraction layer, a material parameter inversion branch, a structure surface identification branch, and a structure surface localization branch. The initial mineral and rock material parameter inversion and structure identification and localization model also includes an autonomously derived theoretical model. Among them, the constraints of the autonomously derived theoretical model are not directly used as independent neural network branches to participate in forward prediction, but are used to construct physical constraint loss in the training stage and for execution result verification and collaborative optimization in the inference stage.

[0009] S3. Construct a theoretical consistency loss term based on the self-derived theoretical model, and add the theoretical consistency loss term to the total loss function of the model. Use the input feature dataset as the training dataset to train the initial mineral and rock material parameter inversion and structure identification and localization model to obtain the mineral and rock material parameter inversion and structure identification and localization model.

[0010] S4. Based on the inversion model of mineral and rock material parameters and the structural identification and localization model, reasoning is performed to obtain the initial prediction results of mineral and rock material parameters, the structural surface identification results and the initial localization results of structural surfaces;

[0011] S5. Based on the self-derived theoretical model, the initial prediction results of the mineral and rock material parameters are checked for consistency in theoretical estimation, and based on the structural response characteristics, the structural surface identification results and the initial structural surface positioning results are checked for consistency in structural response; at the same time, the confidence of the model inference results is evaluated.

[0012] S6. Based on the theoretical estimation consistency verification results, the structural response consistency verification results, and the confidence assessment results, the initial prediction results of the mineral and rock material parameters, the structural surface identification results, and the initial positioning results of the structural surface are collaboratively corrected to obtain the final mineral and rock material parameter results, structural surface identification results, and structural surface positioning results, thus completing the collaborative probing and sensing of mineral and rock material and structure.

[0013] The mineral and rock material parameters include at least one or more of the following: rock strength, elastic modulus, cohesion, and internal friction angle.

[0014] In step S1, the data cleaning includes outlier filtering and missing value linear interpolation completion; the multi-scale time window feature extraction includes extracting 6-dimensional basic statistical features, 4-dimensional dynamic derived features, and 2-dimensional mutation features using a combination of short and long windows through a dual sliding window.

[0015] 2. The dual sliding window specifically employs a combination of a 5-second short window and a 10-second long window; the six-dimensional basic statistical features include the mean torque within the window. In-window variance of torque The mean value of propulsion within the window Variance within the window of propulsion The mean value of rotational speed within the window The mean of propulsion speed within the window ;

[0016] The four-dimensional dynamic derived features include rock-breaking energy ratio. Normalized transition width Torque Thrust-rate-of-drilling gradient ;in, This represents the average thrust within the previous window. This represents the average propagation speed within the previous window; the normalized transition width is the depth feature covered by the interface response as it increases from 10% to 90%, used to characterize the geometric expansion range of the interface perturbation. A wider normalized transition width indicates a more fragmented structural surface, expressed by the following formula: ;in, d represents the geometric depth features covered by the interface response as it increases from 10% to 90%; d represents the geometric thickness of the cutting tooth.

[0017] The two-dimensional abrupt change features include the interface energy jump intensity and the load jump intensity; wherein, the interface energy jump intensity is the increase in mechanical specific energy of the region at the 0.75 upper quantile within the interface window compared to the mechanical specific energy of the stable section of the upper rock. A large interface jump intensity indicates a complete structural surface, and is expressed by the following formula: ;in, The mechanical specific energy of the region at the upper quantile of 0.75 within the interface window; The mechanical specific energy of the stable section of the upper rock is given; the load abrupt change intensity is the increase in the average thrust within the interface window compared to the baseline thrust during the stable drilling stage of the upper rock. A small load amplification intensity indicates the presence of cementing material at the structural surface, expressed by the following formula: ;in, The average propulsion force within the interface window; Baseline thrust for the stable drilling stage of the upper rock;

[0018] Among them, the normalized transition width is used to characterize the spatial influence range of structural interface disturbance, and the interface energy jump intensity and load jump intensity are used to characterize one or more of the following: structural interface response strength, interface integrity, fragmentation or cementation state.

[0019] In step S2, the initial ore and rock material parameter inversion and structure identification and localization model includes a shared temporal feature extraction layer and a branch task layer;

[0020] The shared temporal feature extraction layer is used to extract common temporal features of material and structure from the input feature data;

[0021] The branch task layer includes a multi-parameter inversion branch, a structural surface identification branch, and a structural surface location branch. The parameter inversion branch is processed through several cascaded fully connected layers and then linearly activated, outputting multi-dimensional mechanical parameters including rock strength, elastic modulus, and cohesion-internal friction angle. The structural surface identification branch is processed through fully connected layers and then activated by the Sigmoid function, outputting a judgment result on whether the structural surface exists. The structural surface location branch is processed through fully connected layers and then linearly activated, outputting the depth, location coordinates, and structural surface feature parameters of the structural surface. The structural surface feature parameters include at least one or more of the following: structural surface width, structural surface dip angle, and structural surface type.

[0022] The output features of the shared temporal feature extraction layer are simultaneously input into the three branches of the branch task layer, and the three branches of the branch task layer feed back the gradients to the feature extraction layer through the backpropagation process.

[0023] The shared temporal feature extraction layer is composed of multiple temporal convolutional network modules stacked in series. Each temporal convolutional network module includes a causal convolutional unit, a batch normalization unit, a random deactivation regularization unit, a shallow connection unit, and a residual connection unit connected in series. Each task branch shares the output features of the shared temporal feature extraction layer and jointly constrains the shared feature learning process through their respective task losses to achieve collaborative optimization between mineral material perception and structural response perception.

[0024] In step S3, an autonomous derivation theoretical model is introduced, which is established based on the theoretical mapping relationship between drilling force relationship, cutting response relationship of cutting teeth and ore material parameters;

[0025] The self-derived theoretical model is used to calculate theoretical estimates of ore and rock material parameters based on drilling parameters, and to construct a theoretical consistency loss term based on the theoretical estimates.

[0026] The autonomously derived theoretical model does not participate directly in forward prediction as an independent neural network branch. Instead, it is used to construct a theoretical consistency loss term during the training phase and to perform theoretical estimation consistency verification and collaborative correction on the initial prediction results of mineral and rock material parameters during the inference phase. This improves the physical rationality and generalization ability of the model's prediction results, ensures that the inversion data conforms to the basic physical mechanism conditions, and avoids the black box problem of learning and training based on pure data.

[0027] The self-derived theoretical model includes theoretical estimation formulas for calculating elastic modulus, cohesion, internal friction angle, and uniaxial compressive strength of rock;

[0028] The theoretical estimation formula is established based on drilling propulsion force, torque, rotational speed, drilling speed, and drill bit structural parameters;

[0029] in:

[0030] Theoretical formula for calculating elastic modulus: ;

[0031] Theoretical formula for calculating cohesion-angle of internal friction:

[0032] ;

[0033] Theoretical formula for calculating the uniaxial compressive strength of rock: ;in, Where T is the drilling propulsion force, N is the drilling torque, V is the rotational speed, and E is the rock elastic modulus. For the Poisson's ratio of rocks, The radius of the drill bit's geometric rotation center; The internal friction angle between the rock compression and fracture zone and the intact rock. The internal friction angle of the rock. Let m be the coefficient of friction between the drill bit and the rock contact surface, and m be the number of drill bit teeth. S is the comprehensive characteristic coefficient of the cutting tooth, and S is the comprehensive cutting parameter; This refers to the length of a single tooth.

[0034] Comprehensive characteristic coefficient of cutting teeth With the angle of entry of the cutting tooth Cross-sectional shape factor and the coefficient of friction between the drill bit and the rock contact surface Related, specific calculation formula: ;

[0035] Combined cutting parameters S and tooth entry angle Coefficient of friction of rock contact surface and the internal friction angle between the rock compression fracture zone and the intact rock. Regarding the specific calculation formula: .

[0036] In step S3, the total loss function of the model Express it using the following formula: ;in, Huber loss for the parameter inversion branch; The Huber loss weighting coefficient; Cross-entropy loss for identifying branches on structural surfaces; These are the weighting coefficients for the cross-entropy loss. Mean square error loss for the structural surface positioning branches; The mean squared error loss weighting coefficient; This is a physical constraint regularization term that includes the loss of consistency between parameter theories, the mechanical correlation deviation between parameters, and the correlation deviation between structural surfaces and parameters; The weight coefficient for the theoretical consistency regularization term.

[0037] Step S5 includes the following steps:

[0038] The initial prediction results are subjected to consistency verification; the consistency verification includes parameter physical range consistency verification, physical mechanism consistency verification, and structural surface response consistency verification. When the consistency verification fails, the model is retrained and the parameters are adjusted before prediction is performed.

[0039] Then, a confidence assessment is performed; the confidence assessment includes: first, calculating the model output confidence, physical consistency confidence, and temporal stability confidence; then, based on the model output confidence, physical consistency confidence, and temporal stability confidence, calculating the comprehensive confidence of the ore-rock mechanical parameter simulation results; and based on the identification branch output probability, structural surface response consistency confidence, and temporal stability confidence, calculating the comprehensive confidence of the structural surface task; finally, calculating the total comprehensive confidence based on the comprehensive confidence of the ore-rock mechanical parameter simulation results and the comprehensive confidence of the structural surface task.

[0040] Step S6 includes the following steps:

[0041] After the confidence assessment is completed, the consistency check and confidence assessment results are used to determine whether to perform collaborative correction in order to obtain the final prediction result.

[0042] After the collaborative correction is completed, the final results of the ore and rock mechanical parameters, structural surface identification results, and structural surface positioning results are output, along with the corresponding comprehensive confidence scores. These results are used for on-site engineering prompts, risk warnings, and construction parameter adjustments, thus completing the collaborative probing and perception of ore and rock material and structure.

[0043] The consistency check specifically includes:

[0044] The physical parameter range consistency verification is used to determine whether the initial prediction results of the mineral and rock mechanical parameters and the initial results of the structural surface positioning are within the reasonable range of the project; wherein, the mineral and rock mechanical parameters include at least one or more of uniaxial compressive strength, elastic modulus, cohesion and internal friction angle; when any output result exceeds the corresponding reasonable range, it is determined that the result has insufficient range consistency;

[0045] The physical mechanism consistency check is used to calculate the deviation between the initial predicted value of the model and the output value of the physical theory model. The deviation is characterized by absolute deviation, relative deviation, normalized deviation, or a combination thereof. In the physical mechanism consistency check, for the i-th ore-rock mechanical parameter, the physical consistency deviation is characterized by the absolute deviation, relative deviation, normalized deviation, or a combination thereof between the model predicted value and the physical estimate, and is expressed by the following formula: ;in, The physical consistency deviation of the i-th parameter; The observed value of the i-th parameter output by the model; Theoretical physical estimates of the corresponding parameters; To prevent positive numbers with a denominator of zero;

[0046] The structural surface response consistency check is used to determine the degree of matching between the initial results of structural surface identification and the initial results of structural surface positioning and the abrupt changes in drilling parameters within the corresponding depth range; the abrupt changes in drilling parameters include at least one or more of the following: interface energy jump intensity, load jump intensity, vibration acceleration change, and multi-parameter joint anomaly.

[0047] Preferably, a structural surface response consistency index is constructed. The degree of matching is quantitatively characterized by the structural surface response consistency index, which is obtained by normalizing and weighting the sum of various parameter mutation characteristics within the target depth range.

[0048] ;in, To normalize the amplitude of the interface energy jump intensity, The magnitude of the load abrupt change in intensity. Normalized vibration response amplitude, To normalize the intensity magnitude weight of interface energy jumps, As the load mutation intensity amplitude weight, For the normalized vibration response amplitude weights, and ;

[0049] When the structural surface response consistency index is lower than the preset response threshold hour( The result is determined to have insufficient structural surface response consistency; when the structural surface response consistency index is greater than or equal to the preset response threshold ( The result is determined to meet the structural surface response consistency requirements.

[0050] The confidence assessment specifically includes:

[0051] The model output confidence score is used to characterize the model's degree of confidence in the current output result. For the structural surface identification result, the model output confidence score can be determined based on the probability value of the classification branch output. For the ore and rock mechanical parameter inversion result and the structural surface location result, the model output confidence score can be determined based on the regression output stability, the fluctuation of adjacent time window prediction, the model integration dispersion, the distribution of Monte Carlo random inactivation results, or a combination thereof.

[0052] The physical consistency confidence score is determined based on the deviation between the model prediction and the physical estimate; the physical consistency confidence score can be obtained by monotonically decreasing the physical consistency deviation. For the i-th parameter, its physical consistency confidence score is expressed using the following formula: ;in, Let be the physical consistency confidence level of the i-th parameter; This is the adjustment coefficient for the i-th parameter;

[0053] The time series stability confidence is determined based on the continuity, smoothness, and volatility of the prediction results within adjacent time windows or adjacent drilling depth intervals. If the output of the current window is continuous with the outputs of the preceding and following adjacent windows and the volatility is small, the corresponding time series stability confidence is high; otherwise, it is low.

[0054] The overall confidence level of the inversion results of the ore and rock mechanical parameters is obtained by fusing the model output confidence level, physical consistency confidence level, and temporal stability confidence level, and is expressed by the following formula: ;in, The overall confidence level of the inversion results of ore and rock mechanical parameters; Output confidence scores for the model; For physical consistency confidence; The confidence level for time series stability; Output confidence weights for the model. For the weight of physical consistency confidence, The weights for the confidence scores of time series stability are: ;

[0055] The overall confidence score for the structural surface task is obtained by fusing the probability of the identification branch output, the confidence score of the structural surface response consistency, and the confidence score of the temporal stability, and is expressed by the following formula: ;in, The overall confidence level for the structural surface task; The probability confidence level for identifying branches of the structural surface; The confidence level for the consistency of the structural surface response; The confidence level for time series stability; The probability confidence weights for identifying branches on the structural surface are used. Weights for the confidence level of structural surface response consistency. To calculate the temporal stability confidence weights for the overall confidence of the structural surface task, and ;

[0056] The overall confidence score is calculated based on the overall confidence score of the ore-rock mechanical parameter inversion results and the overall confidence score of the structural surface task, using the following formula: ;in, The overall confidence level; As the first fusion weight, As the second fusion weight, and ;

[0057] All the above weights are determined using any one of the following methods: preset empirical weights, optimal weights determined based on the validation set, or adaptively adjusted weights; or they are determined jointly based on the prediction error, positioning deviation, physical consistency deviation, and recognition accuracy during the sample validation process.

[0058] All the confidence levels mentioned above are represented by normalized values ​​between 0 and 1.

[0059] The coordinated correction specifically refers to:

[0060] When the consistency deviation exceeds the preset threshold, the overall confidence level is lower than the preset threshold, the output result exceeds the reasonable range of the project, or the output fluctuation of adjacent time windows is abnormal, necessary collaborative correction is triggered.

[0061] The collaborative correction includes one or more of the following: correction of ore and rock mechanical parameter results, correction of structural surface identification results, and correction of structural surface positioning results;

[0062] The correction of the ore and rock mechanical parameters is achieved by weighting and fusing the initial model predictions and physical estimates based on the comprehensive confidence level, resulting in the corrected parameter results: ; This is the final output result for the i-th parameter; The observed value of the i-th parameter output by the model; Theoretical physical estimates of the corresponding parameters; The final fusion weight is the i-th parameter, which is determined based on the overall confidence level of the corresponding parameter. When the overall confidence level is high, the weight of the model prediction value is increased; when the overall confidence level is low, the weight of the physical estimate value is increased.

[0063] The collaborative correction of the structural surface results includes correcting the structural surface identification results and the structural surface positioning results; wherein, the correction of the structural surface identification results combines the consistency of the structural surface response and the continuity of adjacent time windows to perform smoothing optimization or threshold correction, and the correction of the structural surface positioning results combines the mapping relationship between the peak position of parameter mutations in adjacent depth intervals and the drilling depth to perform local fine-tuning.

[0064] The aforementioned preset thresholds include one or more of the following: physical consistency deviation threshold, comprehensive confidence threshold, structural surface response consistency threshold, and output fluctuation threshold; the thresholds can be determined based on prior engineering experience, or based on statistical results of geological history samples, cross-validation results, or field test calibration results.

[0065] The present invention also provides a system for implementing the aforementioned method for collaborative penetration sensing of ore and rock material and structure based on drilling monitoring, comprising a data acquisition and preprocessing module, a model building module, a model training module, a model inference module, a consistency verification module, and a collaborative correction module;

[0066] The data acquisition and preprocessing module obtains the drilling time series parameters, performs data cleaning and multi-scale time series window feature extraction on the drilling time series parameters to obtain the input feature dataset, and uploads the data to the model building module;

[0067] The model building module constructs an initial model for inverting mineral and rock material parameters and identifying and locating structures based on a multi-task temporal convolutional network, and uploads the data to the model training module.

[0068] The model training module constructs a theoretical consistency loss term based on the self-derived theoretical model and adds the theoretical consistency loss term to the model's total loss function. The input feature dataset is used as the training dataset to train the initial mineral and rock material parameter inversion and structure identification and localization model, thereby obtaining the mineral and rock material parameter inversion and structure identification and localization model. The data is then uploaded to the model inference module.

[0069] The model inference module performs inference based on the mineral and rock material parameter inversion and structure identification and positioning model to obtain the initial prediction results of mineral and rock material parameters, the structure surface identification results and the initial positioning results of structure surface, and uploads the data to the consistency verification module.

[0070] The consistency verification module performs theoretical estimation consistency verification on the initial prediction results of the mineral and rock material parameters based on the self-derived theoretical model, and performs structural response consistency verification on the structural surface identification results and the initial structural surface positioning results based on the structural response characteristics; at the same time, it evaluates the confidence of the model inference results and uploads the data to the collaborative correction module.

[0071] The collaborative correction module performs collaborative correction on the initial prediction results of the mineral and rock material parameters, the structural surface identification results, and the initial location results of the structural surface based on the consistency verification results of theoretical estimation, the consistency verification results of structural response, and the confidence assessment results, so as to obtain the final mineral and rock material parameter results, structural surface identification results, and structural surface location results, thus completing the collaborative probing and sensing of mineral and rock material and structure.

[0072] This invention discloses a method and system for collaborative penetration testing of rock material and structure during drilling. By constructing a multi-task TCN network of "shared feature extraction - branch task output", it integrates multi-dimensional physical mechanism parameters and drilling time-series data to achieve online collaborative inversion of multivariate parameters of rock strength, elastic modulus, cohesion-internal friction angle force and structural surface identification and positioning, which significantly improves the real-time performance and reliability of mine rock mass quality evaluation. Attached Figure Description

[0073] Figure 1 This is a schematic flowchart of the method of the present invention;

[0074] Figure 2 This is a schematic diagram of the structure of the ore and rock material parameter inversion and structure identification and positioning model in an embodiment of the present invention;

[0075] Figure 3 This is a schematic diagram of the system structure of the present invention;

[0076] Figure 4 This is a schematic diagram of actual application construction in an embodiment of the present invention; the markings are: 1, drill rod; 2, speed-torque composite sensor; 3, stress sensor; 4, wire displacement sensor;

[0077] Figure 5 This is a schematic diagram illustrating the generation of original data and inversion data in an embodiment of the present invention. Detailed Implementation

[0078] This invention provides a method for co-probing and sensing the material and structure of mineral rocks while drilling, the flowchart of which is shown below. Figure 1 As shown, it includes the following steps:

[0079] S1. Obtain the drilling time series parameters, and perform data cleaning and multi-scale time series window feature extraction on the drilling time series parameters to obtain the input feature dataset;

[0080] In step S1, the data cleaning includes outlier filtering and missing value linear interpolation completion; the multi-scale time window feature extraction includes extracting 6-dimensional basic statistical features, 4-dimensional dynamic derived features, and 2-dimensional mutation features using a combination of short and long windows through a dual sliding window.

[0081] The dual sliding window specifically employs a combination of a 5-second short window and a 10-second long window; the six-dimensional basic statistical features include the mean torque within the window. In-window variance of torque The mean value of propulsion within the window Variance within the window of propulsion Average value of rotational speed within the window The mean of propulsion speed within the window ;

[0082] The four-dimensional dynamic derived features include rock-breaking energy ratio. Normalized transition width Torque Thrust-rate-of-drilling gradient ;in, This represents the average thrust within the previous window. This represents the average propagation speed within the previous window; the normalized transition width is the depth feature covered by the interface response as it increases from 10% to 90%, used to characterize the geometric expansion range of the interface perturbation. A wider normalized transition width indicates a more fragmented structural surface, expressed by the following formula: ;in, d represents the geometric depth features covered by the interface response as it increases from 10% to 90%; d represents the geometric thickness of the cutting tooth.

[0083] The two-dimensional abrupt change features include the interface energy jump intensity and the load jump intensity; wherein, the interface energy jump intensity is the increase in mechanical specific energy of the region at the 0.75 upper quantile within the interface window compared to the mechanical specific energy of the stable section of the upper rock. A large interface jump intensity indicates a complete structural surface, and is expressed by the following formula: ;in, The mechanical specific energy of the region at the upper quantile of 0.75 within the interface window; The mechanical specific energy of the stable section of the upper rock is given; the load abrupt change intensity is the increase in the average thrust within the interface window compared to the baseline thrust during the stable drilling stage of the upper rock. A small load amplification intensity indicates the presence of cementing material at the structural surface, expressed by the following formula: ;in, The average propulsion force within the interface window; Baseline thrust for the stable drilling stage of the upper rock;

[0084] Among them, the normalized transition width is used to characterize the spatial influence range of structural interface disturbance, and the interface energy jump intensity and load jump intensity are used to characterize one or more of the following: structural interface response strength, interface integrity, fragmentation or cementation state.

[0085] S2. An initial mineral and rock material parameter inversion and structure identification and localization model is constructed based on a multi-task temporal convolutional network. The initial mineral and rock material parameter inversion and structure identification and localization model includes a shared temporal feature extraction layer, a material parameter inversion branch, a structure surface identification branch, and a structure surface localization branch. The initial mineral and rock material parameter inversion and structure identification and localization model also includes an autonomously derived theoretical model. Among them, the constraints of the autonomously derived theoretical model are not directly used as independent neural network branches to participate in forward prediction, but are used to construct physical constraint loss in the training stage and for execution result verification and collaborative optimization in the inference stage.

[0086] In step S2, the initial ore and rock material parameter inversion and structure identification and localization model includes a shared temporal feature extraction layer and a branch task layer, as shown in the schematic diagram below. Figure 2 As shown;

[0087] The shared temporal feature extraction layer is used to extract common temporal features of material and structure from the input feature data;

[0088] The branch task layer includes a multi-parameter inversion branch, a structural surface identification branch, and a structural surface location branch. The parameter inversion branch is processed through several cascaded fully connected layers and then linearly activated, outputting multi-dimensional mechanical parameters including rock strength, elastic modulus, and cohesion-internal friction angle. The structural surface identification branch is processed through fully connected layers and then activated by the Sigmoid function, outputting a judgment result on whether the structural surface exists. The structural surface location branch is processed through fully connected layers and then linearly activated, outputting the depth, location coordinates, and structural surface feature parameters of the structural surface. The structural surface feature parameters include at least one or more of the following: structural surface width, structural surface dip angle, and structural surface type.

[0089] The output features of the shared temporal feature extraction layer are simultaneously input into the three branches of the branch task layer, and the three branches of the branch task layer feed back the gradients to the feature extraction layer through the backpropagation process.

[0090] The shared temporal feature extraction layer is composed of multiple temporal convolutional network modules stacked in series. Each temporal convolutional network module includes a causal convolutional unit, a batch normalization unit, a random deactivation regularization unit, a shallow connection unit, and a residual connection unit connected in series. Each task branch shares the output features of the shared temporal feature extraction layer and jointly constrains the shared feature learning process through their respective task losses to achieve collaborative optimization between mineral material perception and structural response perception.

[0091] Specifically, the network structure diagram of the mineral and rock material parameter inversion and structure identification and localization model is as follows: Figure 3 As shown.

[0092] S3. Construct a theoretical consistency loss term based on the self-derived theoretical model, and add the theoretical consistency loss term to the total loss function of the model. Use the input feature dataset as the training dataset to train the initial mineral and rock material parameter inversion and structure identification and localization model to obtain the mineral and rock material parameter inversion and structure identification and localization model.

[0093] In step S3, an autonomous derivation theoretical model is introduced, which is established based on the theoretical mapping relationship between drilling force relationship, cutting response relationship of cutting teeth and ore material parameters;

[0094] The self-derived theoretical model is used to calculate theoretical estimates of ore and rock material parameters based on drilling parameters, and to construct a theoretical consistency loss term based on the theoretical estimates.

[0095] The autonomously derived theoretical model does not participate directly in forward prediction as an independent neural network branch. Instead, it is used to construct a theoretical consistency loss term during the training phase and to perform theoretical estimation consistency verification and collaborative correction on the initial prediction results of mineral and rock material parameters during the inference phase. This improves the physical rationality and generalization ability of the model's prediction results, ensures that the inversion data conforms to the basic physical mechanism conditions, and avoids the black box problem of learning and training based on pure data.

[0096] The self-derived theoretical model includes theoretical estimation formulas for calculating elastic modulus, cohesion, internal friction angle, and uniaxial compressive strength of rock;

[0097] The theoretical estimation formula is established based on drilling propulsion force, torque, rotational speed, drilling speed, and drill bit structural parameters;

[0098] in:

[0099] Theoretical formula for calculating elastic modulus: ;

[0100] Theoretical formula for calculating cohesion-angle of internal friction:

[0101] ;

[0102] Theoretical formula for calculating the uniaxial compressive strength of rock: ;in, Where T is the drilling propulsion force, N is the drilling torque, V is the rotational speed, and E is the rock elastic modulus. For the Poisson's ratio of rocks, The radius of the drill bit's geometric rotation center; The internal friction angle between the rock compression and fracture zone and the intact rock. The internal friction angle of the rock. Let m be the coefficient of friction between the drill bit and the rock contact surface, and m be the number of drill bit teeth. S is the comprehensive characteristic coefficient of the cutting tooth, and S is the comprehensive cutting parameter; This refers to the length of a single tooth.

[0103] Comprehensive characteristic coefficient of cutting teeth With the angle of entry of the cutting tooth Cross-sectional shape factor and the coefficient of friction between the drill bit and the rock contact surface Related, specific calculation formula: ;

[0104] Combined cutting parameters S and tooth entry angle Coefficient of friction of rock contact surface and the internal friction angle between the rock compression fracture zone and the intact rock. Regarding the specific calculation formula: .

[0105] In step S3, the total loss function of the model Express it using the following formula: ;in, Huber loss for the parameter inversion branch; The Huber loss weighting coefficient; Cross-entropy loss for identifying branches on structural surfaces; These are the weighting coefficients for the cross-entropy loss. Mean square error loss for locating branches of the structural surface; The mean squared error loss weighting coefficient; This is a physical constraint regularization term that includes the loss of consistency between parameter theories, the mechanical correlation deviation between parameters, and the correlation deviation between structural surfaces and parameters; The weight coefficient for the theoretical consistency regularization term.

[0106] S4. Based on the inversion model of mineral and rock material parameters and the structural identification and localization model, reasoning is performed to obtain the initial prediction results of mineral and rock material parameters, the structural surface identification results and the initial localization results of structural surfaces;

[0107] S5. Based on the self-derived theoretical model, the initial prediction results of the mineral and rock material parameters are checked for consistency in theoretical estimation, and based on the structural response characteristics, the structural surface identification results and the initial structural surface positioning results are checked for consistency in structural response; at the same time, the confidence of the model inference results is evaluated.

[0108] Step S5 includes the following steps:

[0109] The initial prediction results are subjected to consistency verification; the consistency verification includes parameter physical range consistency verification, physical mechanism consistency verification, and structural surface response consistency verification. When the consistency verification fails, the model is retrained and the parameters are adjusted before prediction is performed.

[0110] Then, a confidence assessment is performed; the confidence assessment includes: first, calculating the model output confidence, physical consistency confidence, and temporal stability confidence; then, based on the model output confidence, physical consistency confidence, and temporal stability confidence, calculating the comprehensive confidence of the ore-rock mechanical parameter simulation results; and based on the identification branch output probability, structural surface response consistency confidence, and temporal stability confidence, calculating the comprehensive confidence of the structural surface task; finally, calculating the total comprehensive confidence based on the comprehensive confidence of the ore-rock mechanical parameter simulation results and the comprehensive confidence of the structural surface task.

[0111] S6. Based on the consistency verification results of theoretical estimation, the consistency verification results of structural response, and the confidence assessment results, the initial prediction results of the mineral and rock material parameters, the structural surface identification results, and the initial positioning results of the structural surface are collaboratively corrected to obtain the final mineral and rock material parameter results, structural surface identification results, and structural surface positioning results, thus completing the collaborative probing and sensing of mineral and rock material and structure.

[0112] The mineral and rock material parameters include at least one or more of the following: rock strength, elastic modulus, cohesion, and internal friction angle.

[0113] Step S6 includes the following steps:

[0114] After the confidence assessment is completed, the consistency check and confidence assessment results are used to determine whether to perform collaborative correction in order to obtain the final prediction result.

[0115] After the collaborative correction is completed, the final results of the ore and rock mechanical parameters, structural surface identification results, and structural surface positioning results are output, along with the corresponding comprehensive confidence scores. These results are used for on-site engineering prompts, risk warnings, and construction parameter adjustments, thus completing the collaborative probing and perception of ore and rock material and structure.

[0116] The consistency check specifically includes:

[0117] The physical parameter range consistency verification is used to determine whether the initial prediction results of the mineral and rock mechanical parameters and the initial results of the structural surface positioning are within the reasonable range of the project; wherein, the mineral and rock mechanical parameters include at least one or more of uniaxial compressive strength, elastic modulus, cohesion and internal friction angle; when any output result exceeds the corresponding reasonable range, it is determined that the result has insufficient range consistency;

[0118] The physical mechanism consistency check is used to calculate the deviation between the initial predicted value of the model and the output value of the physical theory model. The deviation is characterized by absolute deviation, relative deviation, normalized deviation, or a combination thereof. In the physical mechanism consistency check, for the i-th ore-rock mechanical parameter, the physical consistency deviation is characterized by the absolute deviation, relative deviation, normalized deviation, or a combination thereof between the model predicted value and the physical estimate, and is expressed by the following formula: ;in, The physical consistency deviation of the i-th parameter; The observed value of the i-th parameter output by the model; Theoretical physical estimates of the corresponding parameters; To prevent positive numbers with a denominator of zero;

[0119] The structural surface response consistency check is used to determine the degree of matching between the initial results of structural surface identification and the initial results of structural surface positioning and the abrupt changes in drilling parameters within the corresponding depth range; the abrupt changes in drilling parameters include at least one or more of the following: interface energy jump intensity, load jump intensity, vibration acceleration change, and multi-parameter joint anomaly.

[0120] Preferably, a structural surface response consistency index is constructed. The degree of matching is quantitatively characterized by the structural surface response consistency index, which is obtained by normalizing and weighting the sum of various parameter mutation characteristics within the target depth range.

[0121] ;in, To normalize the amplitude of the interface energy jump intensity, The magnitude of the load abrupt change in intensity. Normalized vibration response amplitude, To normalize the intensity magnitude weight of interface energy jumps, As the load mutation intensity amplitude weight, For the normalized vibration response amplitude weights, and ;

[0122] When the structural surface response consistency index is lower than the preset response threshold hour( The result is determined to have insufficient structural surface response consistency; when the structural surface response consistency index is greater than or equal to the preset response threshold ( The result is determined to meet the structural surface response consistency requirements.

[0123] The confidence assessment specifically includes:

[0124] The model output confidence score is used to characterize the model's degree of confidence in the current output result. For the structural surface identification result, the model output confidence score can be determined based on the probability value of the classification branch output. For the ore and rock mechanical parameter inversion result and the structural surface location result, the model output confidence score can be determined based on the regression output stability, the fluctuation of adjacent time window prediction, the model integration dispersion, the distribution of Monte Carlo random inactivation results, or a combination thereof.

[0125] The physical consistency confidence score is determined based on the deviation between the model prediction and the physical estimate; the physical consistency confidence score can be obtained by monotonically decreasing the physical consistency deviation. For the i-th parameter, its physical consistency confidence score is expressed using the following formula: ;in, Let be the physical consistency confidence level of the i-th parameter; This is the adjustment coefficient for the i-th parameter;

[0126] The time series stability confidence is determined based on the continuity, smoothness, and volatility of the prediction results within adjacent time windows or adjacent drilling depth intervals. If the output of the current window is continuous with the outputs of the preceding and following adjacent windows and the volatility is small, the corresponding time series stability confidence is high; otherwise, it is low.

[0127] The overall confidence level of the inversion results of the ore and rock mechanical parameters is obtained by fusing the model output confidence level, physical consistency confidence level, and temporal stability confidence level, and is expressed by the following formula: ;in, The overall confidence level of the inversion results of ore and rock mechanical parameters; Output confidence scores for the model; For physical consistency confidence; The confidence level for time series stability; Output confidence weights for the model. For the weight of physical consistency confidence, The weights for the confidence scores of time series stability are: ;

[0128] The overall confidence score for the structural surface task is obtained by fusing the probability of the identification branch output, the confidence score of the structural surface response consistency, and the confidence score of the temporal stability, and is expressed by the following formula: ;in, The overall confidence level for the structural surface task; The probability confidence level for identifying branches of the structural surface; The confidence level for the consistency of the structural surface response; The confidence level for time series stability; The probability confidence weights for identifying branches on the structural surface are used. Weights for the confidence level of structural surface response consistency. To calculate the temporal stability confidence weights for the overall confidence of the structural surface task, and ;

[0129] The overall confidence score is calculated based on the overall confidence score of the ore-rock mechanical parameter inversion results and the overall confidence score of the structural surface task, using the following formula: ;in, The overall confidence level; As the first fusion weight, As the second fusion weight, and ;

[0130] All the above weights are determined using any one of the following methods: preset empirical weights, optimal weights determined based on the validation set, or adaptively adjusted weights; or they are determined jointly based on the prediction error, positioning deviation, physical consistency deviation, and recognition accuracy during the sample validation process.

[0131] All the confidence levels mentioned above are represented by normalized values ​​between 0 and 1.

[0132] The coordinated correction specifically refers to:

[0133] When the consistency deviation exceeds the preset threshold, the overall confidence level is lower than the preset threshold, the output result exceeds the reasonable range of the project, or the output fluctuation of adjacent time windows is abnormal, necessary collaborative correction is triggered.

[0134] The collaborative correction includes one or more of the following: correction of ore and rock mechanical parameter results, correction of structural surface identification results, and correction of structural surface positioning results;

[0135] The correction of the ore and rock mechanical parameters is achieved by weighting and fusing the initial model predictions and physical estimates based on the comprehensive confidence level, resulting in the corrected parameter results: ; This is the final output result for the i-th parameter; The observed value of the i-th parameter output by the model; Theoretical physical estimates of the corresponding parameters; The final fusion weight is the i-th parameter, which is determined based on the overall confidence level of the corresponding parameter. When the overall confidence level is high, the weight of the model prediction value is increased; when the overall confidence level is low, the weight of the physical estimate value is increased.

[0136] The collaborative correction of the structural surface results includes correcting the structural surface identification results and the structural surface positioning results; wherein, the correction of the structural surface identification results combines the consistency of the structural surface response and the continuity of adjacent time windows to perform smoothing optimization or threshold correction, and the correction of the structural surface positioning results combines the mapping relationship between the peak position of parameter mutations in adjacent depth intervals and the drilling depth to perform local fine-tuning.

[0137] The aforementioned preset thresholds include one or more of the following: physical consistency deviation threshold, comprehensive confidence threshold, structural surface response consistency threshold, and output fluctuation threshold; the thresholds can be determined based on prior engineering experience, or based on statistical results of geological history samples, cross-validation results, or field test calibration results.

[0138] This invention also provides a system for realizing the method of collaborative penetration testing to perceive the material and structure of the ore and rock while drilling, the schematic diagram of which is shown below. Figure 3 As shown, it includes a data acquisition and preprocessing module, a model building module, a model training module, a model inference module, a consistency verification module, and a collaborative correction module;

[0139] The data acquisition and preprocessing module obtains the drilling time series parameters, performs data cleaning and multi-scale time series window feature extraction on the drilling time series parameters to obtain the input feature dataset, and uploads the data to the model building module;

[0140] The model building module constructs an initial model for inverting mineral and rock material parameters and identifying and locating structures based on a multi-task temporal convolutional network, and uploads the data to the model training module.

[0141] The model training module constructs a theoretical consistency loss term based on the self-derived theoretical model and adds the theoretical consistency loss term to the model's total loss function. The input feature dataset is used as the training dataset to train the initial mineral and rock material parameter inversion and structure identification and localization model, thereby obtaining the mineral and rock material parameter inversion and structure identification and localization model. The data is then uploaded to the model inference module.

[0142] The model inference module performs inference based on the mineral and rock material parameter inversion and structure identification and positioning model to obtain the initial prediction results of mineral and rock material parameters, the structure surface identification results and the initial positioning results of structure surface, and uploads the data to the consistency verification module.

[0143] The consistency verification module performs theoretical estimation consistency verification on the initial prediction results of the mineral and rock material parameters based on the self-derived theoretical model, and performs structural response consistency verification on the structural surface identification results and the initial structural surface positioning results based on the structural response characteristics; at the same time, it evaluates the confidence of the model inference results and uploads the data to the collaborative correction module.

[0144] The collaborative correction module performs collaborative correction on the initial prediction results of the mineral and rock material parameters, the structural surface identification results, and the initial location results of the structural surface based on the consistency verification results of theoretical estimation, the consistency verification results of structural response, and the confidence assessment results, so as to obtain the final mineral and rock material parameter results, structural surface identification results, and structural surface location results, thus completing the collaborative probing and sensing of mineral and rock material and structure.

[0145] The method of the present invention will be further described below with reference to an embodiment:

[0146] The drilling equipment is modularly modified as follows to adapt to the method of this invention, as shown in the schematic diagram. Figure 4 As shown:

[0147] High-frequency torque-speed integrated sensor: installed at the output end of the drill rig's power head, fixed to the drill rod connection with a flexible coupling, using a high-frequency sensor for sampling, torque measurement accuracy ±1% FS, speed measurement accuracy ±0.1 r / min, and data is transmitted in real time via bus;

[0148] Pressure and hydraulic stress sensors: installed in the rodless chamber of the propulsion cylinder and the main oil circuit of the hydraulic system respectively, with a measurement accuracy of ±0.5% FS, used to synchronously collect propulsion force and system working pressure;

[0149] Wire displacement sensor: Fixed to the end of the drill rig guide rail, the wire is rigidly connected to the power head, the measuring stroke is 0-10m, the resolution is 0.01mm, and it is used to obtain data on the propulsion speed and drilling depth.

[0150] Based on the estimated characteristics of the ore and rock in the construction area (ore and rock type, regional structure, etc.), the following three types of core parameters are configured:

[0151] Physical mechanism parameters: preset physical derivation formulas for mechanical parameters such as rock Poisson's ratio and drill bit-rock contact friction coefficient, and uniformly configure common drilling parameters such as drill bit tooth count and cutting tooth invasion angle;

[0152] Model and task parameters: The number of filters in the TCN network is set to 64-128, the convolution kernel is a one-dimensional convolution kernel with a size of 3, the dilation rate is configured, and the initial weights for multiple tasks are set.

[0153] Constraints and threshold parameters: Set reasonable ranges for mechanical parameters in engineering, thresholds for identifying abrupt changes in structural surfaces, and confidence thresholds.

[0154] The drilling monitoring system was statically calibrated using standard force sensors and a torque calibration bench to ensure that the measurement error of thrust and torque was ≤2%. Dynamic calibration was performed through indoor simulated drilling tests, and the sensor installation position and data transmission delay compensation coefficient were adjusted to ensure that the synchronization error of timing data was ≤10ms.

[0155] The staged drilling process includes:

[0156] During the hole-opening stage (drilling depth 0-2m): a low-speed, low-pressure drilling mode is adopted, and the stability of torque and propulsion force is monitored in real time. The impact interference data during the hole-opening stage is filtered through the data preprocessing module to ensure the validity of the model initialization input. If the interface energy jump intensity, load jump intensity ≥20%, and the width of the normalized transition zone ≥1mm occurs, it is determined to be a surface weathering fracture. The depth coordinates are recorded and continuously tracked.

[0157] During normal drilling (drilling depth greater than 2m): Based on the real-time output of the UCS inversion results from the model, the drilling parameters are dynamically adjusted, and the "high rotation speed + medium thrust" mode is adopted to improve drilling efficiency; when 30MPa < UCS ≤ 80MPa (medium-hard rock), the rotation speed is appropriately reduced while the thrust is increased to balance drilling efficiency and equipment load; when UCS > 80MPa (hard rock), the rotation speed is further reduced and the thrust is increased, and the staged pressurization mode is activated to avoid drill pipe overload and breakage.

[0158] Drilling in the vicinity of the structural surface (when the model output confidence level is ≥0.7): Automatically switch to "fine drilling mode", reduce rotation speed and thrust; increase sensor sampling frequency, enhance the capture of structural surface parameter abrupt change features, and ensure that the positioning accuracy meets the requirement of error ≤0.5m. At the same time, record the location information of parameter abrupt change features, and then calculate the width and angle information of the structural surface.

[0159] Specifically, the diagram illustrating the generation of raw and inverted data is as follows: Figure 5 As shown. Figure 5 The final visualization results of the drilling-while-drilling ore and rock characteristic identification are displayed. As the drilling footage depth increases, multi-dimensional parameters such as torque, rotational speed, thrust, and propulsion speed are collected sequentially. Then, the ore and rock parameters are inverted in real time to output the curve of rock strength change with drilling footage depth. At the same time, the in-situ structural surface information of the hole location is output, including position, angle, and width.

[0160] Drilling data is transmitted to edge computing nodes via 5G industrial modules for data cleaning and preprocessing. A 12-dimensional core feature is extracted using a dual sliding window (5s short window + 10s long window) to generate a standard time-series matrix. For model output and command issuance, the edge computing nodes load a pre-trained TCN model, outputting multiple mechanical parameter values, structural surface identification results, and positioning coordinates in real time, which are transmitted to the drilling rig control system via an industrial bus. When the overall confidence level is <0.85, a parameter verification mechanism is automatically triggered, re-collecting data for that depth segment and performing a secondary inversion. Construction parameters are adaptively adjusted. The drilling rig control system dynamically optimizes drilling parameters based on the correlation of the mechanical parameters output by the model. If the inversion result shows that the equivalent elastic modulus is too small, indicating strong ore plasticity, the advance speed is appropriately reduced. If the internal friction angle is too small, the rotation speed is appropriately increased to improve crushing efficiency.

Claims

1. A method for coordinating drilling exploration and sensing of mineral and rock material and structure, characterized in that, Includes the following steps: S1. Obtain the drilling time series parameters, and perform data cleaning and multi-scale time series window feature extraction on the drilling time series parameters to obtain the input feature dataset; S2. An initial mineral and rock material parameter inversion and structure identification and localization model is constructed based on a multi-task temporal convolutional network. The initial mineral and rock material parameter inversion and structure identification and localization model includes a shared temporal feature extraction layer, a material parameter inversion branch, a structure surface identification branch, and a structure surface localization branch. The initial mineral and rock material parameter inversion and structure identification and localization model also includes an autonomously derived theoretical model. Among them, the constraints of the autonomously derived theoretical model are not directly used as independent neural network branches to participate in forward prediction, but are used to construct physical constraint loss in the training stage and for execution result verification and collaborative optimization in the inference stage. S3. Construct a theoretical consistency loss term based on the self-derived theoretical model, and add the theoretical consistency loss term to the total loss function of the model. Use the input feature dataset as the training dataset to train the initial ore and rock material parameter inversion and structure identification and localization model to obtain the ore and rock parameter inversion and structure identification model. S4. Based on the inversion of mineral and rock parameters and the structural identification model, reasoning is performed to obtain the initial prediction results of mineral and rock material parameters, the structural surface identification results, and the initial location results of structural surfaces; S5. Based on the self-derived theoretical model, the initial prediction results of the mineral and rock material parameters are checked for consistency in theoretical estimation, and based on the structural response characteristics, the structural surface identification results and the initial structural surface positioning results are checked for consistency in structural response; at the same time, the confidence of the model inference results is evaluated. S6. Based on the consistency verification results of theoretical estimation, the consistency verification results of structural response, and the confidence assessment results, the initial prediction results of the mineral and rock material parameters, the structural surface identification results, and the initial positioning results of the structural surface are collaboratively corrected to obtain the final mineral and rock material parameter results, structural surface identification results, and structural surface positioning results, thus completing the collaborative probing and sensing of mineral and rock material and structure. The mineral and rock material parameters include at least one or more of the following: rock strength, elastic modulus, cohesion, and internal friction angle.

2. The method for co-probing and sensing mineral and rock material and structure during drilling according to claim 1, characterized in that, In step S1, the data cleaning includes outlier filtering and missing value linear interpolation completion; the multi-scale time window feature extraction includes extracting 6-dimensional basic statistical features, 4-dimensional dynamic derived features, and 2-dimensional mutation features using a combination of short and long windows through a dual sliding window.

3. The method for co-probing and sensing mineral and rock material and structure during drilling according to claim 2, characterized in that, The dual sliding window in the short-window and long-window combination method is a combination of a 5-second window and a 10-second window; the six-dimensional basic statistical features include the in-window mean of torque. In-window variance of torque The mean value of propulsion within the window Variance within the window of propulsion The mean value of rotational speed within the window The mean of propulsion speed within the window ; The four-dimensional dynamic derived features include rock-breaking energy ratio. Normalized transition width Torque Thrust-rate-of-drilling gradient ;in, This represents the average thrust within the previous window. This represents the average propagation speed within the previous window; the normalized transition width is the depth feature covered by the interface response as it increases from 10% to 90%, used to characterize the geometric expansion range of the interface perturbation. A wider normalized transition width indicates a more fragmented structural surface, expressed by the following formula: ;in, d represents the geometric depth features covered by the interface response as it increases from 10% to 90%; d represents the geometric thickness of the cutting tooth. The two-dimensional abrupt change features include the interface energy jump intensity and the load jump intensity; wherein, the interface energy jump intensity is the increase in mechanical specific energy of the region at the 0.75 upper quantile within the interface window compared to the mechanical specific energy of the stable section of the upper rock. A large interface jump intensity indicates a complete structural surface, and is expressed by the following formula: ;in, The mechanical specific energy of the region at the upper quantile of 0.75 within the interface window; The mechanical specific energy of the stable section of the upper rock is given; the load abrupt change intensity is the increase in the average thrust within the interface window compared to the baseline thrust during the stable drilling stage of the upper rock. A small load amplification intensity indicates the presence of cementing material at the structural surface, expressed by the following formula: ;in, The average propulsion force within the interface window; Baseline thrust for the stable drilling stage of the upper rock; Among them, the normalized transition width is used to characterize the spatial influence range of structural interface disturbance, and the interface energy jump intensity and load jump intensity are used to characterize one or more of the following: structural interface response strength, interface integrity, fragmentation or cementation state.

4. The method for co-probing and sensing the mineral and rock material and structure during drilling according to claim 1, characterized in that, In step S2, the initial ore and rock material parameter inversion and structure identification and localization model includes a shared temporal feature extraction layer and a branch task layer; The shared temporal feature extraction layer is used to extract common temporal features of material and structure from the input feature data; The branch task layer includes a multi-parameter inversion branch, a structural surface identification branch, and a structural surface location branch. The parameter inversion branch is processed through several cascaded fully connected layers and then linearly activated, outputting multi-dimensional mechanical parameters including rock strength, elastic modulus, and cohesion-internal friction angle. The structural surface identification branch is processed through fully connected layers and then activated by the Sigmoid function, outputting a judgment result on whether the structural surface exists. The structural surface location branch is processed through fully connected layers and then linearly activated, outputting the depth, location coordinates, and structural surface feature parameters of the structural surface. The structural surface feature parameters include at least one or more of the following: structural surface width, structural surface dip angle, and structural surface type. The output features of the shared temporal feature extraction layer are simultaneously input into the three branches of the branch task layer, and the three branches of the branch task layer feed back the gradients to the feature extraction layer through the backpropagation process.

5. The method for co-probing and sensing the mineral and rock material and structure during drilling according to claim 4, characterized in that, The shared temporal feature extraction layer is composed of multiple temporal convolutional network modules stacked in series. Each temporal convolutional network module includes a causal convolutional unit, a batch normalization unit, a random deactivation regularization unit, a shallow connection unit, and a residual connection unit connected in series. Each task branch shares the output features of the shared temporal feature extraction layer and jointly constrains the shared feature learning process through their respective task losses to achieve collaborative optimization between mineral material perception and structural response perception.

6. The method for co-probing and sensing mineral and rock material and structure during drilling according to claim 1, characterized in that, In step S3, an autonomous derivation theoretical model is introduced, which is established based on the theoretical mapping relationship between drilling force relationship, cutting response relationship of cutting teeth and ore material parameters; The self-derived theoretical model is used to calculate theoretical estimates of ore and rock material parameters based on drilling parameters, and to construct a theoretical consistency loss term based on the theoretical estimates. The autonomously derived theoretical model does not participate directly in forward prediction as an independent neural network branch. Instead, it is used to construct a theoretical consistency loss term during the training phase and to perform theoretical estimation consistency verification and collaborative correction on the initial prediction results of mineral and rock material parameters during the inference phase. This improves the physical rationality and generalization ability of the model's prediction results, ensures that the inversion data conforms to the basic physical mechanism conditions, and avoids the black box problem of learning and training based on pure data.

7. The method for coordinating drilling exploration and sensing of ore and rock material and structure according to claim 6, characterized in that, The self-derived theoretical model includes theoretical estimation formulas for calculating elastic modulus, cohesion, internal friction angle, and uniaxial compressive strength of rock; The theoretical estimation formula is established based on drilling propulsion force, torque, rotational speed, drilling speed, and drill bit structural parameters; in: Theoretical formula for calculating elastic modulus: ; Theoretical formula for calculating cohesion-angle of internal friction: ； Theoretical formula for calculating the uniaxial compressive strength of rock: ;in, Where T is the drilling propulsion force, N is the drilling torque, V is the rotational speed, and E is the rock elastic modulus. For the Poisson's ratio of rocks, The radius of the drill bit's geometric rotation center; The internal friction angle between the rock compression and fracture zone and the intact rock. The internal friction angle of the rock. Let m be the coefficient of friction between the drill bit and the rock contact surface, and m be the number of drill bit teeth. S is the comprehensive characteristic coefficient of the cutting tooth, and S is the comprehensive cutting parameter; This refers to the length of a single tooth. Comprehensive characteristic coefficient of cutting teeth With the intrusion angle of the cutting teeth Cross-sectional shape factor and the coefficient of friction between the drill bit and the rock contact surface Related, specific calculation formula: ; Combined cutting parameters S and tooth entry angle Coefficient of friction of rock contact surface and the internal friction angle between the rock compression fracture zone and the intact rock. Regarding the specific calculation formula: .

8. The method for co-probing and sensing mineral and rock material and structure during drilling according to claim 6, characterized in that, In step S3, the total loss function of the model Express it using the following formula: ;in, Huber loss for the parameter inversion branch; The Huber loss weighting coefficient; Cross-entropy loss for identifying branches on structural surfaces; These are the weighting coefficients for the cross-entropy loss. Mean square error loss for locating branches of the structural surface; The mean squared error loss weighting coefficient; This is a physical constraint regularization term that includes the loss of consistency between parameter theories, the mechanical correlation deviation between parameters, and the correlation deviation between structure and parameters; The weight coefficient for the theoretical consistency regularization term.

9. A system for implementing the drilling-while-penetrating collaborative penetration testing method for ore and rock material and structure as described in any one of claims 1 to 8, characterized in that, It includes a data acquisition and preprocessing module, a model building module, a model training module, a model inference module, a consistency verification module, and a collaborative correction module; The data acquisition and preprocessing module obtains the drilling time series parameters, performs data cleaning and multi-scale time series window feature extraction on the drilling time series parameters to obtain the input feature dataset, and uploads the data to the model building module; The model building module constructs an initial model for inverting mineral and rock material parameters and identifying and locating structures based on a multi-task temporal convolutional network, and uploads the data to the model training module. The model training module constructs a theoretical consistency loss term based on the self-derived theoretical model and adds the theoretical consistency loss term to the model's total loss function. The input feature dataset is used as the training dataset to train the initial ore and rock material parameter inversion and structure identification and localization model, thereby obtaining the ore and rock parameter inversion and structure identification model. The data is then uploaded to the model inference module. The model inference module performs inference based on the ore and rock parameter inversion and structure recognition model to obtain the initial prediction results of ore and rock material parameters, structure surface recognition results and initial location results of structure surfaces, and uploads the data to the consistency verification module. The consistency verification module performs theoretical estimation consistency verification on the initial prediction results of the mineral and rock material parameters based on the self-derived theoretical model, and performs structural response consistency verification on the structural surface identification results and the initial structural surface positioning results based on the structural response characteristics; at the same time, it evaluates the confidence of the model inference results and uploads the data to the collaborative correction module. The collaborative correction module performs collaborative correction on the initial prediction results of the mineral and rock material parameters, the structural surface identification results, and the initial location results of the structural surface based on the consistency verification results of theoretical estimation, the consistency verification results of structural response, and the confidence assessment results, so as to obtain the final mineral and rock material parameter results, structural surface identification results, and structural surface location results, thus completing the collaborative probing and sensing of mineral and rock material and structure.

Images