Rock mass dynamic mechanical parameter sensing method and system based on drilling parameters

Abstract

The present application relates to the field of geological exploration and geotechnical engineering, in order to obtain the dynamic mechanical parameter characteristics of rock mass in situ, real-time and dynamically in the drilling process, a rock mass dynamic mechanical parameter sensing method and system based on drilling parameters are provided, using vibration, acoustic emission, torque, drilling pressure and other multi-modal signals, through unified feature engineering and multi-task learning, the instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion and internal friction angle can be dynamically predicted simultaneously in a single drilling, breaking through the limitation of traditional method which only estimates a single static parameter; the rock mass dynamic mechanical parameter acquisition model based on multi-modal adaptive graph attention timing network can effectively couple the sensor physical proximity and statistical correlation, and fuse the inflation timing convolution and time perception Transformer module, under the premise of low time delay, the representation ability of sudden events (such as sticking / drilling resistance) and long-term trend is considered, thereby improving the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geological exploration and geotechnical engineering, and particularly relates to a rock mass dynamic mechanical parameter sensing method and system based on drilling parameters. BACKGROUND

[0002] In deep resource mining, tunneling, geothermal development, and oil and gas drilling, the in-situ rock mass mechanical parameters are mainly obtained through laboratory core testing, geophysical logging and exploration, simplified drilling speed model strength estimation, and drilling formation evaluation. Laboratory core testing requires stopping drilling to take cores, which is time-consuming and costly, and destroys the in-situ state, and cannot reflect the dynamic response under drilling disturbance; geophysical logging and exploration have limited resolution and insufficient real-time performance, and the inversion results are macroscopic average characteristics, which are difficult to capture local stress changes and dynamic responses; simplified drilling speed model strength estimation can only estimate a single static parameter (such as UCS), and the precision is limited, and multiple parameter dynamic mechanical characteristics cannot be obtained; part of the drilling formation evaluation mainly focuses on lithology identification, porosity and pore pressure, and the real-time, multi-parameter, in-situ sensing ability of dynamic mechanical parameters is insufficient. At present, there is no method to obtain the dynamic mechanical parameter characteristics of rock mass in-situ, in real time, and dynamically during drilling, which restricts the safe and efficient construction of deep engineering. SUMMARY

[0003] In order to obtain the dynamic mechanical parameter characteristics of rock mass in-situ, in real time, and dynamically during drilling, a rock mass dynamic mechanical parameter sensing method and system based on drilling parameters are provided.

[0004] The technical scheme adopted by the present application to solve the above problems is:

[0005] The rock mass dynamic mechanical parameter sensing system based on drilling parameters comprises:

[0006] A drill bit integrated sensor module, including a probe body, a drill bit measurement unit for obtaining drill rock vibration frequency spectrum, a drill string monitoring unit for monitoring mechanical load, and an acoustic wave acquisition unit for acquiring acoustic waves, the drill bit measurement unit, the drill string monitoring unit, and the acoustic wave acquisition unit are all arranged on the probe body;

[0007] A data preprocessing module for preprocessing the original data collected by the drill bit integrated sensor module, including synchronization alignment, hybrid noise reduction, and feature extraction, and standardizing and feature dimensionality reduction of the extracted features;

[0008] A rock mass dynamic mechanical parameter acquisition model module for constructing a rock mass dynamic mechanical parameter acquisition model based on a multi-modal adaptive graph attention time series network and training;

[0009] A dynamic mechanical parameter acquisition module for acquiring mechanical parameters based on the preprocessed data and the trained rock mass dynamic mechanical parameter acquisition model.

[0010] Further, the drill bit measurement unit comprises a three-axis MEMS acceleration sensor and a piezoelectric impact stress wave sensor; the drill string monitoring unit comprises a torque sensor and an axial drilling pressure sensor; the acoustic wave acquisition unit comprises a piezoelectric acoustic wave transducer, a differential amplifier and a programmable band-pass filter; the drill bit measurement unit and the acoustic wave acquisition unit are arranged behind the drill bit, and the drill string monitoring unit is arranged at both ends of the probe rod.

[0011] Further, it also comprises a pre-warning module which gives an alarm when the dynamic mechanical parameters obtained by the dynamic mechanical parameter acquisition module exceed the corresponding pre-warning threshold, and the pre-warning threshold is determined based on historical statistics and multi-task weight.

[0012] The rock mass dynamic mechanical parameter sensing method based on drilling parameters comprises:

[0013] Step 1, constructing training sample data: using the drill bit integrated sensor module to collect drilling parameters and pre-processing them, including synchronous alignment, hybrid noise reduction and feature extraction, and standardizing and feature dimensionality reduction of the extracted features; obtaining the mechanical parameters corresponding to the drilling parameters through laboratory core true values and finite element simulation, and constructing training sample data based on the pre-processed data and the corresponding mechanical parameters;

[0014] Step 2, constructing a rock mass dynamic mechanical parameter acquisition model based on a multi-modal adaptive graph attention time sequence network and training it based on the training sample data;

[0015] Step 3, using the drill bit integrated sensor module to collect drilling parameters in real time, and using the same pre-processing method as the training sample to pre-process them, and obtaining the mechanical parameters based on the pre-processed data and the trained rock mass dynamic mechanical parameter acquisition model.

[0016] Further, the mechanical parameters are: instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion and internal friction angle.

[0017] Further, the pre-processing step is specifically:

[0018] Aligning the time scales of all sensor channels, real-time dividing the signal stream with M window length and S step, and processing each window independently;

[0019] Performing band-pass filtering of different frequencies according to the data characteristics;

[0020] Performing ensemble empirical mode decomposition on the filtered signal, selecting the first N IMF components for reconstruction; performing singular value decomposition on the reconstructed signal, and retaining the principal components with singular value energy ratio of more than Y%;

[0021] Dynamically estimating the noise spectrum of the acoustic wave signal, adaptively reducing the noise energy according to the frequency band, and retaining the rock mass fracture characteristic spectrum line;

[0022] Parallel extraction of multiple features of the windowed signal, including:

[0023] Time-domain features: calculate mean, variance, peak, RMS energy, kurtosis, skewness, instantaneous mechanical power and drilling power within the window;

[0024] Frequency-domain features: use fast Fourier transform to extract power spectral density, calculate dominant frequency, 1-5 kHz band energy ratio, spectral kurtosis; identify the resonant peak frequency through cepstrum analysis;

[0025] Time-frequency domain features: perform short-time Fourier transform on the band-passed signal to generate time-frequency spectrum, extract energy entropy and spectral centroid trajectory; apply wavelet packet decomposition to obtain multi-scale sub-band coefficients, calculate the energy, information entropy and frequency band variability of each sub-band;

[0026] Dynamic event labeling: use the abrupt point detection algorithm to detect torque abruptness, label the stuck / friction event time, and embed the [0,1] event flag in the feature vector;

[0027] Standardize the extracted features: Z-score normalization for time-domain features, and Min-Max scaling to [0,1] for frequency-domain features;

[0028] PCA dimensionality reduction is performed on time-domain, frequency-domain and time-frequency domain features respectively.

[0029] Further, the rock mass dynamic mechanical parameter acquisition model comprises:

[0030] A sensor topology module for constructing a spatial correlation graph structure of sensor data;

[0031] A graph attention layer module for calculating attention coefficients based on the spatial correlation graph structure and aggregating features based on the attention coefficients;

[0032] A time series convolution graph module for multi-scale extraction of time series dimensions based on the sequence features output by the graph attention layer module, and for optimizing the training process based on local supervision, enhancing feature nonlinear expression and local feature learning accuracy through causal dilated convolution groups and lightweight regression heads;

[0033] A time-aware Transformer enhancement module: based on the long-term and short-term dependence modeling of time series features, relative position encoding is introduced through multi-head self-attention mechanism, and attention weights are adjusted based on exponential decay factor to enhance the simultaneous modeling ability of sudden events and long-term trends;

[0034] A multi-task regression head module: based on the final feature sequence, multiple target parameters are predicted in parallel.

[0035] Furthermore, the rock mass dynamic mechanical parameter acquisition model adopts a layered deployment, with the sensor topology graph module, graph attention layer module, and time-series convolution graph module deployed on the downhole edge computing node, and the time-aware Transformer enhancement module and multi-task regression head module deployed on the ground server.

[0036] Furthermore, a composite loss function L is used during model training: , where y is the truth value; The predicted value output by the model; Weights; symbol It is an L1 norm. ,symbol It is the L2 norm. , This refers to the predicted value of the i-th physical quantity in the final output vector for the current sample. The true value of the i-th physical quantity in the final output vector for the current sample is given. When i=1,…,5, it corresponds to the instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion, and internal friction angle, respectively. ,in Indicates the first The predicted value of the regression head for the s-th sample, For the s-th sample in the th case The true values ​​on the regression heads, where n is the number of samples. , This represents the number of regression heads; These are the model parameters.

[0037] Furthermore, it also includes step 4, which is to issue an alarm when the mechanical parameters acquired by the dynamic mechanical parameter acquisition module exceed the corresponding warning threshold. The warning threshold is determined based on historical statistics and multi-task weights.

[0038] The advantages of this invention compared to the prior art are:

[0039] (i) Joint prediction of multi-source drilling parameters to multi-objective mechanical parameters: By utilizing multi-modal signals such as vibration, acoustic emission, torque, and drilling pressure, and through unified feature engineering and multi-task learning, it is possible to synchronously and dynamically predict instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion, and internal friction angle during a single drilling operation, breaking through the limitation of traditional methods that only estimate a single static parameter.

[0040] (ii) Adaptive Graph Attention Temporal Network: A network structure of "trainable edge weights + GATv2" is proposed to effectively couple the physical proximity and statistical correlation of sensors, and integrate dilated temporal convolution and time-aware Transformer module. Under the premise of low latency, it can take into account the representation ability of sudden events (such as stuck drills / friction) and long-term trends, thereby improving the prediction accuracy of the model.

[0041] (Three) Engineering availability: hierarchical deployment (downhole edge + ground server), online adaptive fine-tuning, dynamic threshold early warning and explainable analysis, significantly improving real-time, stability and availability.

[0042] Therefore, the present application can realize in-situ, real-time, dynamic and multi-parameter joint sensing under the condition of no drilling, and is suitable for safe and efficient construction and risk warning of deep complex strata. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 It is a structural schematic diagram of a rock mass dynamic mechanical parameter sensing system based on drilling parameters.

[0044] Figure 2 It is a structural schematic diagram of a probe rod.

[0045] Figure 3 It is a sectional view of the probe rod.

[0046] Figure 4 It is a flow chart of a rock mass dynamic mechanical parameter sensing method based on drilling parameters.

[0047] Reference signs: 1 is a probe rod body, 2 is a sealing cover plate, 3 is a triaxial MEMS acceleration sensor, 4 is a piezoelectric impact stress wave sensor, 5 is a ring strain type torque sensor, 6 is a piezoelectric ceramic type axial WOB sensor, 7 is a piezoelectric acoustic transducer, 8 is a temperature sensor, 9 is a piezoresistive pressure sensor, and 10 is a turbine flowmeter. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below with examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0049] The present method realizes real-time, dynamic and joint inversion of multiple mechanical parameters during drilling through multi-modal signal fusion and adaptive time series graph neural network architecture, significantly improving the intelligent level of geological parameter acquisition during drilling.

[0050] As shown in Figure 1 The rock mass dynamic mechanical parameter sensing system based on drilling parameters comprises:

[0051] The drill bit integrated sensor module comprises a probe rod body, a drill bit measurement unit for acquiring drill rock vibration spectrum, a drill string monitoring unit for monitoring mechanical load, and an acoustic wave acquisition unit for acquiring acoustic waves, and the drill bit measurement unit, the drill string monitoring unit, and the acoustic wave acquisition unit are arranged on the probe rod body.

[0052] Specifically, asFigure 2 , 3 As shown in FIGS. 1 and 2, the probe body 1 is a cylindrical rod made of high-strength alloy steel, with a sensor mounting hole and a wiring channel inside, and a flow hole in the annular space. The upper and lower ends are provided with threads for connecting with the drill bit and drill rod, and the side wall of the rod body is provided with an axial slot, and the sealing cover plate 2 is connected by bolts to realize modular installation and waterproof of the sensor. The drill bit measurement unit, drill string monitoring unit and acoustic wave acquisition unit are installed in the probe body and fixed in the corresponding holes and slots by bolts or glue. Among them, the drill bit measurement unit: a three-axis MEMS acceleration sensor 3 and a piezoelectric impact stress wave sensor 4 are embedded behind the drill bit to directly capture the drill rock vibration spectrum. The drill string monitoring unit: one set of ring-shaped strain torque sensor 5 and piezoelectric ceramic axial drilling pressure sensor 6 are installed at each end of the rod body to monitor the mechanical load in real time. The acoustic wave acquisition unit: four piezoelectric acoustic transducers 7 are arranged near the drill bit, connected with a differential amplifier and a programmable band-pass filter.

[0053] In addition, the probe body is also provided with an environmental monitoring unit: integrated PT100 temperature sensor 8, piezoresistive pressure sensor 9 and turbine flowmeter 10, which is used to obtain temperature T, pressure p and flow Q in real time, and is used as an online correction quantity of each measurement channel to suppress the influence of temperature drift and drilling fluid on the measurement results. The specific correction method is as follows:

[0054] (1) Temperature drift compensation: a cubic polynomial compensation model with temperature as the independent variable is established for the acceleration, ring-shaped strain torque and piezoelectric drilling pressure channels: , wherein, : the measurement value after temperature drift compensation (the unit is consistent with the channel, such as g, N·m, kN, Pa); : the original output value (the unit is the same as above); coefficient (k=0, 1, 2, 3): temperature compensation polynomial coefficient (matched with the dimension of the channel; obtained by segment calibration in a thermostat and written into the sensor EEPROM (Electrically Erasable Programmable Read-Only Memory)).

[0055] (2) Pressure / flow calibration and temperature-viscosity correction: the pressure channel implements "two-point zero / full scale + multi-point linear fitting" compound calibration; the turbine flowmeter adopts volume metering method to obtain a nonlinear compensation table through 6-point calibration, and combines temperature-viscosity / density correction: , : compensated volume flow ( or L / min); : turbine flowmeter original reading (the unit is the same as above); , : Temperature-dependent / density correction factor (regressed from experimental data, dimensionless); : Dynamic viscosity of drilling fluid at temperature T (Pa·s); : Experimentally measured or model-estimated at temperature T; : Reference temperature : Viscosity at reference temperature : Viscosity at reference temperature : Drilling fluid density at temperature T, pressure p (kg / m3); : Drilling fluid density at reference state : Drilling fluid density at reference state : Drilling fluid density at reference state : Annulus / channel pressure (Pa or MPa).

[0056] (3) Drilling fluid interference (liquid damping / absorption) correction: Estimate drilling fluid viscosity and density at T, p, and calculate band attenuation using empirical model. Apply equivalent inverse filtering to shock / acoustic channel in frequency domain , and make viscosity-normalized correction to MSE, power, etc. MSE (mechanical specific energy): MSE = (WOB x RPM) / ROP; where, : Equivalent inverse filter transfer function (dimensionless, used to compensate for medium attenuation); f: frequency (Hz, from frequency variable after FFT on signal); : Band-dependent amplitude attenuation factor (Np / m; calculated from empirical / semi-empirical model, varies with viscosity μ, density ρ); L: equivalent propagation path length of acoustic / shock signal in medium (m); WOB: weight on bit (kN), from axial WOB sensor; RPM: rotational speed (r / min or r / s); ROP: rate of penetration (m / s or m / h).

[0057] Power: ; : Torque (N·m, from ring-strain torque sensor); : Angular velocity (rad / s, converted from rotational speed).

[0058] (4) Channel cross-sensitivity elimination: Construct cross-sensitivity matrix , and solve by online least squares: , : Real channel vector after decoupling (each component has unit according to channel); : Measured multi-channel vector (original / pre-processed data). I: identity matrix, dimension N x N (N is number of channels). K: cross-sensitivity matrix (dimensionless); element k_ij represents cross-talk weight of channel j to channel i (usually k_ii = 0).

[0059] (5) Zero point / drift suppression in situ: automatic reset of zero offset and update of compensation coefficients of the current temperature section in the drilling static window.

[0060] After the above correction, the temperature drift suppression is better than 0.5% FS / 10°C, the amplitude error caused by drilling fluid is less than 2%, and the long-term zero point drift is less than 0.2% FS.

[0061] Synchronization and calibration: dual-redundant clock sources (downhole OCXO oven-controlled crystal oscillator + GPS disciplined clock), time synchronization error < 1 μs; before drilling, automatically calibrate the frequency response characteristics of each sensor through a standard exciter. Centralized device: built-in lithium sub-battery (endurance 72 h) and RS-485 / CAN bus interface, transmit the matched impedance of multiple analog signals to the ground system.

[0062] Vibration, acoustic emission, torque, and drilling pressure provide complementary observation perspectives. By cross-verifying and fusing the information captured at different physical levels (mechanical, acoustic, and hydraulic), the prediction accuracy, reliability, and stability are improved.

[0063] Data preprocessing module: used for preprocessing the raw data collected by the drill bit integrated sensor module, including synchronization alignment, hybrid noise reduction, and feature extraction, and standardizing and feature dimensionality reduction of the extracted features.

[0064] Rock mass dynamic mechanical parameter acquisition model module: based on a multi-modal adaptive graph attention time series network, a rock mass dynamic mechanical parameter acquisition model is constructed and trained.

[0065] Dynamic mechanical parameter acquisition module: based on the preprocessed data and the trained rock mass dynamic mechanical parameter acquisition model, the mechanical parameters are acquired.

[0066] Correspondingly, as shown in Figure 4 , a rock mass dynamic mechanical parameter sensing method based on drilling parameters is also provided, comprising:

[0067] Step 1, construct training sample data: use the drill bit integrated sensor module to collect drilling parameters and preprocess them, including synchronization alignment, hybrid noise reduction, and feature extraction, and standardize and feature dimensionality reduction of the extracted features; through laboratory core true value and finite element simulation, the mechanical parameters corresponding to the drilling parameters are acquired, and based on the preprocessed data and the corresponding mechanical parameters, the training sample data is constructed.

[0068] The preprocessing process specifically includes:

[0069] 1. Data synchronization alignment

[0070] All sensor channel time stamps are aligned based on GPS / PTP protocol to ensure time synchronization accuracy. The signal stream is split into windows of length M and step S, and each window is processed independently. The choice of window length M and step S is based on the frequency characteristics and sampling rate of the signal, such as 1.0s and 0.1s, respectively.

[0071] 2. Mixed noise reduction

[0072] Band-pass filtering is performed on different frequencies according to the characteristics of the data. For example, 0-1kHz band-pass filtering is applied to mechanical parameters such as weight on bit (WOB) and rotational speed (RPM); 0.5-20kHz band-pass filtering is applied to vibration and acoustic wave data; and low-frequency drift and high-frequency interference filtering is applied.

[0073] Empirical mode decomposition (EEMD)-singular value decomposition (SVD) reconstruction: perform ensemble empirical mode decomposition on the filtered signal, select the first N (set to 5 in this embodiment) IMF components for reconstruction; perform singular value decomposition (SVD) on the reconstructed signal, and retain the principal components with a singular value energy ratio of more than 95%.

[0074] Adaptive spectral subtraction: dynamically estimate the noise spectrum of the acoustic signal, adaptively reduce the noise energy according to the frequency band, and retain the rock mass fracture characteristic spectrum.

[0075] 3. Feature extraction

[0076] Time domain features: calculate the mean, variance, peak value, RMS energy, kurtosis, and skewness in the window; instantaneous mechanical specific energy MSE=(WOB×RPM) / ROP; drilling power .

[0077] Frequency domain features: use fast Fourier transform (FFT) to extract power spectral density (PSD), calculate the dominant frequency, 1-5kHz band energy ratio, and spectral kurtosis; identify the resonance peak frequency through cepstrum analysis.

[0078] Time-frequency domain features: perform short-time Fourier transform (STFT) on the band-passed signal to generate a time-frequency spectrum, extract energy entropy and spectral centroid trajectory; apply wavelet packet decomposition (WPD) to obtain multi-scale sub-band coefficients, and calculate the energy, information entropy, and frequency band variability of each sub-band.

[0079] Dynamic event labeling: use a mutation point detection algorithm such as the CUSUM algorithm to detect torque mutations, label the stuck / abrasion event time, and embed a [0,1] event flag in the feature vector.

[0080] 4. Feature standardization and dimensionality reduction

[0081] The extracted multi-modal features are normalized in the time domain using Z-score (mean=0, variance=1), and the frequency domain features are scaled to [0,1] using Min-Max scaling.

[0082] Feature dimension reduction: PCA dimension reduction (cumulative variance > 95%) is performed on the time domain, frequency domain and time-frequency domain features respectively, and in this embodiment, it is spliced into a 120-dimensional feature vector.

[0083] In addition, in order to increase the amount of data, sample enhancement is also carried out in this embodiment, which expands the data set to 3 times the original by adding Gaussian noise, random scaling, time shift, etc. The expanded data set is divided into 70% training set, 15% validation set and 15% test set.

[0084] Step 2, construct a rock mass dynamic mechanical parameter acquisition model based on a multi-modal adaptive graph attention time sequence network and train based on training sample data.

[0085] The rock mass dynamic mechanical parameter acquisition model is composed of five core functional modules connected in turn, and each module cooperates to complete the whole process from the input of the pretreated sensor signal to the output of the target parameter, which specifically includes:

[0086] Sensor topology graph module: This module is used to construct the spatial correlation graph structure of sensor data, providing the basis topology for subsequent graph attention calculation. The preprocessed signals collected by each sensor are used as the nodes of the graph; the initial value of the edge weight between the nodes is generated by weighted calculation, and the edge weight is set as a trainable parameter, which is dynamically optimized in the network back propagation process to accurately capture the true correlation strength between sensors. The edge weight calculation formula is as follows:

[0087] , wherein is the Euclidean distance between sensors, is mutual information; is the distance scale; is a numerical stability constant; , are the feature vectors of the , the sensor in the current time window after preprocessing; , is the feature vector of any two sensors in the same time window, which is used to calculate the maximum value of the mutual information of all channel pairs to realize normalization.

[0088] Graph attention layer module: Based on the topology graph structure, the spatial correlation features of the nodes are extracted, and the influence of the key nodes is highlighted through the graph attention layer module. The graph attention layer module adopts 2-layer GATv2, and 4 attention heads are set in each layer to realize multi-scale feature aggregation; node feature aggregation is completed through attention coefficient weighting, wherein the attention coefficient is combined with the trainable edge weight calculation, so that the network can adaptively focus on the sensor nodes that are more important for target prediction. The node feature aggregation formula is: , wherein the attention coefficient where: a is the attention weight vector (learnable parameter); represents the node the feature vector before entering this layer of GAT; : the updated node feature (output vector); : a nonlinear activation function (e.g. ReLU / ELU / LeakyReLU) to increase expressiveness and stability; : the set of neighbors of the node . ∈ [0, 1]: the attention weight from neighbor to . For fixed , these weights are softmax-normalized over neighbors, satisfying , learned by the scoring network, reflecting the "importance of neighbor to "; : linear mapping matrix (learnable) to project neighbor features from input dimension to output dimension ; : input feature of neighbor node . A "node" is the feature vector (e.g. 120-dimensional after preprocessing) of a sensor / channel at the current time window; is the trainable edge weight.

[0089] Temporal Convolutional Graph Module (TCGM): Multi-scale temporal dimension extraction is performed on the sequence features output by the graph attention, and local supervision optimization training is introduced. The core is the causal dilated convolution group: the graph attention output sequence of each node is input into 4 parallel convolution channels, each channel has a dilation rate of 1, 2, 4, and 8, and the temporal receptive field is expanded through dilated convolution; after the parallel channel outputs are spliced, they are processed by LayerNorm standardization and ReLU activation function to enhance the nonlinear expression of features; a lightweight regression head (2-layer fully connected network) is connected after each TCGM unit, and the local MAE loss is calculated through deep supervision to assist the early convergence of the network and improve the accuracy of local feature learning.

[0090] The Time-Aware Transformer Enhancement Module further models the short- and long-term dependencies of temporal features, enhancing the ability to simultaneously capture sudden events and long-term trends. It receives the feature sequence output from the TCGM and inputs it into a lightweight TransformerEncoder; it introduces relative position encoding into the multi-head self-attention process to clarify temporal positional relationships; and it multiplies the attention weights by an exponential decay factor. Enhance the ability to simultaneously model emergencies and long-term trends, including and They represent the first and second parts of the feature sequence, respectively. The time step and the first Each time step corresponds to a timestamp (or sampling moment), in seconds (s) or milliseconds (ms), determined by the sampling period. Accumulated.

[0091] Multi-task regression head module: This module performs parallel prediction of multiple target parameters based on the final feature sequence and improves overall prediction accuracy through joint loss optimization. It includes five independent fully connected regression branches, corresponding to the instantaneous elastic modulus E, dynamic Poisson's ratio, etc. Transient uniaxial compressive strength Cohesion C and internal friction angle .

[0092] An adaptive graph attention temporal network is constructed, which utilizes trainable edge weights and GATv2 to dynamically capture the physical associations and statistical dependencies between sensor nodes. It combines the local feature extraction capability of dilated convolution with the long-term context modeling of time-aware Transformer to achieve multi-scale feature representation of drilling temporal signals, thereby improving the model's prediction accuracy.

[0093] When training the model, the loss function used is the composite loss function L. ,in: The true value is given by core experiment / simulation. The output is the mechanical parameter result. , , , , These are the predicted values ​​for the instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion, and internal friction angle of the rock mass, respectively. As weight; These are model parameters; For the first The mean absolute error of the intermediate regression heads (or local outputs) is defined as: ,in Indicates the first The predicted value of each regression head for the s-th sample; n is the number of samples in the batch; For the s-th sample in the th case The true value at the top of the regression line, , The number of intermediate supervision / regression heads set (e.g., the expansion branches or local outputs of TCGM). Use... Monte Carlo dropout (p=0.2) estimates prediction uncertainty and has a regularization effect; the optimization strategy employs the AdamW optimizer, combined with linear warm-up and cosine annealing learning rate scheduling; Label Smoothing, Stochastic Depth, and Spectral Jitter techniques are applied at the network layers. The AdamW optimizer improves the model's generalization ability by handling weight decay more reasonably; the linear warm-up phase starts with a small learning rate and gradually increases it to a preset maximum value to help the model converge stably. Subsequently, the cosine annealing strategy gradually decreases the learning rate in the form of a cosine function, which helps the model escape local optima and explore a wider parameter space; Label Smoothing: improves generalization ability by smoothing the label distribution and reducing the model's overconfidence in the training data; Stochastic Depth: increases the model's diversity and prevents overfitting by randomly dropping some layers in the network during training; Spectral Jitter: enhances the data by adding random noise in the frequency domain, improving the model's robustness to different frequency variations.

[0094] Step 3: The drill bit integrated sensor module is used to collect drilling parameters in real time, and the same preprocessing method as the training samples is used for preprocessing. Based on the preprocessed data and the trained rock mass dynamic mechanical parameters, the mechanical parameters are obtained from the model.

[0095] The preprocessed data is input sequentially into: sensor topology graph module → graph attention layer module (GATv2) → temporal convolutional graph module (TCGM) → time-aware Transformer enhancement module → multi-task regression head. Output parameters: instantaneous elastic modulus E, dynamic Poisson's ratio. Transient uniaxial compressive strength Cohesion C and internal friction angle In this embodiment, the input data sequence is scrolled using a 1-second sliding window and a 0.1-second step size to update the parameter prediction results in real time.

[0096] To improve the processing efficiency, the embodiment adopts a hierarchical deployment of downhole edge + ground server: the sensor topology graph module, the graph attention layer module (GATv2) and the TCGM are deployed in the downhole edge computing node (GPU / FPGA), ensuring that the single-step processing delay is less than or equal to 20 ms; the time-aware Transformer enhancement module and the multi-task regression head are deployed in the ground high-performance server, which undertakes the tasks of reading buffer, centralized inference and visualization.

[0097] To improve the prediction accuracy, online adaptive update is provided: the prediction residual is continuously monitored, and the error trend is tracked through exponential weighted moving average (EWMA); when the residual exceeds the preset threshold, online fine-tuning is triggered only for the Transformer enhancement layer or the regression head; the on-site point measurement verification data is regularly integrated to perform incremental retraining, so as to improve the adaptability and stability of the model to new strata.

[0098] In addition, based on the model prediction results, the embodiment also provides a visualization and early warning module: the ground platform simultaneously supports two-dimensional curve graphs, heat maps and three-dimensional geological models, which can be interactively scaled and time-depth scrolled; dynamic threshold early warning: based on historical statistics and adaptive update of multi-task weight early warning threshold, when the parameter is out of limit, an audible and visual alarm is immediately sounded; explainable analysis: integrating Attention Rollout, Grad-CAM and GAT edge weight visualization, the input features and model decisions are provided to engineers for visual explanation.

[0099] The specific way of adaptive update of early warning threshold based on historical statistics and multi-task weight is: using sliding window robust statistics to give the "center + scale" of each parameter, then using "priority ÷ uncertainty" to form task weight, and adaptively scaling the threshold bandwidth according to the weight and updating it over time.

[0100] The "center" represents the typical level of the parameter under the current working condition, and the exponential weighted moving average (EWMA) is adopted: , wherein is the predicted value at the current time, is the center value at the current time, is the center value calculated at the last time, is the smoothing coefficient.

[0101] The "scale" represents the fluctuation amplitude, and the EWMA absolute deviation is adopted: , is the fluctuation scale at the current time, is the fluctuation scale calculated at the last time, is the smoothing coefficient.

[0102] “Priority” is calculated by “deviation x duration x importance” to get score S, and divided into three levels: normalized deviation is calculated by: ; uncertainty reduction is introduced by: , and continuous threshold-crossing duration τ (s) is counted, w_p is the business weight set for different monitoring parameters / different working conditions (different parameters can be different, default 1), and the score is: . Grading rules (can be calibrated on site): level one (P1, high-level alarm): event score S≥12 or continuous threshold-crossing time τ≥5 s; level two (P2, medium-level alarm): event score ; level three (P3, low-level alarm): event score . To prevent jitter, a 10% hysteresis and a minimum duration of 1 s strategy is adopted.

[0103] “Uncertainty” comes from the prediction standard deviation of MC-Dropout (or the square root of variance), and the normalized and truncated uncertainty coefficient is . Where u_ref is the empirical reference uncertainty (such as the average value of the training set or historical normal working conditions), and u_max is the upper limit of uncertainty truncation (default 2.0), which is used to avoid extreme uncertainty from making the threshold bandwidth too large.

[0104] The upper and lower thresholds of the adaptive threshold are calculated as follows, and are updated rolling with time: 、 , where k is the alarm sensitivity (k=2 for early warning and k=3 for threshold crossing), and κ is the uncertainty amplification coefficient (recommended 0.5).

Claims

1. A rock mass dynamic mechanical parameter perception system based on drilling parameters, characterized in that, Comprise: Drill bit integrated sensor module: including probe rod body, drill bit measurement unit for acquiring rock drilling vibration frequency spectrum, drill string monitoring unit for monitoring mechanical load and acoustic wave acquisition unit for collecting acoustic waves, drill bit measurement unit, drill string monitoring unit and acoustic wave acquisition unit are arranged on the probe rod body; Data preprocessing module: for preprocessing the original data collected by the drill bit integrated sensor module, including synchronization alignment, hybrid noise reduction and feature extraction, and standardizing and feature dimensionality reduction of the extracted features; Rock mass dynamic mechanical parameter acquisition model module: based on a multi-modal adaptive graph attention time series network, a rock mass dynamic mechanical parameter acquisition model is constructed and trained; Dynamic mechanical parameter acquisition module: based on the preprocessed data and the trained rock mass dynamic mechanical parameter acquisition model, the mechanical parameters are acquired; Wherein, the mechanical parameters are: instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion and internal friction angle; The rock mass dynamic mechanical parameter acquisition model comprises: Sensor topology graph module, for constructing a spatial correlation graph structure of sensor data; Graph attention layer module, based on the spatial correlation graph structure, the attention coefficients are calculated, and the features are aggregated based on the attention coefficients; Time series convolution graph module: based on the sequence features output by the graph attention layer module, multi-scale extraction of time series dimension is performed, and the training process is optimized based on local supervision, the feature nonlinear expression and local feature learning accuracy are enhanced through the causal inflation convolution group and the lightweight regression head; Time-aware Transformer enhancement module: based on the long-short term dependence relationship modeling of time series features, the relative position encoding is introduced through the multi-head self-attention mechanism, and the attention weight is adjusted based on the exponential decay factor, to enhance the simultaneous modeling ability of sudden events and long-term trends; Multi-task regression head module: based on the final feature sequence, the multi-target parameters are predicted in parallel.

2. The system for sensing the dynamic mechanical parameters of rock mass based on the while-drilling parameters according to claim 1, characterized in that, The drill bit measurement unit comprises a three-axis MEMS acceleration sensor and a piezoelectric impact stress wave sensor; the drill string monitoring unit comprises a torque sensor and an axial drilling pressure sensor; the acoustic wave acquisition unit comprises a piezoelectric acoustic wave transducer, a differential amplifier and a programmable band-pass filter; the drill bit measurement unit and the acoustic wave acquisition unit are arranged behind the drill bit, and the drill string monitoring unit is arranged at both ends of the probe rod.

3. The system for sensing the dynamic mechanical parameters of rock mass based on the while-drilling parameters according to claim 1, characterized in that, It also includes a warning module, which alarms when the mechanical parameters acquired by the dynamic mechanical parameter acquisition module exceed the corresponding warning threshold, and the warning threshold is determined based on historical statistics and multi-task weights.

4. A rock mass dynamic mechanical parameter sensing method based on a while-drilling parameter, characterized in that, Comprise: Step 1, construct training sample data: adopt the drill bit integrated sensor module to collect the while-drilling parameters and preprocess them, including synchronization alignment, hybrid noise reduction and feature extraction, and standardize and feature dimensionality reduction of the extracted features; acquire the mechanical parameters corresponding to the while-drilling parameters through laboratory core true value and finite element simulation, and construct the training sample data based on the preprocessed data and the corresponding mechanical parameters; Step 2, based on the multi-modal adaptive graph attention time series network, a rock mass dynamic mechanical parameter acquisition model is constructed and trained based on the training sample data; Step 3, real-time acquisition of while-drilling parameters using drill bit integrated sensor modules, and pre-processing using the same pre-processing method as the training samples, obtaining mechanical parameters based on pre-processed data and trained rock mass dynamic mechanical parameter acquisition model; Wherein, the mechanical parameters are: instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion and internal friction angle; The rock mass dynamic mechanical parameter acquisition model comprises: A sensor topology graph module for constructing a spatial correlation graph structure of sensor data; A graph attention layer module for calculating attention coefficients based on the spatial correlation graph structure and performing feature aggregation based on the attention coefficients; A time series convolution graph module for multi-scale extraction of time series dimensions based on the sequence features output by the graph attention layer module, and for enhancing feature nonlinear expression and local feature learning accuracy through causal inflation convolution group and lightweight regression head based on local supervised optimization training process; A time-aware Transformer enhancement module: modeling the long-term and short-term dependence relationship of time series features, introducing relative position encoding through multi-head self-attention mechanism, and adjusting attention weights based on exponential decay factor to enhance the simultaneous modeling capability of sudden events and long-term trends; A multi-task regression head module: realizing parallel prediction of multiple target parameters based on the final feature sequence.

5. The method according to claim 4, wherein, The pre-processing steps are as follows: Aligning the time scales of all sensor channels, dividing the signal stream into M window lengths and S step lengths in real time, and processing each window independently; Band-pass filtering of different frequencies according to data characteristics; Performing ensemble empirical mode decomposition on the filtered signal, selecting the first N IMF components for reconstruction, and performing singular value decomposition on the reconstructed signal to retain the principal components with a singular value energy ratio of more than 95%; Dynamically estimating the noise spectrum of the acoustic signal, adaptively reducing the noise energy according to the frequency band, and retaining the rock mass fracture characteristic spectrum; Parallelly extracting multiple features of the windowed signal, including: Time domain features: calculating mean, variance, peak value, RMS energy, kurtosis, skewness, instantaneous mechanical specific energy, and drilling power in the window; Frequency domain features: extracting power spectral density using fast Fourier transform, calculating dominant frequency, 1-5 kHz band energy ratio, and spectral kurtosis; identifying the resonance peak frequency through cepstrum analysis; Time-frequency domain features: performing short-time Fourier transform on the band-passed signal to generate a time-frequency spectrum, extracting energy entropy and spectral centroid trajectory; applying wavelet packet decomposition to obtain multi-scale sub-band coefficients and calculate the energy, information entropy, and frequency band variability of each sub-band; Dynamic event labeling: detecting torque mutations using a mutation point detection algorithm, labeling the stuck / abrasion event time, and embedding a [0,1] event flag in the feature vector; Standardizing the extracted features: Z-score normalization for time domain features, and Min-Max scaling to [0,1] for frequency domain features; PCA dimensionality reduction for time domain, frequency domain, and time-frequency domain features.

6. The method of claim 4, wherein, The rock mass dynamic mechanical parameter acquisition model is deployed in layers, with the sensor topology graph module, graph attention layer module, and time series convolution graph module deployed on the downhole edge computing node, and the time-aware Transformer enhancement module and multi-task regression head module deployed on the ground server.

7. The method of claim 4, wherein, The model training process uses a composite loss function L: , where y is the truth value; The predicted value output by the model; Weights; symbol It is an L1 norm. ,symbol It is the L2 norm. , This refers to the predicted value of the i-th physical quantity in the final output vector for the current sample. The true value of the i-th physical quantity in the final output vector for the current sample is given. When i=1,…,5, it corresponds to the instantaneous elastic modulus, dynamic Poisson's ratio, transient uniaxial compressive strength, cohesion, and internal friction angle, respectively. ,in Indicates the first The predicted value of the regression head for the s-th sample, For the s-th sample in the th case The true values ​​on the regression heads, where n is the number of samples. , This represents the number of regression heads; These are the model parameters.

8. The method of claim 4, wherein, Further comprising step 4, when the dynamic mechanical parameter acquisition module obtains the mechanical parameter exceeding the corresponding early warning threshold, an alarm is given, and the early warning threshold is determined based on historical statistics and multi-task weight.

Images