Slope lithology detection system and method for super-large open-pit coal mine based on while-drilling lithology identification

Abstract

This invention provides a lithology detection system and method for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification, relating to the field of open-pit coal mine slope geological exploration technology. The system includes: a drilling subsystem for providing drilling power; a data acquisition subsystem for acquiring sensor signals during the drilling process; a signal conversion subsystem electrically connected to the data acquisition subsystem for receiving and converting sensor signals; a data processing subsystem electrically connected to the signal conversion subsystem for receiving and preprocessing the converted sensor signals; and an intelligent lithology identification subsystem for extracting temporal statistical features from the preprocessed sensor signals, selecting the optimal feature combination, classifying and identifying lithology, and outputting a depth-domain lithology profile. This invention utilizes the aforementioned lithology detection system and method for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification to achieve integrated output of real-time drilling-while-drilling lithology identification, model self-updating, and geological modeling data for ultra-large open-pit coal mine slopes.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology for open-pit coal mine slopes, and in particular to a lithological exploration system and method for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification. Background Technology

[0002] Currently, slope stability in ultra-large open-pit coal mines is one of the core issues for mine safety. The distribution of slope lithology (such as coal, mudstone, sandstone, and conglomerate) directly affects the mechanical behavior and failure mode of the slope. Therefore, accurately grasping slope lithology information is crucial. Existing slope lithology detection methods mainly rely on traditional geological exploration methods such as core drilling, manual logging, and laboratory testing.

[0003] However, the aforementioned traditional methods suffer from significant technical bottlenecks: firstly, they are time-consuming and costly, making it difficult to meet the practical needs of rapid detection of large-scale slopes; secondly, core sampling has poor representativeness, and discrete sampling points cannot fully reflect the spatial continuous distribution of slope lithology. Furthermore, although there are existing technologies that use sensors mounted on drilling rigs to collect drilling parameters for lithology identification, most current methods are designed for underground coal mine roadway roof scenarios, and their outputs are primarily used for anchor bolt support design, not directly for slope geological modeling, resulting in a mismatch between application scenarios. At the algorithmic level, existing technologies mostly employ single neural networks or simple feature inputs, exhibiting defects such as feature redundancy, model complexity, and insufficient real-time performance. They also generally lack confidence assessment mechanisms, making them prone to misjudgments at lithological interfaces.

[0004] Therefore, there is an urgent need for an integrated detection system and method specifically designed for ultra-large open-pit coal mine slope scenarios, capable of real-time lithology identification during drilling, possessing high reliability and confidence assessment capabilities, and outputting standardized geological modeling data, so as to achieve efficient and accurate detection of slope lithology and automatic construction of three-dimensional geological models. Summary of the Invention

[0005] The purpose of this invention is to provide a lithological detection system and method for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification. This system aims to solve the prominent problems of low efficiency, high cost, and poor representativeness of existing slope lithology detection methods. It achieves real-time drilling-while-drilling lithology identification, model self-updating, and integrated output of geological modeling data for ultra-large open-pit coal mine slopes, thereby improving technical adaptability and model robustness and providing accurate geological basis for slope stability evaluation and coal resource distribution.

[0006] To achieve the above objectives, the present invention provides a lithological detection system for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification, including a drilling subsystem, a data acquisition subsystem, a signal conversion subsystem, a data processing subsystem, and an intelligent lithology identification subsystem; The drilling subsystem is used to provide drilling power; The data acquisition subsystem is used to collect sensor signals from the drilling subsystem during the drilling process; The signal conversion subsystem is electrically connected to the data acquisition subsystem and is used to receive and convert sensor signals. The data processing subsystem is electrically connected to the signal conversion subsystem and is used to receive and preprocess the converted sensor signals. The intelligent lithology identification subsystem is used to extract time-domain statistical features from preprocessed sensor signals, select the optimal feature combination, classify and identify lithology, and output depth-domain lithology profiles.

[0007] Preferably, the sensor signals include drill pressure signals, torque signals, rotational speed signals, displacement signals, and vibration signals; the data acquisition subsystem includes a drill pressure sensor, a torque sensor, a rotational speed sensor, a footage displacement sensor, and a vibration impact load sensor; the drill pressure sensor is located between the bottom and the center of the drilling subsystem and connected by a rigid threaded sleeve, and is used to collect drilling propulsion force; the torque sensor is located between the drive spindle and the drill pipe of the drilling subsystem and connected by a rigid sleeve, and is used to collect real-time drilling torque; the rotational speed sensor is integrated with the torque sensor and measures coaxially, and is used to collect drill pipe rotational speed; the footage displacement sensor is located on the upper part of the air leg of the drilling subsystem and connected to the air leg handle by a pull rope, and is used to output displacement signals in real time; the vibration impact load sensor is located above the motor drive assembly of the drilling subsystem and connected by magnetic attraction, and is used to collect axial vibration signals and radial vibration signals.

[0008] Preferably, the signal conversion subsystem includes an explosion-proof signal box and a first power transformer, a second power transformer, an air switch, a junction box, a first XSE5 digital display alarm controller, a second XSE5 digital display alarm controller, a signal conversion module, a first RS_485 serial port and a second RS_485 serial port, all housed within the explosion-proof signal box. The signal conversion subsystem is electrically connected to each sensor in the data acquisition subsystem via signal lines.

[0009] Preferably, the data processing subsystem is an explosion-proof laptop computer; the data processing subsystem preprocesses the converted sensor signals, including removing outliers, normalizing the data, and splitting the dataset into training and testing sets.

[0010] Preferably, the intelligent lithology identification subsystem includes a vibration time-domain feature extraction module, a feature optimization module, a lithology classification and identification module, a confidence threshold rejection module, and a depth domain alignment module; The vibration time-domain feature extraction module is used to perform high-pass filtering, sliding window segmentation, and 3D modeling on the vibration signal in the preprocessed sensor signal. After removing outliers, extract time-domain statistical features; The feature optimization module adopts a recursive feature elimination method, takes the 1DCNN-BiLSTM-CBAM model as the base evaluator, gradually eliminates the time-domain statistical features with the lowest importance in a recursive manner, determines the optimal number of feature combinations in combination with 5-fold cross-validation, and screens out the optimal feature combination; The lithology classification and identification module splices the optimal feature combination and drilling working condition parameters into a feature vector, inputs the 1DCNN-BiLSTM-CBAM hybrid neural network model, and outputs the lithology category and confidence; The confidence threshold rejection module is provided with a confidence threshold T, when the confidence is ≥ T, the lithology label is output normally, and when the confidence is < T, it is marked as a layer to be confirmed; The depth domain alignment module reads the displacement signal output by the footage displacement sensor and converts it into borehole depth and rate of penetration, adopts the trapezoidal integration method to resample the time-series identification results to the depth domain, and outputs the lithology profile in the depth domain.

[0011] Preferably, the time-domain statistical features include root amplitude, peak-to-peak value, absolute mean value, mean value, maximum absolute value, standard deviation, waveform index, peak index, pulse index, skewness, kurtosis, margin index and root mean square value; the optimal feature combination includes waveform index, peak-to-peak value, absolute mean value, skewness, kurtosis, root mean square value and standard deviation; the drilling working condition parameters include weight on bit signal, torque signal, rotational speed signal and borehole depth; the 1DCNN-BiLSTM-CBAM hybrid neural network model includes an input layer, a 1DCNN layer, a BiLSTM layer, a CBAM module, a fully connected layer and a Softmax output layer; the input layer is configured to receive the feature vector, the 1DCNN layer is configured to extract local time-series features from the feature vector, the BiLSTM layer is configured to capture the bidirectional long-term time-series dependency relationship, the CBAM module is configured to perform adaptive weighting enhancement on the feature vector output by the BiLSTM layer, and the weighted feature vector is finally processed by the fully connected layer and the Softmax output layer to obtain the lithology category and confidence.

[0012] Preferably, the system further comprises a model training and verification module, the model training and verification module uses a historical drilling data training set to train the 1DCNN-BiLSTM-CBAM hybrid neural network model, and uses a historical drilling data test set to independently verify the model performance of the 1DCNN-BiLSTM-CBAM hybrid neural network model, calculates the overall accuracy, the recall rate, precision rate and F1 score of each type of lithology, and outputs a confusion matrix.

[0013] Preferably, it also includes a three-dimensional geological modeling module. The three-dimensional geological modeling module registers the lithological profiles of several boreholes in depth domain according to spatial coordinates to form a three-dimensional lithological constraint point set. It adopts the radial basis function implicit modeling method to construct implicit functions for various types of lithology. Through voxelization and isosurface reconstruction, it generates a three-dimensional lithological entity model and outputs a vertical geological profile map along the borehole, a three-dimensional visualization model of rock layer distribution, and a standardized geological modeling data list, i.e., geological modeling data.

[0014] Preferably, it also includes a model dynamic update module, which sets an incremental learning strategy: when the accumulated new drilling lithology data reaches a preset threshold, the model incremental update is automatically triggered, the new data is merged into the original training set and validation set, and the transfer learning method is used to fine-tune the original model to generate a new version of the model and replace the original model.

[0015] This invention also provides a method for lithological detection of ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification, which, using the above-mentioned system, includes the following steps: Step S1: Collect sensor signals during the drilling process, convert them, transmit them, and perform preprocessing. Step S2: Extract time-domain statistical features from the vibration signals in the preprocessed sensor signals and select the optimal feature combination, classify and identify lithology, and output the depth-domain lithology profile. Step S3: Register the depth domain lithological profile according to spatial coordinates and generate geological modeling data using the radial basis function implicit modeling method; Step S4: Continuously collect the depth domain lithological profile output in step S2 and count the newly added drilling lithological data; when the accumulated newly added drilling lithological data reaches a preset threshold, incrementally update the neural network model.

[0016] Therefore, the present invention employs the above-mentioned lithological detection system and method for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification, and the beneficial technical effects are as follows: (1) This invention realizes real-time high-precision identification of slope lithology during drilling. By simultaneously collecting drilling pressure signals, torque signals, rotation speed signals, displacement signals and vibration signals during drilling and performing intelligent analysis, a continuous profile with a lithology label every 1cm depth can be output at the same time as drilling. It does not rely on traditional core drilling, manual logging and indoor testing, fundamentally solving the problems of long cycle, high cost and poor representativeness of traditional methods, and greatly improving the detection efficiency.

[0017] (2) This invention selects the optimal combination of seven features from 13 time-domain statistical features by using a recursive feature elimination method, namely waveform index, peak-to-peak value, absolute average value, skewness, kurtosis, root mean square value and standard deviation. This effectively reduces the feature dimension, reduces data redundancy, improves the computational efficiency and real-time performance of the model, and solves the algorithm defects of feature redundancy, model complexity and insufficient real-time performance in the prior art.

[0018] (3) This invention adopts a 1DCNN-BiLSTM-CBAM hybrid neural network model, which integrates the 1DCNN layer to extract local temporal features, the BiLSTM layer to capture long temporal dependencies, and the CBAM module to perform adaptive weighted enhancement of channel attention and spatial attention, thereby improving the accuracy and robustness of lithology classification. After independent verification on 30% of the test set, the overall accuracy is not less than 88%, and the recall rate of each lithology is not less than 80%, which effectively solves the problem of insufficient recognition accuracy of existing single neural network models.

[0019] (4) By setting a confidence threshold rejection mechanism, the present invention actively marks the result as a layer to be confirmed instead of forcibly outputting lithology labels when the confidence of the model output is lower than the threshold. This effectively avoids misjudgment caused by insufficient model confidence in transition areas such as lithological interfaces, solves the problem that the existing technology generally lacks a confidence assessment mechanism and is prone to misjudgment at lithological interfaces, and improves the reliability of the identification results.

[0020] (5) This invention uses the displacement signal obtained by the advance displacement sensor to perform conversion through the depth domain alignment module, and at the same time uses the trapezoidal integral method to accurately resample the identification results of the time series to the depth domain, and outputs a continuous profile with a lithology label for every 1cm depth, realizing a high-precision mapping between the time domain identification results and the geological depth, and providing accurate, continuous and standardized data input for subsequent three-dimensional geological modeling.

[0021] (6) This invention registers the lithological profiles of multiple boreholes in the depth domain according to spatial coordinates, generates a continuous and smooth three-dimensional lithological entity model by using the radial basis function implicit modeling method, and outputs a vertical geological profile map along the borehole and a three-dimensional visualization model of rock layer distribution. This realizes the automated and integrated output of detection data to a three-dimensional geological model, enabling the detection results to directly serve slope stability evaluation and coal mine resource distribution analysis, and solves the problem that the output results of existing drilling identification technology cannot be directly applied to slope geological modeling.

[0022] (7) By setting an incremental learning strategy, when the accumulated new drilling lithology data reaches a preset threshold (such as 500 meters of new footage or 20 boreholes), the model is automatically triggered to update incrementally. The transfer learning method is used to fine-tune the original model in a small number of rounds with a low learning rate, generate a new version of the model and replace the old model, so that the model can be continuously optimized with the continuous accumulation of exploration data, avoiding the defect of traditional static models that cannot be updated by themselves, and realizing the continuous evolution of geological models. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the system structure of an embodiment of the ultra-large open-pit coal mine slope lithology detection system based on drilling lithology identification of the present invention; Figure 2 This is a flowchart illustrating an embodiment of the present invention's method for detecting lithology on ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification.

[0024] Figure Labels 1. Drilling subsystem; 2. Data acquisition subsystem; 201. Drill pressure sensor; 202. Torque sensor; 203. Rotation speed sensor; 204. Cut-off displacement sensor; 205. Vibration and impact load sensor; 3. Signal conversion subsystem; 4. Data processing subsystem. Detailed Implementation

[0025] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0027] Example 1 like Figure 1 As shown, the ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification includes a drilling subsystem 1, a data acquisition subsystem 2, a signal conversion subsystem 3, a data processing subsystem 4, and an intelligent lithology identification subsystem.

[0028] Drilling subsystem 1 provides drilling power. Drilling subsystem 1 is a pneumatic anchor drill, connected to a blower and a water pump to ensure the required air pressure, air volume, and water pressure for operation. During exploration operations, the drill is placed vertically, the drill pressure sensor 201 is brought into contact with the ground, the drill rod is installed, and drilling begins.

[0029] Data acquisition subsystem 2 is used to acquire sensor signals from drilling subsystem 1 during the drilling process. These sensor signals include drill pressure signals, torque signals, rotational speed signals, displacement signals, and vibration signals. Data acquisition subsystem 2 includes a drill pressure sensor 201, a torque sensor 202, a rotational speed sensor 203, a cut-off displacement sensor 204, and a vibration and impact load sensor 205.

[0030] The drill pressure sensor 201 is located between the bottom and the center of the drilling subsystem 1, connected via a rigid threaded sleeve, and is used to collect drilling thrust, with an accuracy ≤ ±1%FS and a sampling frequency ≥ 10Hz. The torque sensor 202 is located between the drive spindle and the drill pipe of the drilling subsystem 1, connected via a rigid sleeve, and is used to collect real-time drilling torque, with an accuracy ≤ ±1%FS and a response time ≤ 10ms. The speed sensor 203 is integrated with the torque sensor 202, measuring coaxially, and is used to collect drill pipe speed, with a measurement range of 0~300r / min and a resolution ≤ 1r / min. The feed displacement sensor 204 is located on the upper part of the air leg of the drilling subsystem 1, connected to the air leg handle via a pull rope, and is used to output displacement signals in real time, with an accuracy ≤ ±1mm. The vibration and impact load sensor 205 is located above the motor drive assembly of the drilling subsystem 1, connected via magnetic attraction, and is used to collect axial and radial vibration signals.

[0031] The aforementioned drill pressure sensor 201, torque sensor 202, speed sensor 203, feed displacement sensor 204, and vibration impact load sensor 205 respectively collect drill pressure signal, torque signal, speed signal, displacement signal, and vibration signal; wherein, the drill pressure signal represents the drilling propulsion force, the torque signal represents the real-time drilling torque, the speed signal represents the drill pipe speed, and the vibration signal includes axial vibration signal and radial vibration signal.

[0032] The signal conversion subsystem 3 includes an explosion-proof signal box and, housed within the box, a first power transformer, a second power transformer, an air switch, a junction box, a first XSE5 digital display alarm controller, a second XSE5 digital display alarm controller, a signal conversion module, a first RS_485 serial port, and a second RS_485 serial port. All these components are connected via signal lines. In this embodiment, the explosion-proof signal box is a coal mine safety type explosion-proof signal box.

[0033] The signal conversion subsystem 3 is electrically connected to each sensor in the data acquisition subsystem 2 via signal lines. It is used to receive and convert sensor signals, complete the amplification, filtering, analog-to-digital conversion and intrinsically safe explosion-proof processing of weak signals, and transmit the converted sensor signals to the data processing subsystem 4.

[0034] The data processing subsystem 4 is electrically connected to the signal conversion subsystem 3 and is used to receive and preprocess the converted sensor signals. The data processing subsystem 4 is an explosion-proof laptop computer used to receive, store, and process signals transmitted by the signal conversion subsystem 3. The data processing subsystem 4 preprocesses the converted sensor signals, including removing outliers using the 3σ principle, normalization, and splitting the dataset into a 70% training set and a 30% test set.

[0035] The intelligent lithology identification subsystem is embedded in the data processing subsystem 4. It is used to extract time-domain statistical features from the preprocessed sensor signals, select the optimal feature combination, classify and identify lithology, and output depth-domain lithology profiles.

[0036] The intelligent lithology identification subsystem includes a vibration time-domain feature extraction module, a feature optimization module, a lithology classification and identification module, a confidence threshold rejection module, and a depth domain alignment module.

[0037] The vibration time-domain feature extraction module is used to perform high-pass filtering (removing low-frequency rotation noise of the drilling rig) and sliding window segmentation on the vibration signal in the preprocessed sensor signal, and 3D modeling. After removing outliers, time-domain statistical features are extracted. These features include root square amplitude, peak-to-peak value, absolute mean, mean, maximum absolute value, standard deviation, waveform index, peak index, impulse index, skewness, kurtosis, margin index, and root mean square value, totaling 13 items.

[0038] The feature selection module employs the Recursive Feature Elimination (RFE) method. Using a 1DCNN-BiLSTM-CBAM model as the base evaluator, the model is trained with all 13 temporal statistical features. The contribution of each temporal statistical feature is evaluated through model weights or feature importance scores. The least important temporal statistical features are recursively eliminated. After each feature is eliminated, the model is retrained and the importance scores of the remaining features are recalculated. This process is repeated. Combined with 5-fold cross-validation, the optimal feature combination is determined to be 7 dimensions. This process continues until the remaining features reach 7 dimensions. Finally, the 7-dimensional optimal feature combination with the strongest recognition power is selected from the 13 temporal statistical features, including waveform indicators, peak-to-peak value, absolute mean, skewness, kurtosis, root mean square value, and standard deviation.

[0039] The lithology classification and identification module combines the optimal 7-dimensional feature set with drilling condition parameters to form an 11-dimensional feature vector. This vector is then input into a 1DCNN-BiLSTM-CBAM hybrid neural network model, outputting four lithology categories: coal, mudstone, sandstone, and conglomerate, along with their confidence levels. Drilling condition parameters include drill pressure, torque, rotational speed, and borehole depth.

[0040] The 1DCNN-BiLSTM-CBAM hybrid neural network model includes an input layer, a 1DCNN layer, a BiLSTM layer, a CBAM module, a fully connected layer, and a Softmax output layer; the input layer is configured to receive 11-dimensional feature vectors, the 1DCNN layer (one-dimensional convolutional neural network layer) has three layers with a kernel size of 3 and the number of channels being 64, 128, and 256 respectively, and is configured to extract local temporal features from the feature vectors; the BiLSTM layer (bidirectional long short-term memory network layer) has 128 hidden units, and is configured to capture the forward and backward bidirectional long-term temporal dependencies; the CBAM module (convolutional block attention module) is configured to perform adaptive weighting enhancement on the feature vectors output by the BiLSTM layer sequentially through the channel attention mechanism and the spatial attention mechanism; the weighted feature vectors are finally processed by the fully connected layer and the Softmax output layer to obtain the probability distribution of four lithology types, the one with the highest probability is the lithology category discrimination result, and the maximum probability value is the confidence, and the lithology category and the confidence are output.

[0041] The confidence threshold rejection module is provided with a confidence threshold T, with a default value T=0.6, which can be adjusted according to actual engineering requirements. When the confidence is ≥ T, the lithology label is output normally; when the confidence is < T, it is marked as a layer to be confirmed, and no specific lithology name is forcibly output, so as to avoid misjudgment at the lithology interface.

[0042] The depth domain alignment module reads the displacement signal output by the footage displacement sensor 204 and converts it into borehole depth and rate of penetration, adopts the trapezoidal integration method to resample the recognition results of the time sequence to the depth domain, and outputs a continuous profile with one lithology label per 1 cm depth, that is, the lithology profile in the depth domain.

[0043] The system of the present invention further includes a model training and verification module. The model training and verification module needs to collect historical drilling data with labeled lithology (no less than 5000 samples), and divide the historical drilling data into a historical drilling data training set and a historical drilling data test set at a ratio of 7:3. The historical drilling data training set is used to train the 1DCNN-BiLSTM-CBAM hybrid neural network, the optimizer adopts Adam, the loss function is cross-entropy loss, the initial learning rate is set to 0.001, the batch size is set to 64, the number of training epochs is 100, and the early stopping method is adopted to prevent overfitting. After the training is completed, the historical drilling data test set is used to independently verify the model performance of the 1DCNN-BiLSTM-CBAM hybrid neural network model, calculate the overall accuracy, the recall rate, precision rate and F1-score of each type of lithology, and output the confusion matrix, so as to finally obtain the model performance evaluation report. The verification results need to satisfy: the overall accuracy ≥ 88%, and the recall rate of each type of lithology ≥ 80%, otherwise the model hyperparameters need to be adjusted or the training data needs to be expanded. The verification result serves as the performance benchmark before model deployment.

[0044] The system of the present invention also includes a three-dimensional geological modeling module, which is embedded in or external to the data processing subsystem 4.

[0045] The 3D geological modeling module registers the depth-domain lithological profiles of multiple boreholes according to spatial coordinates (X, Y, Z) to form a 3D lithological constraint point set. Employing a radial basis function implicit modeling method, it constructs implicit functions for each type of lithology. Through voxelization and isosurface reconstruction, it generates a continuous and smooth 3D lithological solid model and outputs a vertical geological profile along the borehole, a 3D visualization model of the rock layer distribution, and a standardized geological modeling data list, i.e., the geological modeling data. The standardized geological modeling data list includes information such as borehole number, starting depth, ending depth, lithology name, and confidence level.

[0046] The system of this invention also includes a model dynamic update module, which is embedded in the data processing subsystem 4 and has no independent hardware. The model dynamic update module sets an incremental learning strategy: when the cumulative amount of newly added drilling lithological data (including new borehole data and manually verified "layers to be confirmed" data) reaches a preset threshold (such as adding 500 meters of footage or adding 20 boreholes), the model incremental update is automatically triggered. The new data is merged into the original historical drilling data training set and historical drilling data validation set at a ratio of 7:3. The transfer learning method is used to fine-tune the original model, that is, using the original model as the initial weight, a low learning rate (such as 0.0001) is used to perform 20 rounds of fine-tuning training to generate a new version of the model and replace the original model, thereby realizing the continuous optimization of the geological model.

[0047] like Figure 2 As shown, the method for lithological detection of ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification, using the above-mentioned system, includes the following steps: Step S1: Collect sensor signals during the drilling process, convert them, transmit them, and perform preprocessing.

[0048] Connect the drilling subsystem 1 to the blower and water pump to ensure the required air pressure, air volume, and water pressure. Place the drilling rig vertically, ensuring the drill pressure sensor 201 is in contact with the ground, install the drill rod, and begin drilling. During drilling, sensor signals are synchronously and in real-time acquired by five types of sensors in the data acquisition subsystem 2. The acquired sensor signals are transmitted to the signal conversion subsystem 3 via signal lines, processed by the signal conversion subsystem 3, and then transmitted to the data processing subsystem 4. The data processing subsystem 4 preprocesses the received converted sensor signals.

[0049] Step S2: Extract time-domain statistical features from the vibration signals in the preprocessed sensor signals, select the optimal feature combination, classify and identify lithology, and output the depth-domain lithology profile.

[0050] Step S3: Register the lithological profile in the depth domain according to spatial coordinates, and generate geological modeling data using the radial basis function implicit modeling method.

[0051] Step S4: Continuously collect the depth domain lithology profile output in step S2 and count the newly added drilling lithology data; when the accumulated newly added drilling lithology data reaches the preset threshold, retrieve the historical drilling data of labeled lithology stored in the model training and verification module, incrementally update the neural network model, and send the updated model to the intelligent lithology identification subsystem for lithology identification in the subsequent drilling process.

[0052] Therefore, the present invention adopts the above-mentioned lithological detection system and method for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification, which solves the prominent problems of low efficiency, high cost and poor representativeness of existing slope lithology detection methods. It realizes the integrated output of real-time drilling-while-drilling lithology identification, model self-updating and geological modeling data for ultra-large open-pit coal mine slopes, improves the technical adaptability and model robustness, and provides accurate geological basis for slope stability evaluation and coal resource distribution.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A lithological detection system for ultra-large open-pit coal mine slopes based on drilling-while-drilling lithology identification, characterized in that, It includes a drilling subsystem, a data acquisition subsystem, a signal conversion subsystem, a data processing subsystem, and an intelligent lithology identification subsystem; The drilling subsystem is used to provide drilling power; The data acquisition subsystem is used to acquire sensor signals during the drilling process of the drilling subsystem; The signal conversion subsystem is electrically connected to the data acquisition subsystem and is used to receive and convert sensor signals. The data processing subsystem is electrically connected to the signal conversion subsystem and is used to receive and preprocess the converted sensor signals. The intelligent lithology identification subsystem is used to extract time-domain statistical features from preprocessed sensor signals, select the optimal feature combination, classify and identify lithology, and output depth-domain lithology profiles.

2. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification as described in claim 1, characterized in that, The sensor signals include drill pressure signals, torque signals, rotational speed signals, displacement signals, and vibration signals. The data acquisition subsystem includes a drill pressure sensor, a torque sensor, a rotational speed sensor, a footage displacement sensor, and a vibration impact load sensor. The drill pressure sensor is located between the bottom and the center of the drilling subsystem and is connected via a rigid threaded sleeve; it is used to collect drilling propulsion force. The torque sensor is located between the drive spindle and the drill pipe of the drilling subsystem and is connected via a rigid sleeve; it is used to collect real-time drilling torque. The rotational speed sensor is integrated with the torque sensor and measures coaxially; it is used to collect drill pipe rotational speed. The footage displacement sensor is located on the upper part of the air leg of the drilling subsystem and is connected to the air leg handle via a pull rope; it is used to output displacement signals in real time. The vibration impact load sensor is located above the motor drive assembly of the drilling subsystem and is connected via magnetic attraction; it is used to collect axial and radial vibration signals.

3. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification as described in claim 1, characterized in that, The signal conversion subsystem includes an explosion-proof signal box and a first power transformer, a second power transformer, an air switch, a junction box, a first XSE5 digital display alarm controller, a second XSE5 digital display alarm controller, a signal conversion module, a first RS_485 serial port, and a second RS_485 serial port, all housed within the explosion-proof signal box. The signal conversion subsystem is electrically connected to each sensor in the data acquisition subsystem via signal lines.

4. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification as described in claim 1, characterized in that, The data processing subsystem is an explosion-proof laptop computer; the data processing subsystem preprocesses the converted sensor signals, including removing outliers, normalizing the data, and splitting the dataset into training and testing sets.

5. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification as described in claim 1, characterized in that, The intelligent lithology identification subsystem includes a vibration time-domain feature extraction module, a feature optimization module, a lithology classification and identification module, a confidence threshold rejection module, and a depth domain alignment module. The vibration time-domain feature extraction module is used to perform high-pass filtering, sliding window segmentation, and 3D modeling on the vibration signal in the preprocessed sensor signal. After removing outliers, extract time-domain statistical features; The feature selection module adopts a recursive feature elimination method, using the 1DCNN-BiLSTM-CBAM model as the base evaluator. It gradually eliminates the least important temporal statistical features through recursion, and determines the optimal number of feature combinations by combining 5-fold cross-validation, thus selecting the optimal feature combination. The lithology classification and identification module concatenates the optimal feature combination with drilling condition parameters into a feature vector, inputs it into a 1DCNN-BiLSTM-CBAM hybrid neural network model, and outputs the lithology category and confidence level. The confidence threshold rejection module is provided with a confidence threshold T, when the confidence is ≥T, the lithology label is output normally, and when the confidence is <T, it is marked as a layer to be confirmed; The depth domain alignment module reads the displacement signal output by the footage displacement sensor, converts it into borehole depth and mechanical drilling rate, resamples the time-series identification results to the depth domain using the trapezoidal integration method, and outputs the depth domain lithology profile.

6. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification according to claim 5, characterized in that, The time-domain statistical features include root amplitude, peak-to-peak value, absolute mean value, mean value, maximum absolute value, standard deviation, waveform index, peak index, pulse index, skewness, kurtosis, margin index and root mean square value; the optimal feature combination includes waveform index, peak-to-peak value, absolute mean value, skewness, kurtosis, root mean square value and standard deviation; the drilling working condition parameters include weight on bit signal, torque signal, rotational speed signal and borehole depth; the 1DCNN-BiLSTM-CBAM hybrid neural network model includes an input layer, a 1DCNN layer, a BiLSTM layer, a CBAM module, a fully connected layer and a Softmax output layer; the input layer is used for receiving feature vectors, the 1DCNN layer is used for extracting local time-series features from the feature vectors, the BiLSTM layer is used for capturing the bidirectional long-term time-series dependency relationship, the CBAM module is used for performing adaptive weighting enhancement on the feature vectors output by the BiLSTM layer, and the weighted feature vectors are finally processed by the fully connected layer and the Softmax output layer to obtain the lithology category and confidence.

7. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification according to claim 1, characterized in that, It further comprises a model training and verification module, which trains the 1DCNN-BiLSTM-CBAM hybrid neural network model using a historical drilling data training set, independently verifies the model performance of the 1DCNN-BiLSTM-CBAM hybrid neural network model using a historical drilling data test set, calculates the overall accuracy, the recall rate, precision rate and F1 score of each type of lithology, and outputs a confusion matrix.

8. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification according to claim 1, characterized in that, It further comprises a 3D geological modeling module, which registers the depth domain lithology profiles of a plurality of boreholes according to spatial coordinates to form a 3D lithology constraint point set, adopts a radial basis function implicit modeling method to construct implicit functions for each type of lithology respectively, generates a 3D lithology solid model through voxelization and isosurface reconstruction, and outputs a vertical geological profile along the borehole, a 3D visualization model of stratum distribution and a standardized geological modeling data list, namely geological modeling data.

9. The ultra-large open-pit coal mine slope lithology detection system based on drilling-while-drilling lithology identification according to claim 1, characterized in that, It further comprises a model dynamic updating module, which is provided with an incremental learning strategy: when the newly added logging-while-drilling lithology data accumulates to reach a preset threshold, the incremental update of the model is automatically triggered, the new data is merged into the original training set and verification set, fine-tuning is performed on the basis of the original model in a transfer learning manner, and a new version of the model is generated to replace the original model.

10. A method for lithological detection of slopes in ultra-large open-pit coal mines based on drilling-while-drilling lithology identification, employing the system described in any one of claims 1-9, characterized in that, The method comprises the following steps: Step S1: Collect sensor signals during the drilling process, transmit the signals after conversion and perform preprocessing; Step S2: Extract time-domain statistical features from the vibration signals in the preprocessed sensor signals, screen out the optimal feature combination, classify and identify lithology, and output a depth domain lithology profile; Step S3: Register the depth domain lithological profile according to spatial coordinates and generate geological modeling data using the radial basis function implicit modeling method; Step S4: Continuously collect the depth domain lithological profile output in step S2 and count the newly added drilling lithological data; when the accumulated newly added drilling lithological data reaches a preset threshold, incrementally update the neural network model.

Images